You can tell it what you want (\u201cSend this NDA to John for signature\u201d), and it will:\n \u2022 Find the right document and contacts\n \u2022 Detect and place signing fields\n \u2022 Assign signer roles\n \u2022 Suggest titles and messages\n \u2022 Send requests via email, SMS, WhatsApp, or Microsoft Teams\n \u2022 Track status and automatically follow up with signers
It supports the full e-signature lifecycle: self-signing, sending requests, templates, reminders, resends, and real-time status updates. Follow-ups are generated automatically with tone and timing adjusted based on progress.
visit https://docendorse.com/esignature to learn more"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["an","ai","assistant","that"],"value":"Show HN: DocEndorse \u2013 An AI assistant that runs your e-sign workflow in chat"}},"_tags":["story","author_kariopaul","story_46785751","show_hn"],"author":"kariopaul","created_at":"2026-01-27T20:09:46Z","created_at_i":1769544586,"num_comments":0,"objectID":"46785751","points":1,"story_id":46785751,"story_text":"DocEndorse is an AI-powered e-signature assistant that lets you prepare, send, track, and follow up on documents through natural language instead of forms and drag-and-drop editors.
You can tell it what you want (\u201cSend this NDA to John for signature\u201d), and it will:\n \u2022 Find the right document and contacts\n \u2022 Detect and place signing fields\n \u2022 Assign signer roles\n \u2022 Suggest titles and messages\n \u2022 Send requests via email, SMS, WhatsApp, or Microsoft Teams\n \u2022 Track status and automatically follow up with signers
It supports the full e-signature lifecycle: self-signing, sending requests, templates, reminders, resends, and real-time status updates. Follow-ups are generated automatically with tone and timing adjusted based on progress.
visit https://docendorse.com/esignature to learn more","title":"Show HN: DocEndorse \u2013 An AI assistant that runs your e-sign workflow in chat","updated_at":"2026-03-05T23:27:05Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"thinkmoon"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"With ThinkMoon you can build your own Alpha Arena: plug in models from OpenRouter, OpenAI, Anthropic (or our custom ones), feed them real-time data (candles, order book, ticker) \u2014 then send them to trade real crypto (via Binance Futures).
You define the rules: risk limits, which coins (BTC, ETH, SOL, XRP\u2026 whatever), even mix different LLMs and compare their performance.
Every move the AI makes is logged \u2014 full prompt, reasoning, market snapshot \u2014 so you can audit or learn from it.
There\u2019s a live dashboard with P&L, open positions, and the AI\u2019s chain of thought. Alerts via Telegram or Slack when trades happen.
On top of that: stop-loss, take-profit, position limits and a kill-switch if drawdown gets too high.
We\u2019re now testing fine-tuned models like DeepSeek 3.2 Speciale and Grok 4.1 \u2014 early results look promising, with more consistent performance than generic models.
Would love to hear what features you\u2019d want in an AI trading assistant.
(LLM orchestration & logic framework: LangChain)"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"Show HN: Thinkmoon.ai \u2013 Build your own Alpha Arena, trade crypto with AI Agents"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://demo.thinkmoon.ai/"}},"_tags":["story","author_thinkmoon","story_46207237","show_hn"],"author":"thinkmoon","created_at":"2025-12-09T16:52:37Z","created_at_i":1765299157,"num_comments":0,"objectID":"46207237","points":1,"story_id":46207237,"story_text":"With ThinkMoon you can build your own Alpha Arena: plug in models from OpenRouter, OpenAI, Anthropic (or our custom ones), feed them real-time data (candles, order book, ticker) \u2014 then send them to trade real crypto (via Binance Futures).
You define the rules: risk limits, which coins (BTC, ETH, SOL, XRP\u2026 whatever), even mix different LLMs and compare their performance.
Every move the AI makes is logged \u2014 full prompt, reasoning, market snapshot \u2014 so you can audit or learn from it.
There\u2019s a live dashboard with P&L, open positions, and the AI\u2019s chain of thought. Alerts via Telegram or Slack when trades happen.
On top of that: stop-loss, take-profit, position limits and a kill-switch if drawdown gets too high.
We\u2019re now testing fine-tuned models like DeepSeek 3.2 Speciale and Grok 4.1 \u2014 early results look promising, with more consistent performance than generic models.
Would love to hear what features you\u2019d want in an AI trading assistant.
(LLM orchestration & logic framework: LangChain)","title":"Show HN: Thinkmoon.ai \u2013 Build your own Alpha Arena, trade crypto with AI Agents","updated_at":"2026-03-05T23:07:27Z","url":"https://demo.thinkmoon.ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Houshalter"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Perhaps you could get around the high error rate of speech recognition by having it ask the user again when it's not certain, and use prior probabilities on what they likely said ("wreck a nice beach" is less likely than "recognize speech".)
I wonder if a generative model would be acceptable at producing speech. It works really great on handwriting (http://www.cs.toronto.edu/~graves/handwriting.cgi?text=This+...).
For the underlying AI, this is a great demo of an AI assistant that can do things like you describe (https://www.youtube.com/watch?v=54HpAmzaIbs) . Good natural language processing, it can learn new facts and even asks questions about things it wants to learn more about."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["that"],"value":"Facebook: Do not release your new app feature that listens to users"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"http://action.sumofus.org/a/Facebook-app-taps-phones/"}},"_tags":["comment","author_Houshalter","story_7812812"],"author":"Houshalter","comment_text":"Perhaps you could get around the high error rate of speech recognition by having it ask the user again when it's not certain, and use prior probabilities on what they likely said ("wreck a nice beach" is less likely than "recognize speech".)
I wonder if a generative model would be acceptable at producing speech. It works really great on handwriting (http://www.cs.toronto.edu/~graves/handwriting.cgi?text=This+...).
For the underlying AI, this is a great demo of an AI assistant that can do things like you describe (https://www.youtube.com/watch?v=54HpAmzaIbs) . Good natural language processing, it can learn new facts and even asks questions about things it wants to learn more about.","created_at":"2014-05-28T23:42:40Z","created_at_i":1401320560,"objectID":"7813736","parent_id":7813483,"points":null,"story_id":7812812,"story_title":"Facebook: Do not release your new app feature that listens to users","story_url":"http://action.sumofus.org/a/Facebook-app-taps-phones/","updated_at":"2023-09-06T23:00:42Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Davidepedrotti"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Hi HN,
I\u2019m working on MapBuddy, an AI assistant that helps small, non-technical business owners create an optimized Google Business Profile (GBP). The goal is to make it easy for local businesses (e.g. plumbers, caf\u00e9s, hairdressers) to get found online with a well-built profile\u2014no marketing skills needed.
Right now I only have a landing page and a very early chatbot mockup. Before building further, I want to learn how to get early adopters, validate the idea, and understand how others have approached sales and outreach in the local/SMB space.
Have you successfully launched a tool for small/local businesses? \nHow did you find your first users or paying clients? \nWhat strategies or channels worked best for reaching non-tech business owners?
Thanks so much for any thoughts or stories!"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["an","ai"],"value":"How can I get first users for an AI tool for local businesses?"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://sites.google.com/view/mapboostai/mapbuddy"}},"_tags":["comment","author_Davidepedrotti","story_44607468"],"author":"Davidepedrotti","comment_text":"Hi HN,
I\u2019m working on MapBuddy, an AI assistant that helps small, non-technical business owners create an optimized Google Business Profile (GBP). The goal is to make it easy for local businesses (e.g. plumbers, caf\u00e9s, hairdressers) to get found online with a well-built profile\u2014no marketing skills needed.
Right now I only have a landing page and a very early chatbot mockup. Before building further, I want to learn how to get early adopters, validate the idea, and understand how others have approached sales and outreach in the local/SMB space.
Have you successfully launched a tool for small/local businesses? \nHow did you find your first users or paying clients? \nWhat strategies or channels worked best for reaching non-tech business owners?
Thanks so much for any thoughts or stories!","created_at":"2025-07-18T17:26:30Z","created_at_i":1752859590,"objectID":"44607469","parent_id":44607468,"story_id":44607468,"story_title":"How can I get first users for an AI tool for local businesses?","story_url":"https://sites.google.com/view/mapboostai/mapbuddy","updated_at":"2025-07-18T17:28:41Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"juliennakache"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Medbill AI | Founding Senior Engineer + Founding Designer | Full Time | Remote - US timezones
We are building an AI assistant that helps people with medical billing and health insurance issues. Thanks to AI and new healthcare regulations around patient data access and price transparency, it is now possible for everyone, including you, to have their own assistant who works around the clock to save them time and money on medical bills and health insurance issues.
Reach out if you\u2019re interested in the role or want to learn more about what we\u2019re up to!
https://www.medbill.ai/about"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Who is hiring? (January 2024)"}},"_tags":["comment","author_juliennakache","story_38842977"],"author":"juliennakache","comment_text":"Medbill AI | Founding Senior Engineer + Founding Designer | Full Time | Remote - US timezones
We are building an AI assistant that helps people with medical billing and health insurance issues. Thanks to AI and new healthcare regulations around patient data access and price transparency, it is now possible for everyone, including you, to have their own assistant who works around the clock to save them time and money on medical bills and health insurance issues.
Reach out if you\u2019re interested in the role or want to learn more about what we\u2019re up to!
https://www.medbill.ai/about","created_at":"2024-01-03T03:45:18Z","created_at_i":1704253518,"objectID":"38850457","parent_id":38842977,"story_id":38842977,"story_title":"Ask HN: Who is hiring? (January 2024)","updated_at":"2024-09-20T16:03:13Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"koopuluri"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Congrats on getting this off the ground, and thank you for putting it out there for us to learn from!
I've been curious how Rewind worked under the hood because I've been playing with an idea in my head: an AI assistant that helps you protect your attention.
