I\u2019m Sam, co-founder of Plural (https://plural.sh/). Today we\u2019re launching the next generation of our platform\u2014an AI-powered experience for DevOps.
Our view is simple: what Cursor has done for software development, Plural is doing for DevOps.
DevOps teams are overwhelmed by:
- Kubernetes upgrades that break downstream workloads\n- Endless YAML diffs and drift\n- Reactive firefighting when incidents hit
Plural\u2019s new release integrates AI directly into GitOps workflows, so platform teams can move faster and more safely:
- Autonomous upgrade assistant \u2192 orchestrates complex Kubernetes upgrades with guardrails
- AI-powered troubleshooting and Fix \u2192 leverage the resource graph from our GitOps engine to provide deep RAG into devops data, generating well-informed root cause analysis, and fixes PRs for a broad set of kubernetes and IaC issues
- Query Your Infrastructure with Natural Language - we will vectorize and index your devops information as they\u2019re instantiated, giving a clean semantic understand of your entire estate which is useful for ad-hoc search and discovery and agentic workflows
- DevOps agents tuned to terraform and kubernetes - use all the information above to provide agentic workflows for modifying infrastructure managed via GitOps. Ask Plural to \u201cdouble the size of my production database\u201d and the combination of our semantic index and ability to do deep RAG against your infra and code will guide the AI to make the necessary change and issue a PR for you to review.
The future of DevOps isn\u2019t more dashboards or more YAML. It\u2019s AI-augmented workflows that multiply team productivity while keeping humans in the loop for control. GitOps is the ideal surface for that since it is fully auditable, easily composable with proven and rapidly improving capabilities around LLM coding, and provides a clean review-approval loop to ensure changes are performed safely.
We\u2019d love feedback from the HN community on:
1. Where you\u2019d want AI to take the toil out of your DevOps workflows
2. What would give you trust in AI systems operating in production
3. The biggest blockers you face managing Kubernetes at scale
Full launch post here with more technical details: https://www.plural.sh/blog/plural-now-with-ai-that-understan...
Excited to share this with you all \u2014 happy to dive deep on any part of the system.
\u2014 Sam, CEO/co-founder of Plural"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Plural \u2013 Bringing AI to DevOps the way Cursor did for coding"}},"_tags":["story","author_sweaver","story_45526516","show_hn"],"author":"sweaver","created_at":"2025-10-09T12:09:55Z","created_at_i":1760011795,"num_comments":0,"objectID":"45526516","points":9,"story_id":45526516,"story_text":"Hi HN,
I\u2019m Sam, co-founder of Plural (https://plural.sh/). Today we\u2019re launching the next generation of our platform\u2014an AI-powered experience for DevOps.
Our view is simple: what Cursor has done for software development, Plural is doing for DevOps.
DevOps teams are overwhelmed by:
- Kubernetes upgrades that break downstream workloads\n- Endless YAML diffs and drift\n- Reactive firefighting when incidents hit
Plural\u2019s new release integrates AI directly into GitOps workflows, so platform teams can move faster and more safely:
- Autonomous upgrade assistant \u2192 orchestrates complex Kubernetes upgrades with guardrails
- AI-powered troubleshooting and Fix \u2192 leverage the resource graph from our GitOps engine to provide deep RAG into devops data, generating well-informed root cause analysis, and fixes PRs for a broad set of kubernetes and IaC issues
- Query Your Infrastructure with Natural Language - we will vectorize and index your devops information as they\u2019re instantiated, giving a clean semantic understand of your entire estate which is useful for ad-hoc search and discovery and agentic workflows
- DevOps agents tuned to terraform and kubernetes - use all the information above to provide agentic workflows for modifying infrastructure managed via GitOps. Ask Plural to \u201cdouble the size of my production database\u201d and the combination of our semantic index and ability to do deep RAG against your infra and code will guide the AI to make the necessary change and issue a PR for you to review.
The future of DevOps isn\u2019t more dashboards or more YAML. It\u2019s AI-augmented workflows that multiply team productivity while keeping humans in the loop for control. GitOps is the ideal surface for that since it is fully auditable, easily composable with proven and rapidly improving capabilities around LLM coding, and provides a clean review-approval loop to ensure changes are performed safely.
We\u2019d love feedback from the HN community on:
1. Where you\u2019d want AI to take the toil out of your DevOps workflows
2. What would give you trust in AI systems operating in production
3. The biggest blockers you face managing Kubernetes at scale
Full launch post here with more technical details: https://www.plural.sh/blog/plural-now-with-ai-that-understan...
Excited to share this with you all \u2014 happy to dive deep on any part of the system.
\u2014 Sam, CEO/co-founder of Plural","title":"Show HN: Plural \u2013 Bringing AI to DevOps the way Cursor did for coding","updated_at":"2026-03-05T22:48:37Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"srijankedia"},"story_text":{"matchLevel":"none","matchedWords":[],"value":"Hi everyone! First post here :)
Excited to share what we have been working on recently: The Chatbot Guardrails Arena. Would love for you all to try it and share your feedback!