You would describe the kind of content that you consider a distraction, and any other constraints you have (e.g. "Don't let me watch cat videos unless I'm on a break".
And whenever it sees you watching anything that fits your prompt, it'll pop up on the screen and start a conversation with you to try and understand whether you actually need to consume the content you're looking at.
An AI that intervenes when you're going off track (based purely on how YOU define going off track). Current website blocking approaches aren't useful because they're all-or-nothing. I don't ever want to block entire sites because often there's useful content there relevant for my work. I want to block content on a much more granular level.
And I'd love for an "attention audit" at the end of each day. Attention is our most valuable asset, and I believe protecting it is a worthwhile endeavor... I'd just like some help doing so :)."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Rem: Remember Everything (open source)"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/jasonjmcghee/rem"}},"_tags":["comment","author_koopuluri","story_38787892"],"author":"koopuluri","children":[38790083,38790494],"comment_text":"Congrats on getting this off the ground, and thank you for putting it out there for us to learn from!
I've been curious how Rewind worked under the hood because I've been playing with an idea in my head: an AI assistant that helps you protect your attention.
You would describe the kind of content that you consider a distraction, and any other constraints you have (e.g. "Don't let me watch cat videos unless I'm on a break".
And whenever it sees you watching anything that fits your prompt, it'll pop up on the screen and start a conversation with you to try and understand whether you actually need to consume the content you're looking at.
An AI that intervenes when you're going off track (based purely on how YOU define going off track). Current website blocking approaches aren't useful because they're all-or-nothing. I don't ever want to block entire sites because often there's useful content there relevant for my work. I want to block content on a much more granular level.
And I'd love for an "attention audit" at the end of each day. Attention is our most valuable asset, and I believe protecting it is a worthwhile endeavor... I'd just like some help doing so :).","created_at":"2023-12-28T03:09:42Z","created_at_i":1703732982,"objectID":"38789558","parent_id":38787892,"story_id":38787892,"story_title":"Show HN: Rem: Remember Everything (open source)","story_url":"https://github.com/jasonjmcghee/rem","updated_at":"2024-09-20T16:05:22Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"juliennakache"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Medbill AI | Founding Senior Engineer + Founding Designer | Full Time | Remote - US timezones
We are building an AI assistant that helps people with medical billing and health insurance issues. Thanks to AI and new healthcare regulations around patient data access and price transparency, it is now possible for everyone, including you, to have their own assistant who works around the clock to save them money on medical bills and navigate the disjointed US healthcare system.
Reach out if you\u2019re interested in the role or want to learn more about what we\u2019re up to!
https://www.medbill.ai/about"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Who is hiring? (December 2023)"}},"_tags":["comment","author_juliennakache","story_38490811"],"author":"juliennakache","children":[38493817,38493918],"comment_text":"Medbill AI | Founding Senior Engineer + Founding Designer | Full Time | Remote - US timezones
We are building an AI assistant that helps people with medical billing and health insurance issues. Thanks to AI and new healthcare regulations around patient data access and price transparency, it is now possible for everyone, including you, to have their own assistant who works around the clock to save them money on medical bills and navigate the disjointed US healthcare system.
Reach out if you\u2019re interested in the role or want to learn more about what we\u2019re up to!
https://www.medbill.ai/about","created_at":"2023-12-01T20:32:19Z","created_at_i":1701462739,"objectID":"38491929","parent_id":38490811,"story_id":38490811,"story_title":"Ask HN: Who is hiring? (December 2023)","updated_at":"2024-09-20T15:53:12Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"sneha1995"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"ChatGPT, EmbedAI, Bard, or Aria? Which is the best assistant for you?
I found that each of the four mentioned AI assistants has its unique features and advantages.
ChatGPT provides a robust and comprehensive chatbot tool that can facilitate businesses to improve customer communication, from customer service inquiries to lead generation. GPT-3 powers it, which allows it to understand natural language and context. Additionally, ChatGPT integrates with several messaging platforms such as Facebook Messenger and WhatsApp, and the integration process is effortless.
EmbedAI offers a powerful AI assistant platform that provides businesses with a range of services such as customer support, data extraction, and more. With a user-friendly drag and drop interface that allows you to create conversational flows and analyze conversational data, EmbedAI is useful for businesses that want to leverage AI but don't have the expertise.
Bard is a specialized virtual writing assistant that focuses on helping writers improve their writing. Bard uses AI to analyze your writing and provide suggestions to improve grammar, syntax, tone, and more. Bard can also generate topic ideas and even suggest emotional arcs for short stories.
Aria AI is an AI-powered digital assistant that can schedule meetings, share emails, and manage your to-do list. Aria AI integrates various platforms such as email, calendars, and messaging apps, which makes it easy to use. Aria AI also learns from your preferences and patterns, making it more effective and efficient over time.
In conclusion, which AI assistant is the best for you depends on your particular needs and goals. If you're looking for a chatbot that provides excellent customer support, ChatGPT may be the right choice. However, if you're a writer, Bard may be the ideal choice for improving your writing. On the other hand, EmbedAI may be the best choice for businesses looking for an all-around AI assistant, while Aria is perfect for personal organization."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["assistant"],"value":"ChatGPT, EmbedAI, Bard, or Aria? Which is the best assistant for you?"}},"_tags":["story","author_sneha1995","story_37431642","ask_hn"],"author":"sneha1995","created_at":"2023-09-08T09:55:39Z","created_at_i":1694166939,"num_comments":0,"objectID":"37431642","points":5,"story_id":37431642,"story_text":"ChatGPT, EmbedAI, Bard, or Aria? Which is the best assistant for you?
I found that each of the four mentioned AI assistants has its unique features and advantages.
ChatGPT provides a robust and comprehensive chatbot tool that can facilitate businesses to improve customer communication, from customer service inquiries to lead generation. GPT-3 powers it, which allows it to understand natural language and context. Additionally, ChatGPT integrates with several messaging platforms such as Facebook Messenger and WhatsApp, and the integration process is effortless.
EmbedAI offers a powerful AI assistant platform that provides businesses with a range of services such as customer support, data extraction, and more. With a user-friendly drag and drop interface that allows you to create conversational flows and analyze conversational data, EmbedAI is useful for businesses that want to leverage AI but don't have the expertise.
Bard is a specialized virtual writing assistant that focuses on helping writers improve their writing. Bard uses AI to analyze your writing and provide suggestions to improve grammar, syntax, tone, and more. Bard can also generate topic ideas and even suggest emotional arcs for short stories.
Aria AI is an AI-powered digital assistant that can schedule meetings, share emails, and manage your to-do list. Aria AI integrates various platforms such as email, calendars, and messaging apps, which makes it easy to use. Aria AI also learns from your preferences and patterns, making it more effective and efficient over time.
In conclusion, which AI assistant is the best for you depends on your particular needs and goals. If you're looking for a chatbot that provides excellent customer support, ChatGPT may be the right choice. However, if you're a writer, Bard may be the ideal choice for improving your writing. On the other hand, EmbedAI may be the best choice for businesses looking for an all-around AI assistant, while Aria is perfect for personal organization.","title":"ChatGPT, EmbedAI, Bard, or Aria? Which is the best assistant for you?","updated_at":"2024-09-20T15:00:49Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"troyanstudios"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Hi HN! I\u2019m a young founder working on a new wearable device that\u2019s a bit different from the current crop. Think of it as a wearable AI companion that goes beyond single-use cases like memory aids or smart glasses for the visually impaired.
The device itself is simple \u2013 a small camera, mic, and speaker that attaches to your shirt and connects to your phone. But the real innovation is in the software: a flexible AI helper that adapts to your specific needs, whether that\u2019s helping a colorblind painter identify colors, a person with ADHD who needs reminders to stay on task, or someone who often misplaces items and needs help tracking them down.
Here\u2019s where it gets exciting: instead of requiring technical skills to set up, the AI "learns" what you need from simple prompts. Imagine telling it, "I always lose my keys \u2013 can you help?" and it finds or creates the right helper. It also \u201cborrows\u201d and shares skills as needed, creating a highly personalized, adaptive assistant without complex setup or coding.
I\u2019m curious to know: what small problems or annoyances would you want solved by a device like this? What situations or daily challenges could an AI wearable assistant help with that current devices don\u2019t? Also, if anyone\u2019s interested in collaborating, let\u2019s connect!
Thanks for any input \u2013 your ideas could make this a product that truly makes life easier in ways that matter."},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["an","ai"],"value":"Feedback Needed: Building an AI Wearable for Solving Everyday Problems"}},"_tags":["story","author_troyanstudios","story_41955515","ask_hn"],"author":"troyanstudios","created_at":"2024-10-26T15:51:25Z","created_at_i":1729957885,"num_comments":0,"objectID":"41955515","points":1,"story_id":41955515,"story_text":"Hi HN! I\u2019m a young founder working on a new wearable device that\u2019s a bit different from the current crop. Think of it as a wearable AI companion that goes beyond single-use cases like memory aids or smart glasses for the visually impaired.
The device itself is simple \u2013 a small camera, mic, and speaker that attaches to your shirt and connects to your phone. But the real innovation is in the software: a flexible AI helper that adapts to your specific needs, whether that\u2019s helping a colorblind painter identify colors, a person with ADHD who needs reminders to stay on task, or someone who often misplaces items and needs help tracking them down.
Here\u2019s where it gets exciting: instead of requiring technical skills to set up, the AI "learns" what you need from simple prompts. Imagine telling it, "I always lose my keys \u2013 can you help?" and it finds or creates the right helper. It also \u201cborrows\u201d and shares skills as needed, creating a highly personalized, adaptive assistant without complex setup or coding.
I\u2019m curious to know: what small problems or annoyances would you want solved by a device like this? What situations or daily challenges could an AI wearable assistant help with that current devices don\u2019t? Also, if anyone\u2019s interested in collaborating, let\u2019s connect!
Thanks for any input \u2013 your ideas could make this a product that truly makes life easier in ways that matter.","title":"Feedback Needed: Building an AI Wearable for Solving Everyday Problems","updated_at":"2024-10-26T15:53:17Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"pscanf"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"> Imagine a world where your data isn\u2019t trapped in distant data centers. Instead, it\u2019s close to home\u2014in a secure data wallet or pod, under your control. Now imagine pairing that with a loyal personal AI assistant, a private, local tool that lives with you, learns from you (with your permission), and acts on your behalf. Your AI. Not theirs.
This is almost exactly what I say on the landing page\u00b9 of the product I'm building (an open-source personal database, with an AI assistant on top).
I want to believe this can be a reality, and I'm trying to make it become one, but there are two significant challenges:
1. AI = cloud. Taking my app as an example, it'll be at least 2 years before consumer hardware will be able to run the smallest model that performs somewhat decently (gpt-oss-20b). And of course in 2 years that model will be beyond obsolete. Would a regular user pay the price of a subpar experience in order to get data ownership and privacy? It's a very hard sell.
2. Apps/services are very jealous of their users' data. As a user I have to jump through incredible hoops just to get a point-in-time copy of my data. If I can get that at all. There is no incentive for apps to allow their users to own their data. On the contrary, it's better if they don't, so they remain locked in the app. Also, regular Joe and Jane users are not really asking to have access to their data, because there's no benefit for them either.
That is, I think, the key to overcome challenge #2: giving regular Joes and Janes an immediate and obvious benefit. If they see that only by owning their data they can do $INCREDIBLY_VALUABLE_THING, then they will themselves start demanding companies access to it, or they will jump through the hoops to get it. (That's the way I'm going about it. I'm nowhere near the end goal, of course, but I see promising results.\u00b2)
I have no idea how to overcome challenge #1 yet. Mainly because currently there aren't really any big downsides to using cloud models. Or, at least, we haven't seen them yet. Maybe if OpenAI starts injecting ads in GPT-8 responses, people will reconsider using a "stupider" but local, ad-free model.
\u00b9 https://superego.dev/
\u00b2 https://pscanf.com/s/350/"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","that"],"value":"AI is killing privacy. We can't let that happen"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://www.fastcompany.com/91435189/ai-privacy-openai-tracking-apps"}},"_tags":["comment","author_pscanf","story_45947434"],"author":"pscanf","comment_text":"> Imagine a world where your data isn\u2019t trapped in distant data centers. Instead, it\u2019s close to home\u2014in a secure data wallet or pod, under your control. Now imagine pairing that with a loyal personal AI assistant, a private, local tool that lives with you, learns from you (with your permission), and acts on your behalf. Your AI. Not theirs.
This is almost exactly what I say on the landing page\u00b9 of the product I'm building (an open-source personal database, with an AI assistant on top).
I want to believe this can be a reality, and I'm trying to make it become one, but there are two significant challenges:
1. AI = cloud. Taking my app as an example, it'll be at least 2 years before consumer hardware will be able to run the smallest model that performs somewhat decently (gpt-oss-20b). And of course in 2 years that model will be beyond obsolete. Would a regular user pay the price of a subpar experience in order to get data ownership and privacy? It's a very hard sell.
2. Apps/services are very jealous of their users' data. As a user I have to jump through incredible hoops just to get a point-in-time copy of my data. If I can get that at all. There is no incentive for apps to allow their users to own their data. On the contrary, it's better if they don't, so they remain locked in the app. Also, regular Joe and Jane users are not really asking to have access to their data, because there's no benefit for them either.
That is, I think, the key to overcome challenge #2: giving regular Joes and Janes an immediate and obvious benefit. If they see that only by owning their data they can do $INCREDIBLY_VALUABLE_THING, then they will themselves start demanding companies access to it, or they will jump through the hoops to get it. (That's the way I'm going about it. I'm nowhere near the end goal, of course, but I see promising results.\u00b2)
I have no idea how to overcome challenge #1 yet. Mainly because currently there aren't really any big downsides to using cloud models. Or, at least, we haven't seen them yet. Maybe if OpenAI starts injecting ads in GPT-8 responses, people will reconsider using a "stupider" but local, ad-free model.
\u00b9 https://superego.dev/
\u00b2 https://pscanf.com/s/350/","created_at":"2025-11-16T21:22:41Z","created_at_i":1763328161,"objectID":"45948521","parent_id":45947434,"story_id":45947434,"story_title":"AI is killing privacy. We can't let that happen","story_url":"https://www.fastcompany.com/91435189/ai-privacy-openai-tracking-apps","updated_at":"2026-03-05T23:00:49Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"xilong88"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Hello everyone!\nI\u2019m a developer working on *LogAI* \u2014 an AI-powered super memory assistant that helps you truly *remember*, *reflect*, and *act* on the things that matter most to you.
Like many of you, I\u2019ve always been a note-taker \u2014 tracking my *tasks*, *ideas*, *daily life*, *health data*, *expenses*, and more.\nBut existing apps are often too *scattered*, too *similar*, or too *complicated* to actually help.\nHere are some of the pain points I\u2019ve encountered:
### Problems:
*Fragmented tools, bloated features*\nWant to track everything?\nYou might end up with 5+ apps, 5+ subscriptions, and 10+ different interface logics to learn.
*Lots of data, no insight*\nYour data is scattered across apps \u2014 hard to search and impossible to connect.\nAnd so-called \u201cinsights\u201d? Usually just a few charts \u2014 no real guidance or meaning.
*High barrier to logging*\nEvery app has its own way to record things.\nMost (I\u2019ve counted \u2014 over 70%) don\u2019t support *AI input* at all.
I\u2019ve been searching for a tool that could truly *quantify myself* \u2014 and couldn\u2019t find one.\nSo I decided to *build it myself*.
---
### Introducing LogAI
*LogAI* is the \u201csecond brain\u201d I\u2019ve always wanted \u2014 a tool that lets you record *anything, in any format, in the shortest time possible.*
You just need to type or say something like:
> \u201cRan for 50 minutes at 8 AM\u201d\n> \u201cStudied programming for 2 hours\u201d\n> \u201cTook an antibiotic at 4 PM\u201d
That\u2019s it.\nNo complex forms, no extra steps.
Whether it\u2019s *food*, *fitness*, *finance*, *medication*, *dreams*, *ideas*, *photos*, or *voice notes*, LogAI can automatically handle them \u2014 for example, recognizing calories or parsing medical reports.
More importantly, it gives you *feedback*:\nReal insights, emotional understanding, and actionable suggestions, such as:
> \u201cAnalyze my workout patterns and suggest improvements.\u201d\n> \u201cDid I take my medicine today?\u201d\n> \u201cHow\u2019s my study progress this week?\u201d
---
### Where are we now?
LogAI is still in development.\nThe past few months have been intense \u2014 countless late nights, debugging, rewriting, scrapping and rebuilding.
Why?\nBecause we don\u2019t just want to build something that works \u2014\nwe want to build something that feels *truly good to use*.
And making a tool that\u2019s *universal, intelligent, and delightful* is incredibly hard.
[https://logai.chat](https://logai.chat)
I\u2019d love to answer your questions or hear your feedback \u2014 *version 1.0 is now released*, and you\u2019re welcome to try it out!
Thank you for reading \u2014 I can\u2019t wait to use it together with you!\nFeel free to leave a comment or share any suggestions!"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","assistant"],"value":"I Built LogAI: The AI Memory Assistant I Always Wanted"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.logai.chat/"}},"_tags":["comment","author_xilong88","story_45633466"],"author":"xilong88","comment_text":"Hello everyone!\nI\u2019m a developer working on *LogAI* \u2014 an AI-powered super memory assistant that helps you truly *remember*, *reflect*, and *act* on the things that matter most to you.
Like many of you, I\u2019ve always been a note-taker \u2014 tracking my *tasks*, *ideas*, *daily life*, *health data*, *expenses*, and more.\nBut existing apps are often too *scattered*, too *similar*, or too *complicated* to actually help.\nHere are some of the pain points I\u2019ve encountered:
### Problems:
*Fragmented tools, bloated features*\nWant to track everything?\nYou might end up with 5+ apps, 5+ subscriptions, and 10+ different interface logics to learn.
*Lots of data, no insight*\nYour data is scattered across apps \u2014 hard to search and impossible to connect.\nAnd so-called \u201cinsights\u201d? Usually just a few charts \u2014 no real guidance or meaning.
*High barrier to logging*\nEvery app has its own way to record things.\nMost (I\u2019ve counted \u2014 over 70%) don\u2019t support *AI input* at all.
I\u2019ve been searching for a tool that could truly *quantify myself* \u2014 and couldn\u2019t find one.\nSo I decided to *build it myself*.
---
### Introducing LogAI
*LogAI* is the \u201csecond brain\u201d I\u2019ve always wanted \u2014 a tool that lets you record *anything, in any format, in the shortest time possible.*
You just need to type or say something like:
> \u201cRan for 50 minutes at 8 AM\u201d\n> \u201cStudied programming for 2 hours\u201d\n> \u201cTook an antibiotic at 4 PM\u201d
That\u2019s it.\nNo complex forms, no extra steps.
Whether it\u2019s *food*, *fitness*, *finance*, *medication*, *dreams*, *ideas*, *photos*, or *voice notes*, LogAI can automatically handle them \u2014 for example, recognizing calories or parsing medical reports.