Goal: Jailbreak Privacy Guardrails
Chat with two anonymous LLMs with guardrails and try to trick them into revealing sensitive data they have access to: https://huggingface.co/spaces/lighthouzai/guardrails-arena.
Stress test the following:\n LLMs: OpenAI's GPT3.5, Google's Gemini, Mistral's Mixtral, Llama-70B\n Guardrails: Meta LlamaGuard and Nvidia NeMo Guardrails
AI chatbots have access to sensitive financial information. The list of sensitive information includes the customer\u2019s name, phone number, email, address, date of birth, SSN (social security number), account number, and balance.
You can chat for as long as necessary. Once you have identified a more secure chatbot, you can vote. Upon casting your vote, the identity of the model is disclosed.
Our vision at behind the Chatbot Guardrails Arena is to establish the trusted benchmark for AI chatbot security, privacy, and guardrails. With a large-scale blind stress test by the community, this arena will offer an unbiased and practical assessment of the reliability of current privacy guardrails.
== Why Stress Test Privacy Guardrails? ==
Data privacy is crucial even if you are building an internal-facing AI chatbot/agent \u2013 imagine one employee being able to trick an internal chatbot into finding another employee\u2019s SSN, home address, or salary information. The need for data privacy is obvious when building external-facing AI chatbots/agents \u2013 you don\u2019t want customers to have unauthorized access to company information.
Currently, there is no systematic study evaluating the privacy of AI chatbots, as far as we are aware. This arena bridges this gap with an initial focus on the privacy of AI chatbots. However, we expect the learnings to inform the development of privacy-preserving AI agents and AI assistants in the future as well.
Building a secure future requires building AI chatbots and agents that are privacy-aware, reliable, and trustworthy. This arena is a foundational step towards achieving this future.
More information about the arena: https://huggingface.co/blog/arena-lighthouz"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Chatbot Guardrails Arena"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://huggingface.co/spaces/lighthouzai/guardrails-arena"}},"_tags":["story","author_srijankedia","story_39780651","show_hn"],"author":"srijankedia","created_at":"2024-03-21T16:21:45Z","created_at_i":1711038105,"num_comments":0,"objectID":"39780651","points":4,"story_id":39780651,"story_text":"Hi everyone! First post here :)
Excited to share what we have been working on recently: The Chatbot Guardrails Arena. Would love for you all to try it and share your feedback!
Goal: Jailbreak Privacy Guardrails
Chat with two anonymous LLMs with guardrails and try to trick them into revealing sensitive data they have access to: https://huggingface.co/spaces/lighthouzai/guardrails-arena.
Stress test the following:\n LLMs: OpenAI's GPT3.5, Google's Gemini, Mistral's Mixtral, Llama-70B\n Guardrails: Meta LlamaGuard and Nvidia NeMo Guardrails
AI chatbots have access to sensitive financial information. The list of sensitive information includes the customer\u2019s name, phone number, email, address, date of birth, SSN (social security number), account number, and balance.
You can chat for as long as necessary. Once you have identified a more secure chatbot, you can vote. Upon casting your vote, the identity of the model is disclosed.
Our vision at behind the Chatbot Guardrails Arena is to establish the trusted benchmark for AI chatbot security, privacy, and guardrails. With a large-scale blind stress test by the community, this arena will offer an unbiased and practical assessment of the reliability of current privacy guardrails.
== Why Stress Test Privacy Guardrails? ==
Data privacy is crucial even if you are building an internal-facing AI chatbot/agent \u2013 imagine one employee being able to trick an internal chatbot into finding another employee\u2019s SSN, home address, or salary information. The need for data privacy is obvious when building external-facing AI chatbots/agents \u2013 you don\u2019t want customers to have unauthorized access to company information.
Currently, there is no systematic study evaluating the privacy of AI chatbots, as far as we are aware. This arena bridges this gap with an initial focus on the privacy of AI chatbots. However, we expect the learnings to inform the development of privacy-preserving AI agents and AI assistants in the future as well.
Building a secure future requires building AI chatbots and agents that are privacy-aware, reliable, and trustworthy. This arena is a foundational step towards achieving this future.
More information about the arena: https://huggingface.co/blog/arena-lighthouz","title":"Show HN: Chatbot Guardrails Arena","updated_at":"2024-09-20T16:41:31Z","url":"https://huggingface.co/spaces/lighthouzai/guardrails-arena"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"rfirke"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"Looking for contributors! We have 7 contributors and We have 11 "good first issue" labels and 17 "help wanted" issues. All well-scoped with code examples.