More importantly, it gives you *feedback*:\nReal insights, emotional understanding, and actionable suggestions, such as:
> \u201cAnalyze my workout patterns and suggest improvements.\u201d\n> \u201cDid I take my medicine today?\u201d\n> \u201cHow\u2019s my study progress this week?\u201d
---
### Where are we now?
LogAI is still in development.\nThe past few months have been intense \u2014 countless late nights, debugging, rewriting, scrapping and rebuilding.
Why?\nBecause we don\u2019t just want to build something that works \u2014\nwe want to build something that feels *truly good to use*.
And making a tool that\u2019s *universal, intelligent, and delightful* is incredibly hard.
[https://logai.chat](https://logai.chat)
I\u2019d love to answer your questions or hear your feedback \u2014 *version 1.0 is now released*, and you\u2019re welcome to try it out!
Thank you for reading \u2014 I can\u2019t wait to use it together with you!\nFeel free to leave a comment or share any suggestions!","created_at":"2025-10-19T11:23:25Z","created_at_i":1760873005,"objectID":"45633467","parent_id":45633466,"story_id":45633466,"story_title":"I Built LogAI: The AI Memory Assistant I Always Wanted","story_url":"https://www.logai.chat/","updated_at":"2026-03-05T22:50:23Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mcello"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"AmplifyHumans \u2014 AI companions that grow with you, learn from your world, and 10\u00d7 your thinking + execution.
Think: second brain \u00d7 chief of staff \u00d7 assistant on steroids.
Join early \u2192 www.amplifyhumans.com
In a few days, I\u2019m sending a 5-min survey to early users. You\u2019ll get:\n Exclusive insights from early adopters\n Priority access to the private MVP demo
This isn\u2019t a demo-day chatbot.\nIt\u2019s a long-term bet to build tools that compound you.
What would you want in an AI that truly amplifies you?"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["want","an","ai","that"],"value":"Would You Want an AI That Amplifies You (Not Just Chats)?"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.amplifyhumans.com"}},"_tags":["comment","author_mcello","story_44327496"],"author":"mcello","comment_text":"AmplifyHumans \u2014 AI companions that grow with you, learn from your world, and 10\u00d7 your thinking + execution.
Think: second brain \u00d7 chief of staff \u00d7 assistant on steroids.
Join early \u2192 www.amplifyhumans.com
In a few days, I\u2019m sending a 5-min survey to early users. You\u2019ll get:\n Exclusive insights from early adopters\n Priority access to the private MVP demo
This isn\u2019t a demo-day chatbot.\nIt\u2019s a long-term bet to build tools that compound you.
What would you want in an AI that truly amplifies you?","created_at":"2025-06-20T13:20:25Z","created_at_i":1750425625,"objectID":"44327497","parent_id":44327496,"story_id":44327496,"story_title":"Would You Want an AI That Amplifies You (Not Just Chats)?","story_url":"https://www.amplifyhumans.com","updated_at":"2025-06-20T13:25:05Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Caritaspax"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"An AI-First Linux Desktop That Treats the User Like a Human
One-Sentence Pitch\nA Linux-based operating environment where a user can simply ask for the software or task they want, and an AI agent safely installs, configures, and operates tools alongside them \u2014 not as a controller, but as a hands-on partner.
The Vibe (Important)\nSlightly cyberpunk, but not dystopian
More \u201cdeck hacker with a conscience\u201d than megacorp AI
The machine feels helpful, not omniscient
Power without pretense
The Problem\nModern computers still assume users:\nKnow what software exists
Know how to install it
Know how to configure it
Know how to use it
This is false.\nPeople know what they want to do, not which package, dependency tree, or UI ritual will get them there.\nCurrent AI assistants help with answers, but not ownership of the system itself.
The Core Idea\nReplace the desktop metaphor (\u201cfiles, apps, menus\u201d) with an intent-first system:\n\u201cI want to edit audio like a podcast.\u201d\n \u201cI need to batch-resize these photos.\u201d\n \u201cI want to model this idea and see what happens.\u201d\nThe system figures out how.
Architecture (High Level)\nLinux stays exactly where it belongs:\n Quiet, stable, invisible.\nOn top of it:\n1. Intent Layer (LLM Agent)\nInterprets natural language goals
Plans steps
Asks clarifying questions when necessary
Never assumes permission
2. Tool Orchestration Layer\nSearches local system first
Then trusted repositories
Then source builds (sandboxed)
Chooses between:
Native packages
Flatpak/AppImage
Containers
Explains choices briefly
3. Safety & Trust Layer\nExplicit approvals for system changes
Snapshots before risky operations
Full action log:
\u201cHere\u2019s what I changed. Want to undo it?\u201d
4. Co-Pilot Interface\nPersistent side panel or overlay
Knows what apps are open
Can:
Operate CLI tools
Use application APIs
Guide UI interactions
Never hides what it\u2019s doing
What Makes This Different\nThis is not:\nA chatbot bolted onto a desktop
A voice assistant guessing commands
A system that \u201cjust does things\u201d
This is:\nA translator between human intent and machine reality
A system that treats the user as author, not operator
A refusal to make people learn the machine\u2019s language
Example Flow\nUser:\n\u201cI want to clean up background noise in this recording.\u201d\nSystem:\n\u201cI can use Audacity or a command-line tool. Audacity is already installed.\n Want me to do it, or walk you through?\u201d\nUser:\n\u201cDo it.\u201d\nSystem:\nOpens Audacity
Applies noise reduction
Plays before/after
Waits
Design Principles (Non-Negotiable)\nThe user is never surprised
Nothing irreversible without consent
The AI explains enough, not everything
The system never claims authority
Power is visible, not magical
Why This Matters\nComputers are incredible tools that still feel like:\nBureaucratic systems
Ritual machines
Tests of worthiness
This re-centers computing around human intention, not technical literacy.\nIt makes powerful tools approachable without making them shallow.
The Sci-Fi Payoff\nYou don\u2019t \u201copen apps.\u201d\nYou sit down and say:\n\u201cLet\u2019s work.\u201d\nAnd the machine listens.
Final Note\nI can\u2019t build this.\nBut someone absolutely should.\nIf you\u2019re an engineer reading this:\nPlease don\u2019t make it creepy
Please don\u2019t monetize the soul out of it
Please let the machine remain a servant
Somebody build this."},"title":{"matchLevel":"none","matchedWords":[],"value":"Somebody Build This"}},"_tags":["story","author_Caritaspax","story_46494866","ask_hn"],"author":"Caritaspax","children":[46494903,46494953,46495002,46508518],"created_at":"2026-01-05T03:01:59Z","created_at_i":1767582119,"num_comments":4,"objectID":"46494866","points":3,"story_id":46494866,"story_text":"An AI-First Linux Desktop That Treats the User Like a Human
One-Sentence Pitch\nA Linux-based operating environment where a user can simply ask for the software or task they want, and an AI agent safely installs, configures, and operates tools alongside them \u2014 not as a controller, but as a hands-on partner.
The Vibe (Important)\nSlightly cyberpunk, but not dystopian
More \u201cdeck hacker with a conscience\u201d than megacorp AI
The machine feels helpful, not omniscient
Power without pretense
The Problem\nModern computers still assume users:\nKnow what software exists
Know how to install it
Know how to configure it
Know how to use it
This is false.\nPeople know what they want to do, not which package, dependency tree, or UI ritual will get them there.\nCurrent AI assistants help with answers, but not ownership of the system itself.
The Core Idea\nReplace the desktop metaphor (\u201cfiles, apps, menus\u201d) with an intent-first system:\n\u201cI want to edit audio like a podcast.\u201d\n \u201cI need to batch-resize these photos.\u201d\n \u201cI want to model this idea and see what happens.\u201d\nThe system figures out how.
Architecture (High Level)\nLinux stays exactly where it belongs:\n Quiet, stable, invisible.\nOn top of it:\n1. Intent Layer (LLM Agent)\nInterprets natural language goals
Plans steps
Asks clarifying questions when necessary
Never assumes permission
2. Tool Orchestration Layer\nSearches local system first
Then trusted repositories
Then source builds (sandboxed)
Chooses between:
Native packages
Flatpak/AppImage
Containers
Explains choices briefly
3. Safety & Trust Layer\nExplicit approvals for system changes
Snapshots before risky operations
Full action log:
\u201cHere\u2019s what I changed. Want to undo it?\u201d
4. Co-Pilot Interface\nPersistent side panel or overlay
Knows what apps are open
Can:
Operate CLI tools
Use application APIs
Guide UI interactions
Never hides what it\u2019s doing
What Makes This Different\nThis is not:\nA chatbot bolted onto a desktop
A voice assistant guessing commands
A system that \u201cjust does things\u201d
This is:\nA translator between human intent and machine reality
A system that treats the user as author, not operator
A refusal to make people learn the machine\u2019s language
Example Flow\nUser:\n\u201cI want to clean up background noise in this recording.\u201d\nSystem:\n\u201cI can use Audacity or a command-line tool. Audacity is already installed.\n Want me to do it, or walk you through?\u201d\nUser:\n\u201cDo it.\u201d\nSystem:\nOpens Audacity
Applies noise reduction
Plays before/after
Waits
Design Principles (Non-Negotiable)\nThe user is never surprised
Nothing irreversible without consent
The AI explains enough, not everything
The system never claims authority
Power is visible, not magical
Why This Matters\nComputers are incredible tools that still feel like:\nBureaucratic systems
Ritual machines
Tests of worthiness
This re-centers computing around human intention, not technical literacy.\nIt makes powerful tools approachable without making them shallow.
The Sci-Fi Payoff\nYou don\u2019t \u201copen apps.\u201d\nYou sit down and say:\n\u201cLet\u2019s work.\u201d\nAnd the machine listens.
Final Note\nI can\u2019t build this.\nBut someone absolutely should.\nIf you\u2019re an engineer reading this:\nPlease don\u2019t make it creepy
Please don\u2019t monetize the soul out of it
Please let the machine remain a servant
Somebody build this.","title":"Somebody Build This","updated_at":"2026-03-05T23:19:40Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jrussino"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"I agree with you, and I cringed a little when I read the following from the OP:
> There are some fundamental reasons why conversational 3rd party platforms are hard.
In my mind the big fundamental problem here is the "3rd party". I'd love to have an "AI assistant" or an "AI buddy" that could watch everything I do and say and write and really get to know me super well... as long as I can be confident that I own and control everything it observes and learns. I sure as hell don't want a 3rd party involved! But alas, I don't see a way we get there that doesn't involve Amazon or Meta or Google or OpenAI sitting between me and my "AI" tools, at least in the short run."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Amazon virtually kills efforts to develop Alexa Skills"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://arstechnica.com/gadgets/2024/04/amazon-virtually-kills-efforts-to-develop-alexa-skills-disappointing-dozens/"}},"_tags":["comment","author_jrussino","story_40008170"],"author":"jrussino","children":[40010805],"comment_text":"I agree with you, and I cringed a little when I read the following from the OP:
> There are some fundamental reasons why conversational 3rd party platforms are hard.
In my mind the big fundamental problem here is the "3rd party". I'd love to have an "AI assistant" or an "AI buddy" that could watch everything I do and say and write and really get to know me super well... as long as I can be confident that I own and control everything it observes and learns. I sure as hell don't want a 3rd party involved! But alas, I don't see a way we get there that doesn't involve Amazon or Meta or Google or OpenAI sitting between me and my "AI" tools, at least in the short run.","created_at":"2024-04-12T04:42:02Z","created_at_i":1712896922,"objectID":"40009447","parent_id":40009262,"story_id":40008170,"story_title":"Amazon virtually kills efforts to develop Alexa Skills","story_url":"https://arstechnica.com/gadgets/2024/04/amazon-virtually-kills-efforts-to-develop-alexa-skills-disappointing-dozens/","updated_at":"2024-09-20T16:45:19Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mapt"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"> Technical people often react strongly against bundling concerns that can be kept separate. But the product view beats the architectural view every time. The question isn't "should we or should we not bundle these concerns?" but rather "given that these concerns are bundled up in actual real-world use and perception anyway, what underlying design can we come up with so that the resulting architecture makes sense?"
Visceral reaction: Outside of FLOSS, technical people often react strongly against bundling concerns that can be kept separate because bundling them is often the first step to an over-aggressive capitalism entirely consuming their utility. The "Feed" that eventually becomes a heavy advertising venue or even a brainwashing tactic, has destroyed Twitter, and Facebook, and others as user experiences, rendered things like Youtube into potent infohazards, become a foundation for post-free-speech worlds in which the Platform is expected to Moderate Content for rightthink because the Platform is (for the sake of engagement!) deciding what to show you in an Editorial Capacity. The corruption of Google Search, or the Amazon review functionality, or then the Amazon search functionality, is a major short-term loss for human agency at least as big as the writer envisions, and it's being done because bundling different concerns provides an opening, gives the corporate board an erection during quarterly P&L briefings. Over the Possibilities.
If you care about the user experience, consider how long my grandfather's wrench was permitted to remain a wrench, rather than autonomously transforming into a screwdriver, or a brick, or a nice welcoming block of cheese, or a magazine subscription, or a bonfire, or Ebola. The owning entity only has to learn how to use the wrench once a generation; The interface does not drastically change to combine my love life, my choice in cereal, and my ability to tighten bolts. I have to re-learn some online tools once per YEAR because somebody is bundling something in a way that is in the short term slightly more profitable. I have a closet full of useful tools that don't exist on my cognitive plane any more and I'm not sure I want to investigate deeply enough to figure out what cosmic horror they became. In order to preserve my agency, I need to be able to flip back and forth between those tools and summon up capacities that I have not engaged in for several years; The state of web applications (and by extension, the browser) makes me as a tool-using ape feel like I have dementia with the number of holes that now exist in my knowledge versus my past self.
You don't need a bigger browser. You need a predictable environment that you can buy and own, that doesn't tear itself to shreds when you're not actively handing it a geometrically larger amount of money every five minutes. Most layering violations (outside of FLOSS) that you observe are trying to pick your pocket with one hand and scramble your neocortex ("The way you THOUGHT that the platform worked") with an icepick with the other.
No, my use of tabs may not be Technically Optimal. It is a way of organizing information. But it's predictable, and it's within my cognitive grasp, and it lets me do a great many things without entirely losing track of them. I don't want to have to relearn an entire means of organizing my information because you thought that my tabs belonged in your bookmark service\u2020, or in your AI personal assistant\u2020, or that they should be tidy and Bring You Joy. I don't want somebody to rearrange the papers on my desk; That would be profoundly disturbing because it breaks the model for how I think, how I predict things, and how I remain effective.
\u2020Which you DEFINITELY will refrain from charging a subscription for. For a year, maybe even two, before seizing that chunk of my exocortex."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"You're Gonna Need a Bigger Browser"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://berjon.com/bigger-browser/"}},"_tags":["comment","author_mapt","story_38144530"],"author":"mapt","children":[38146881],"comment_text":"> Technical people often react strongly against bundling concerns that can be kept separate. But the product view beats the architectural view every time. The question isn't "should we or should we not bundle these concerns?" but rather "given that these concerns are bundled up in actual real-world use and perception anyway, what underlying design can we come up with so that the resulting architecture makes sense?"
Visceral reaction: Outside of FLOSS, technical people often react strongly against bundling concerns that can be kept separate because bundling them is often the first step to an over-aggressive capitalism entirely consuming their utility. The "Feed" that eventually becomes a heavy advertising venue or even a brainwashing tactic, has destroyed Twitter, and Facebook, and others as user experiences, rendered things like Youtube into potent infohazards, become a foundation for post-free-speech worlds in which the Platform is expected to Moderate Content for rightthink because the Platform is (for the sake of engagement!) deciding what to show you in an Editorial Capacity. The corruption of Google Search, or the Amazon review functionality, or then the Amazon search functionality, is a major short-term loss for human agency at least as big as the writer envisions, and it's being done because bundling different concerns provides an opening, gives the corporate board an erection during quarterly P&L briefings. Over the Possibilities.
If you care about the user experience, consider how long my grandfather's wrench was permitted to remain a wrench, rather than autonomously transforming into a screwdriver, or a brick, or a nice welcoming block of cheese, or a magazine subscription, or a bonfire, or Ebola. The owning entity only has to learn how to use the wrench once a generation; The interface does not drastically change to combine my love life, my choice in cereal, and my ability to tighten bolts. I have to re-learn some online tools once per YEAR because somebody is bundling something in a way that is in the short term slightly more profitable. I have a closet full of useful tools that don't exist on my cognitive plane any more and I'm not sure I want to investigate deeply enough to figure out what cosmic horror they became. In order to preserve my agency, I need to be able to flip back and forth between those tools and summon up capacities that I have not engaged in for several years; The state of web applications (and by extension, the browser) makes me as a tool-using ape feel like I have dementia with the number of holes that now exist in my knowledge versus my past self.
You don't need a bigger browser. You need a predictable environment that you can buy and own, that doesn't tear itself to shreds when you're not actively handing it a geometrically larger amount of money every five minutes. Most layering violations (outside of FLOSS) that you observe are trying to pick your pocket with one hand and scramble your neocortex ("The way you THOUGHT that the platform worked") with an icepick with the other.
No, my use of tabs may not be Technically Optimal. It is a way of organizing information. But it's predictable, and it's within my cognitive grasp, and it lets me do a great many things without entirely losing track of them. I don't want to have to relearn an entire means of organizing my information because you thought that my tabs belonged in your bookmark service\u2020, or in your AI personal assistant\u2020, or that they should be tidy and Bring You Joy. I don't want somebody to rearrange the papers on my desk; That would be profoundly disturbing because it breaks the model for how I think, how I predict things, and how I remain effective.
\u2020Which you DEFINITELY will refrain from charging a subscription for. For a year, maybe even two, before seizing that chunk of my exocortex.","created_at":"2023-11-05T00:36:47Z","created_at_i":1699144607,"objectID":"38146764","parent_id":38144530,"story_id":38144530,"story_title":"You're Gonna Need a Bigger Browser","story_url":"https://berjon.com/bigger-browser/","updated_at":"2025-08-14T22:53:53Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"pancomplex"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Hey HN! My name is Christian, and I\u2019m the co-founder of https://frigade.ai. We\u2019ve built an AI agent that automatically learns how to use any web-based product, and in turn guides users directly in the UI, automatically generates documentation, and even takes actions on a user\u2019s behalf. Think of it as Clippy from the old MS Office. But on steroids. And actually helpful.
You can see the agent and tool-calling SDK in action here: https://www.youtube.com/watch?v=UPe0t3A1Vpg
How is this different from other AI customer support products?
Most AI "copilots" are really just glorified chatbots. They skim your help center and spit out some nonspecific bullet points. Basically some \u2018hopes and prayers\u2019 that your users will figure it out. Ultimately, this puts the burden on the user to follow through. And assumes companies are keeping their help center up-to-date with every product change. That means constant screenshots of new product UI or features for accurate instructions.These solutions leverage only a fraction of what\u2019s possible with AI, which can now reason about software interfaces extensively.
With Frigade AI, we guide the user directly in the product and build on-demand tours based on the current user\u2019s state and context. The agents can also take actions immediately on a user\u2019s behalf, e.g. inviting a colleague to a workspace or retrieving billing information (via our tool calling SDK).
This was only made possible recently. The latest frontier models (GPT 4.1, Claude 4, Gemini 2.5, etc.) are able to reason about UIs and workflows in a way that simply didn\u2019t work just 6 months ago. That\u2019s why we\u2019re so excited to bring this technology to the forefront of complex legacy SaaS applications that are not yet AI enabled.
How does it work?
1. Invite agent@frigade.ai to your product. You can send multiple invitations based on distinct roles.
2. Our agent automatically explores and reasons about your application.
3. Attach any existing help center resources or training documentation to supplement the agent\u2019s understanding. Totally optional.
4. Install the agent assistant Javascript snippet (just a few lines).
5. That\u2019s it. Your users can now start asking questions and get on demand product tours and questions answered in real time without any overhead.
This process takes only a few minutes. Once running, you can improve the agent by rating and providing feedback to the responses it provides. If you want to integrate further, you can also hook up your own code to our tool calling SDK to enable the agent to look up customer info, issue refunds, etc. directly. These calls can be made with just a few lines of code by describing the tool and its parameters in natural language and passing a single Javascript promise (e.g. make an API call, call a function in your app, etc.).
Would love to hear what the HN crowd thinks about this approach! Are you building your own AI agent from scratch, or looking to embed one off the shelf?"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["an","ai","that","learns"],"value":"Show HN: An AI agent that learns your product and guides your users"},"url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://frigade.ai"}},"_tags":["story","author_pancomplex","story_44733892","show_hn"],"author":"pancomplex","children":[44737042,44737098,44737249,44737413,44737757,44738363,44740302,44741380,44741549,44742163,44742303,44744374,44746019],"created_at":"2025-07-30T13:24:50Z","created_at_i":1753881890,"num_comments":31,"objectID":"44733892","points":69,"story_id":44733892,"story_text":"Hey HN! My name is Christian, and I\u2019m the co-founder of https://frigade.ai. We\u2019ve built an AI agent that automatically learns how to use any web-based product, and in turn guides users directly in the UI, automatically generates documentation, and even takes actions on a user\u2019s behalf. Think of it as Clippy from the old MS Office. But on steroids. And actually helpful.
You can see the agent and tool-calling SDK in action here: https://www.youtube.com/watch?v=UPe0t3A1Vpg
How is this different from other AI customer support products?
Most AI "copilots" are really just glorified chatbots. They skim your help center and spit out some nonspecific bullet points. Basically some \u2018hopes and prayers\u2019 that your users will figure it out. Ultimately, this puts the burden on the user to follow through. And assumes companies are keeping their help center up-to-date with every product change. That means constant screenshots of new product UI or features for accurate instructions.These solutions leverage only a fraction of what\u2019s possible with AI, which can now reason about software interfaces extensively.
With Frigade AI, we guide the user directly in the product and build on-demand tours based on the current user\u2019s state and context. The agents can also take actions immediately on a user\u2019s behalf, e.g. inviting a colleague to a workspace or retrieving billing information (via our tool calling SDK).
This was only made possible recently. The latest frontier models (GPT 4.1, Claude 4, Gemini 2.5, etc.) are able to reason about UIs and workflows in a way that simply didn\u2019t work just 6 months ago. That\u2019s why we\u2019re so excited to bring this technology to the forefront of complex legacy SaaS applications that are not yet AI enabled.
How does it work?
1. Invite agent@frigade.ai to your product. You can send multiple invitations based on distinct roles.
2. Our agent automatically explores and reasons about your application.
3. Attach any existing help center resources or training documentation to supplement the agent\u2019s understanding. Totally optional.
4. Install the agent assistant Javascript snippet (just a few lines).
5. That\u2019s it. Your users can now start asking questions and get on demand product tours and questions answered in real time without any overhead.
This process takes only a few minutes. Once running, you can improve the agent by rating and providing feedback to the responses it provides. If you want to integrate further, you can also hook up your own code to our tool calling SDK to enable the agent to look up customer info, issue refunds, etc. directly. These calls can be made with just a few lines of code by describing the tool and its parameters in natural language and passing a single Javascript promise (e.g. make an API call, call a function in your app, etc.).
Would love to hear what the HN crowd thinks about this approach! Are you building your own AI agent from scratch, or looking to embed one off the shelf?","title":"Show HN: An AI agent that learns your product and guides your users","updated_at":"2025-08-20T15:24:33Z","url":"https://frigade.ai"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"dialupdotnet"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Hey HN, I didn't want my Gateway 4DX2-66 from 1993 to be left out of the AI revolution, so I built an AI Assistant for Windows 3.1, based on the OpenAI API.
The most interesting parts of building this were:
- Getting TLS 1.2 and 1.3 working on Windows 3.1 so that WinGPT could directly connect to OpenAI's server without relying on a modern machine for TLS termination.
- Learning about the memory segmentation architecture on 16-bit Windows, including the difference between far and near pointers.
- Building the UI in plain C code with the Windows API and limited selection of controls in Windows 3.1.
You can check it out here and learn more about what it was like to build it: https://www.dialup.net/wingpt/"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai","assistant"],"value":"Show HN: WinGPT \u2013 AI assistant for Windows 3.1"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://www.dialup.net/wingpt/"}},"_tags":["story","author_dialupdotnet","story_36472854","show_hn"],"author":"dialupdotnet","children":[36473686,36473893,36473947,36474009,36474031,36474135,36474212,36474266,36475109,36475440,36475498,36475638,36476080,36476185,36476435,36476582,36476739,36476921,36477355,36477478,36477516,36477521,36477740,36477741,36478281,36492801,36505969,36510796,36517180],"created_at":"2023-06-25T21:38:36Z","created_at_i":1687729116,"num_comments":90,"objectID":"36472854","points":382,"story_id":36472854,"story_text":"Hey HN, I didn't want my Gateway 4DX2-66 from 1993 to be left out of the AI revolution, so I built an AI Assistant for Windows 3.1, based on the OpenAI API.
The most interesting parts of building this were:
- Getting TLS 1.2 and 1.3 working on Windows 3.1 so that WinGPT could directly connect to OpenAI's server without relying on a modern machine for TLS termination.
- Learning about the memory segmentation architecture on 16-bit Windows, including the difference between far and near pointers.
- Building the UI in plain C code with the Windows API and limited selection of controls in Windows 3.1.
You can check it out here and learn more about what it was like to build it: https://www.dialup.net/wingpt/","title":"Show HN: WinGPT \u2013 AI assistant for Windows 3.1","updated_at":"2025-04-15T02:54:25Z","url":"https://www.dialup.net/wingpt/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"SaarthRanka"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"Hi HN! I'm Saarth. I'm in 9th grade, and I built Atticus.
Atticus is a fast and cheap Mac assistant that can do tasks for you. You hold down Control-Option-Space, speak or type your task, be it complex or a simple browser search, and Atticus will get on it.
The interesting part is that simple tasks, like opening apps, playing a Spotify playlist, or web queries, don't touch AI, and instead are processed locally and typically finish in under a second. For example, prompting Atticus to open apps takes around 0.08 seconds, sometimes even finishing before you finish your sentence, sort of like Jev.
Any tasks that are more complicated will be sent to Claude via claude -p, meaning you never need an API key and Atticus will run on your existing Claude Pro or Max plan. By using this approach, costs are also cut down immensely, with, on average, tasks costing a fraction of a cent.
To achieve Atticus' speed, it remembers shortcuts from each task it completes. When it does a task right for the first time, it saves its steps and replays them the next time without AI. For example, a workflow of creating a new meeting took about 30 seconds the first time around. Atticus then logs the exact steps to memory and takes 5-8 seconds on the next pass. To add to this, Atticus also learns about you as you use it more and from what's on your computer, including your name, your apps, and things you want it to remember, personalizing its answers to you.
To combat the safety concerns over giving an AI unrestricted access to your device, Atticus has toggleable restrictions, so before anything is deleted, sent, bought, or written to a file, the user is prompted before that action.
Some features I'm currently working on include a meeting copilot that capitalizes on Atticus' speed, allowing Atticus to listen in on meetings and answer questions in real time as they come in, personalized to the context of the meeting and what it knows about you.
To run Atticus, macOS 26 or higher is required for its voice transcription features. The voice transcription is also done locally by macOS.
Atticus is free, open source, and also very fresh, and I'm completely aware of its many rough edges, which is why I'd love to hear from everyone about what breaks and what you guys would like to see Atticus do!"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["assistant","that"],"value":"Show HN: Atticus \u2013 Hold-to-talk voice assistant for macOS that runs on Claude -p"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/SaarthurR/atticus-mac"}},"_tags":["story","author_SaarthRanka","story_50049588","show_hn"],"author":"SaarthRanka","children":[50049589,50052190,50052949,50053168,50053903,50054667,50055941,50058964,50064434,50064732,50065467,50065503,50065511,50066597,50067423,50067562,50068414,50071941,50080444,50080477],"created_at":"2026-10-12T00:15:12Z","created_at_i":1791764112,"num_comments":30,"objectID":"50049588","points":115,"story_id":50049588,"story_text":"Hi HN! I'm Saarth. I'm in 9th grade, and I built Atticus.
Atticus is a fast and cheap Mac assistant that can do tasks for you. You hold down Control-Option-Space, speak or type your task, be it complex or a simple browser search, and Atticus will get on it.
The interesting part is that simple tasks, like opening apps, playing a Spotify playlist, or web queries, don't touch AI, and instead are processed locally and typically finish in under a second. For example, prompting Atticus to open apps takes around 0.08 seconds, sometimes even finishing before you finish your sentence, sort of like Jev.
Any tasks that are more complicated will be sent to Claude via claude -p, meaning you never need an API key and Atticus will run on your existing Claude Pro or Max plan. By using this approach, costs are also cut down immensely, with, on average, tasks costing a fraction of a cent.
To achieve Atticus' speed, it remembers shortcuts from each task it completes. When it does a task right for the first time, it saves its steps and replays them the next time without AI. For example, a workflow of creating a new meeting took about 30 seconds the first time around. Atticus then logs the exact steps to memory and takes 5-8 seconds on the next pass. To add to this, Atticus also learns about you as you use it more and from what's on your computer, including your name, your apps, and things you want it to remember, personalizing its answers to you.
To combat the safety concerns over giving an AI unrestricted access to your device, Atticus has toggleable restrictions, so before anything is deleted, sent, bought, or written to a file, the user is prompted before that action.
Some features I'm currently working on include a meeting copilot that capitalizes on Atticus' speed, allowing Atticus to listen in on meetings and answer questions in real time as they come in, personalized to the context of the meeting and what it knows about you.
To run Atticus, macOS 26 or higher is required for its voice transcription features. The voice transcription is also done locally by macOS.
Atticus is free, open source, and also very fresh, and I'm completely aware of its many rough edges, which is why I'd love to hear from everyone about what breaks and what you guys would like to see Atticus do!","title":"Show HN: Atticus \u2013 Hold-to-talk voice assistant for macOS that runs on Claude -p","updated_at":"2026-10-16T04:29:48Z","url":"https://github.com/SaarthurR/atticus-mac"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mrayushsoni"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"I\u2019ve been thinking about the future of design workflows and wondering if anyone here is working on an AI design agent, not just another image generator, but something that actually understands design intent and executes creative direction end-to-end.
Something that could:\n- Take prompts like \u201ccreate a campaign for a new streetwear drop\u201d and generate cohesive brand visuals (logo, apparel mockups, ad creatives, etc.)\n- Learn a brand\u2019s visual identity over time\n- Suggest new content ideas or remix existing assets\n- Work more like a creative assistant than a \u201cprompt in, image out\u201d tool
Basically, a creative automation platform that behaves like a designer who knows your brand, not just a diffusion model with style filters.
Is anyone building something like this, or thinking in that direction? Would love to see examples, prototypes, or even open-source experiments.
Also curious: what would you want an AI design agent to handle for you? Mockups? Social posts? Full brand systems?"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["an","ai"],"value":"Ask HN: Is someone building an AI design agent?"}},"_tags":["story","author_mrayushsoni","story_45800619","ask_hn"],"author":"mrayushsoni","children":[45800725],"created_at":"2025-11-03T16:12:08Z","created_at_i":1762186328,"num_comments":1,"objectID":"45800619","points":1,"story_id":45800619,"story_text":"I\u2019ve been thinking about the future of design workflows and wondering if anyone here is working on an AI design agent, not just another image generator, but something that actually understands design intent and executes creative direction end-to-end.
Something that could:\n- Take prompts like \u201ccreate a campaign for a new streetwear drop\u201d and generate cohesive brand visuals (logo, apparel mockups, ad creatives, etc.)\n- Learn a brand\u2019s visual identity over time\n- Suggest new content ideas or remix existing assets\n- Work more like a creative assistant than a \u201cprompt in, image out\u201d tool
Basically, a creative automation platform that behaves like a designer who knows your brand, not just a diffusion model with style filters.
Is anyone building something like this, or thinking in that direction? Would love to see examples, prototypes, or even open-source experiments.
Also curious: what would you want an AI design agent to handle for you? Mockups? Social posts? Full brand systems?","title":"Ask HN: Is someone building an AI design agent?","updated_at":"2026-03-05T22:55:40Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"DonHopkins"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["want","an","ai","assistant","that","learns"],"value":"There's a lot of existing historic literature about this stuff that not a lot of people\nseem to be aware of. I've been trying to put those ideas to practical use, and document\nwhere the ideas came from.
I-Beam is cursor-mirror's agent, and it's constitutionally programmed to be the anti-Clippy:
https://github.com/SimHacker/moollm/tree/main/skills/cursor-...
Its design and constitution is based on decades of research, publications, and discussion\nin the HCI and AI community by people like Pattie Maes, Ben Shneiderman, Ted Selker, Byron\nReeves, Cliff Nass, B. J. Fogg, Allen Cypher, Henry Lieberman, Brad Myers, Jaron Lanier,\nSeymour Papert, Marvin Minsky, Douglas Engelbart, Will Wright, Scott McCloud, and others:
https://github.com/SimHacker/moollm/blob/main/skills/cursor-...
>I-Beam is the anti-Clippy, and the reason it can say so is that Clippy is the most cited failure in interface history and almost nobody citing it knows what the research said. Popular contempt for a paperclip is not a design principle. The record is. Ten articles below, each one a finding somebody published, argued or measured, and the operational rule it produces. Anything I-Beam does that cannot be traced to an article here is a preference, not a constraint, and should be labelled as one.
>The 1997 debate ended in agreement. That is the first thing to know, because the field kept the framing and dropped the resolution -- roughly five hundred papers cite "Shneiderman versus Maes" as the canonical opposition of HCI, and the transcript is two researchers narrowing their differences in public and enjoying it. I-Beam does not take a side in a debate whose participants stopped taking sides. It is built to satisfy both sets of constraints at once, which is possible, and was possible in 1997.
The full reading on the debate, which separates the two stagings and documents the convergence:
https://github.com/SimHacker/WillWrightShowForFood/blob/main...
An interface to agency, not agents instead of an interface:
https://github.com/SimHacker/moollm/blob/main/designs/INTERF...
>The 1997 argument between Ben Shneiderman and Pattie Maes at IUI was never settled, it was shipped in one direction. Maes's interface agents won the product war: the assistant, the recommender, the chat window that stands between you and the thing you are working on. Shneiderman's objection was not that software should be dumb. It was that automation must arrive as comprehensible, predictable, and controllable machinery, with the object of interest continuously visible and every action rapid, incremental, and reversible.
>That objection describes a filesystem in a git repository, and nobody involved planned it that way.
>"An interface to agency" is Don's formulation of Shneiderman's position, not a phrase of Shneiderman's. His own vocabulary is direct manipulation, universal usability, supertools, and human-centered AI. The formulation is a good one because it names what the alternative gets wrong: agency is the thing you want, and an agent is only one way to package it.
Here are some sources, and the articles I linked to above explain their history. This debate\nabout agents and these papers are pretty well known in the HCI field and academia, but they\ndon't tend to teach them at the AI and Web Dev boot camps that are producing most of the\npeople who keep repeating the same mistakes.
Clifford Nass was the Stanford professor who performed the brilliant research that Microsoft\ntook and totally fucked up and misinterpreted with Microsoft Bob and Clippy, giving agents a\nbad name, and making Clippy the most infamous and obnoxious agent in the history of the known\nuniverse:
https://en.wikipedia.org/wiki/Clifford_Nass
His student B. J. Fogg published "Silicon sycophants: the effects of computers that flatter,"\nwhich found that praise unconnected to anything the subject did works as well as sincere\npraise, and worked on subjects who knew it was noncontingent. Fogg and Nass, IJHCS 46(5),\n1997, 551-561:
https://doi.org/10.1006/ijhc.1996.0104
The replications, the performance cost, and the dose-response curve:
https://github.com/SimHacker/moollm/blob/main/skills/no-ai-s...
Shneiderman and Maes, "Direct Manipulation vs. Interface Agents," interactions 4(6),\nNov/Dec 1997, 42-61:
https://doi.org/10.1145/267505.267514
Selker, "New paradigms for using computers," CACM 39(8), August 1996, 60-69. COACH, the\nfootball coach metaphor, and the five-times result:
https://doi.org/10.1145/232014.232030
Selker, "COACH: A Teaching Agent that Learns," CACM 37(7), July 1994, 92-99:
https://doi.org/10.1145/176789.176799
Reeves and Nass, The Media Equation, 1996:
https://en.wikipedia.org/wiki/The_Media_Equation
Nass, "Computers as Social Actors," at Ted Selker's NPUC workshop at IBM Almaden, 1996. IBM\ntranscribed the whole talk and the Wayback Machine still has it, including the part where Phil\nAgre tells Nass his presentation is "ethically troubling all the way down" and asks him what he\nthinks about embedding obedience research in user interfaces. Nass answers that discovery has no\nethical component, use does, and that's for the individual. Then Selker cuts in: "Except, except\nwhen you are in your consulting role." Nass and Reeves had consulted for Microsoft on the social\ninterface, and Bob shipped the year before:
https://web.archive.org/web/19980210054622/http://www.almade...
Alan Cooper on the tragic misunderstanding, in his own voice, which I quoted before in the 2022 Hacker News discussion on The Twisted Life of Clippy:
https://news.ycombinator.com/item?id=32820734
https://archive.org/details/g4tv.com-video4080
>Alan Cooper (the "Father of Visual Basic") said: "Clippy was based on a really tragic misunderstanding of a truly profound bit of scientific research. At Stanford University, Clifford Nass and Byron Reeves, two brilliant scientists, had done some pioneering work proving conclusively that human beings react to computers with the same set of emotional reactions that they use to react to other human beings. [...] The work of Nass and Reeves proved that when people talk to computers, when they hit the keyboard and move the mouse, the part of their brain that's being activated is the part that has that emotional reaction to people dealing with people. Here's where the great mistake was made. That's really good research up to that point. But then the great mistake was made, which was: well if people react to computers as though they're people, we have to put the faces of people on computers. Which in my opinion is exactly the incorrect reaction. If people are going to react to computers as though they're humans, the one thing you don't have to do is anthropomorphize them, because they're already using that part of the brain. Clippy was a program based on the research that Nass and Reeves did, and it was a tragic misinterpretation of their work."
Social science research influences computer product design:
https://web.archive.org/web/20180313075429/https://web.stanf...
Lanier, "Early Computing's Long, Strange Trip," American Scientist, July-August 2005, with\nthe Engelbart and Minsky exchange first-hand. American Scientist broke the link, so this is\nthe Wayback copy:
https://web.archive.org/web/20150626081918/http://www.americ...
>The book also captures an important early conflict between two cultures of computing that seemed compatible on the surface but actually had opposing aims. On the one side was the human-centered design work of Engelbart, based initially at the Stanford Research Institute, and on the other was artificial intelligence culture, centered on the Stanford AI lab. Engelbart once told me a story that illustrates the conflict succinctly. He met Marvin Minsky\u2014one of the founders of the field of AI\u2014and Minsky told him how the AI lab would create intelligent machines. Engelbart replied, "You're going to do all that for the machines? What are you going to do for the people?" This conflict between machine- and human-centered design continues to this day.
Cypher, "EAGER: Programming Repetitive Tasks by Example," CHI '91:
https://doi.org/10.1145/108844.108850
Cypher (ed.), Watch What I Do: Programming by Demonstration, MIT Press 1993, full text:
Papert, Mindstorms, 1980:
https://archive.org/details/mindstormschildr00pape
Wright, Dollhouse preview lecture, April 1996, transcript:
https://github.com/SimHacker/moollm/blob/main/designs/sims/s..."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"A warning about 'model welfare'"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["ai"],"value":"https://mustafa-suleyman.ai/a-warning-about-model-welfare"}},"_tags":["comment","author_DonHopkins","story_49727580"],"author":"DonHopkins","comment_text":"There's a lot of existing historic literature about this stuff that not a lot of people\nseem to be aware of. I've been trying to put those ideas to practical use, and document\nwhere the ideas came from.
I-Beam is cursor-mirror's agent, and it's constitutionally programmed to be the anti-Clippy:
https://github.com/SimHacker/moollm/tree/main/skills/cursor-...
Its design and constitution is based on decades of research, publications, and discussion\nin the HCI and AI community by people like Pattie Maes, Ben Shneiderman, Ted Selker, Byron\nReeves, Cliff Nass, B. J. Fogg, Allen Cypher, Henry Lieberman, Brad Myers, Jaron Lanier,\nSeymour Papert, Marvin Minsky, Douglas Engelbart, Will Wright, Scott McCloud, and others:
https://github.com/SimHacker/moollm/blob/main/skills/cursor-...
>I-Beam is the anti-Clippy, and the reason it can say so is that Clippy is the most cited failure in interface history and almost nobody citing it knows what the research said. Popular contempt for a paperclip is not a design principle. The record is. Ten articles below, each one a finding somebody published, argued or measured, and the operational rule it produces. Anything I-Beam does that cannot be traced to an article here is a preference, not a constraint, and should be labelled as one.
>The 1997 debate ended in agreement. That is the first thing to know, because the field kept the framing and dropped the resolution -- roughly five hundred papers cite "Shneiderman versus Maes" as the canonical opposition of HCI, and the transcript is two researchers narrowing their differences in public and enjoying it. I-Beam does not take a side in a debate whose participants stopped taking sides. It is built to satisfy both sets of constraints at once, which is possible, and was possible in 1997.
The full reading on the debate, which separates the two stagings and documents the convergence:
https://github.com/SimHacker/WillWrightShowForFood/blob/main...
An interface to agency, not agents instead of an interface:
https://github.com/SimHacker/moollm/blob/main/designs/INTERF...
>The 1997 argument between Ben Shneiderman and Pattie Maes at IUI was never settled, it was shipped in one direction. Maes's interface agents won the product war: the assistant, the recommender, the chat window that stands between you and the thing you are working on. Shneiderman's objection was not that software should be dumb. It was that automation must arrive as comprehensible, predictable, and controllable machinery, with the object of interest continuously visible and every action rapid, incremental, and reversible.
>That objection describes a filesystem in a git repository, and nobody involved planned it that way.
>"An interface to agency" is Don's formulation of Shneiderman's position, not a phrase of Shneiderman's. His own vocabulary is direct manipulation, universal usability, supertools, and human-centered AI. The formulation is a good one because it names what the alternative gets wrong: agency is the thing you want, and an agent is only one way to package it.
Here are some sources, and the articles I linked to above explain their history. This debate\nabout agents and these papers are pretty well known in the HCI field and academia, but they\ndon't tend to teach them at the AI and Web Dev boot camps that are producing most of the\npeople who keep repeating the same mistakes.
Clifford Nass was the Stanford professor who performed the brilliant research that Microsoft\ntook and totally fucked up and misinterpreted with Microsoft Bob and Clippy, giving agents a\nbad name, and making Clippy the most infamous and obnoxious agent in the history of the known\nuniverse:
https://en.wikipedia.org/wiki/Clifford_Nass
His student B. J. Fogg published "Silicon sycophants: the effects of computers that flatter,"\nwhich found that praise unconnected to anything the subject did works as well as sincere\npraise, and worked on subjects who knew it was noncontingent. Fogg and Nass, IJHCS 46(5),\n1997, 551-561:
https://doi.org/10.1006/ijhc.1996.0104
The replications, the performance cost, and the dose-response curve:
https://github.com/SimHacker/moollm/blob/main/skills/no-ai-s...
Shneiderman and Maes, "Direct Manipulation vs. Interface Agents," interactions 4(6),\nNov/Dec 1997, 42-61:
https://doi.org/10.1145/267505.267514
Selker, "New paradigms for using computers," CACM 39(8), August 1996, 60-69. COACH, the\nfootball coach metaphor, and the five-times result:
https://doi.org/10.1145/232014.232030
Selker, "COACH: A Teaching Agent that Learns," CACM 37(7), July 1994, 92-99:
https://doi.org/10.1145/176789.176799
Reeves and Nass, The Media Equation, 1996:
https://en.wikipedia.org/wiki/The_Media_Equation
Nass, "Computers as Social Actors," at Ted Selker's NPUC workshop at IBM Almaden, 1996. IBM\ntranscribed the whole talk and the Wayback Machine still has it, including the part where Phil\nAgre tells Nass his presentation is "ethically troubling all the way down" and asks him what he\nthinks about embedding obedience research in user interfaces. Nass answers that discovery has no\nethical component, use does, and that's for the individual. Then Selker cuts in: "Except, except\nwhen you are in your consulting role." Nass and Reeves had consulted for Microsoft on the social\ninterface, and Bob shipped the year before:
https://web.archive.org/web/19980210054622/http://www.almade...
Alan Cooper on the tragic misunderstanding, in his own voice, which I quoted before in the 2022 Hacker News discussion on The Twisted Life of Clippy:
https://news.ycombinator.com/item?id=32820734
https://archive.org/details/g4tv.com-video4080
>Alan Cooper (the "Father of Visual Basic") said: "Clippy was based on a really tragic misunderstanding of a truly profound bit of scientific research. At Stanford University, Clifford Nass and Byron Reeves, two brilliant scientists, had done some pioneering work proving conclusively that human beings react to computers with the same set of emotional reactions that they use to react to other human beings. [...] The work of Nass and Reeves proved that when people talk to computers, when they hit the keyboard and move the mouse, the part of their brain that's being activated is the part that has that emotional reaction to people dealing with people. Here's where the great mistake was made. That's really good research up to that point. But then the great mistake was made, which was: well if people react to computers as though they're people, we have to put the faces of people on computers. Which in my opinion is exactly the incorrect reaction. If people are going to react to computers as though they're humans, the one thing you don't have to do is anthropomorphize them, because they're already using that part of the brain. Clippy was a program based on the research that Nass and Reeves did, and it was a tragic misinterpretation of their work."
Social science research influences computer product design:
https://web.archive.org/web/20180313075429/https://web.stanf...
Lanier, "Early Computing's Long, Strange Trip," American Scientist, July-August 2005, with\nthe Engelbart and Minsky exchange first-hand. American Scientist broke the link, so this is\nthe Wayback copy:
https://web.archive.org/web/20150626081918/http://www.americ...
>The book also captures an important early conflict between two cultures of computing that seemed compatible on the surface but actually had opposing aims. On the one side was the human-centered design work of Engelbart, based initially at the Stanford Research Institute, and on the other was artificial intelligence culture, centered on the Stanford AI lab. Engelbart once told me a story that illustrates the conflict succinctly. He met Marvin Minsky\u2014one of the founders of the field of AI\u2014and Minsky told him how the AI lab would create intelligent machines. Engelbart replied, "You're going to do all that for the machines? What are you going to do for the people?" This conflict between machine- and human-centered design continues to this day.
Cypher, "EAGER: Programming Repetitive Tasks by Example," CHI '91:
https://doi.org/10.1145/108844.108850
Cypher (ed.), Watch What I Do: Programming by Demonstration, MIT Press 1993, full text:
Papert, Mindstorms, 1980:
https://archive.org/details/mindstormschildr00pape
Wright, Dollhouse preview lecture, April 1996, transcript:
https://github.com/SimHacker/moollm/blob/main/designs/sims/s...","created_at":"2026-09-16T16:02:14Z","created_at_i":1789574534,"objectID":"49729076","parent_id":49727580,"story_id":49727580,"story_title":"A warning about 'model welfare'","story_url":"https://mustafa-suleyman.ai/a-warning-about-model-welfare","updated_at":"2026-09-16T17:50:29Z"}],"hitsPerPage":20,"nbHits":344,"nbPages":18,"page":0,"params":"query=want+an+AI+assistant+that+learns&advancedSyntax=true&analyticsTags=backend","processingTimeMS":45,"processingTimingsMS":{"_request":{"roundTrip":15},"afterFetch":{"format":{"highlighting":3,"total":3},"merge":{"mergeLoop":{"prepareNextHit":3,"total":3},"total":3},"total":3},"fetch":{"query":10,"scanning":30,"total":41},"total":45},"query":"want an AI assistant that learns","serverTimeMS":49}