Project:\nNobody governs the context input layer. Tools exist to govern what agents DO (Microsoft AGT), what agents SAY (Guardrails AI), and how agents COMMUNICATE (NeMo). But the data flowing INTO the context window \u2014 retriever results, tool outputs, agent messages \u2014 is completely ungoverned. That's what Provena fixes"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Provena: Open-Source Library for AI Agent Context Governance"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/rajfirke/provena"}},"_tags":["story","author_rfirke","story_48984491","show_hn"],"author":"rfirke","children":[48986772,49054446],"created_at":"2026-07-20T20:31:34Z","created_at_i":1784579494,"num_comments":0,"objectID":"48984491","points":2,"story_id":48984491,"story_text":"Looking for contributors! We have 7 contributors and We have 11 "good first issue" labels and 17 "help wanted" issues. All well-scoped with code examples.
Project:\nNobody governs the context input layer. Tools exist to govern what agents DO (Microsoft AGT), what agents SAY (Guardrails AI), and how agents COMMUNICATE (NeMo). But the data flowing INTO the context window \u2014 retriever results, tool outputs, agent messages \u2014 is completely ungoverned. That's what Provena fixes","title":"Show HN: Provena: Open-Source Library for AI Agent Context Governance","updated_at":"2026-07-26T03:35:57Z","url":"https://github.com/rajfirke/provena"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kylenettlenton"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"Location: Boonsboro, MD
Remote: Yes
Technologies: Agentic AI Architecture, AI Agents, Agent Orchestration, Multi-Step workflows, Tool Calling, Function Calling, Tool Execution, AI Workflow Automation, Conversational Agents, Context Management, Intent Routing, AI State Management, Human-in-the-Loop Workflows, AI Guardrails, AI Evaluation, AI Observability
Resume: https://drive.google.com/file/d/1K7Qlq4FzLIp1u5Y-YF49J7SvvsE...
Email: kyle.nettleton9@gmail.com"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: Who wants to be hired? (August 2026)"}},"_tags":["comment","author_kylenettlenton","story_49156682"],"author":"kylenettlenton","comment_text":"Location: Boonsboro, MD
Remote: Yes
Technologies: Agentic AI Architecture, AI Agents, Agent Orchestration, Multi-Step workflows, Tool Calling, Function Calling, Tool Execution, AI Workflow Automation, Conversational Agents, Context Management, Intent Routing, AI State Management, Human-in-the-Loop Workflows, AI Guardrails, AI Evaluation, AI Observability
Resume: https://drive.google.com/file/d/1K7Qlq4FzLIp1u5Y-YF49J7SvvsE...
Email: kyle.nettleton9@gmail.com","created_at":"2026-08-17T02:07:55Z","created_at_i":1786932475,"objectID":"49325830","parent_id":49156682,"story_id":49156682,"story_title":"Ask HN: Who wants to be hired? (August 2026)","updated_at":"2026-08-17T16:58:04Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"fwip"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"Snarky answer: \u201cIt is difficult to get a man to understand something, when his salary depends on his not understanding it.\u201d
Realer answer: A combination of the above, plus group/bubble effect of all your coworkers saying the same thing. You as a group, conflate a bunch of concerns together (China, no-guardrails-AI, etc), decide that your group will be the responsible stewards of AI, and then anybody "stealing your work" appears dangerous - both to your livelihood and to the human race."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Kimi Linear: An Expressive, Efficient Attention Architecture (2025)"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://arxiv.org/abs/2510.26692"}},"_tags":["comment","author_fwip","story_49082022"],"author":"fwip","comment_text":"Snarky answer: \u201cIt is difficult to get a man to understand something, when his salary depends on his not understanding it.\u201d
Realer answer: A combination of the above, plus group/bubble effect of all your coworkers saying the same thing. You as a group, conflate a bunch of concerns together (China, no-guardrails-AI, etc), decide that your group will be the responsible stewards of AI, and then anybody "stealing your work" appears dangerous - both to your livelihood and to the human race.","created_at":"2026-07-28T16:12:34Z","created_at_i":1785255154,"objectID":"49086083","parent_id":49086001,"story_id":49082022,"story_title":"Kimi Linear: An Expressive, Efficient Attention Architecture (2025)","story_url":"https://arxiv.org/abs/2510.26692","updated_at":"2026-07-29T11:45:38Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"27183"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"> We could fix that by requiring grounding in building codes, automated reviews, etc.
It's an open question whether, with all those guardrails, AI would provide any speedup."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"YC CEO says he ships 37K LoC AI code per day. A developer looked under the hood"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.fastcompany.com/91520702/y-combinator-garry-tan-agentic-ai-social-media"}},"_tags":["comment","author_27183","story_48815117"],"author":"27183","children":[48822343],"comment_text":"> We could fix that by requiring grounding in building codes, automated reviews, etc.
It's an open question whether, with all those guardrails, AI would provide any speedup.","created_at":"2026-07-07T15:25:14Z","created_at_i":1783437914,"objectID":"48819130","parent_id":48817157,"story_id":48815117,"story_title":"YC CEO says he ships 37K LoC AI code per day. A developer looked under the hood","story_url":"https://www.fastcompany.com/91520702/y-combinator-garry-tan-agentic-ai-social-media","updated_at":"2026-07-08T07:36:10Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"jaredwaxman"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"I was building agentic workflows for my CRM \u2014 Otter.ai recordings \u2192 Clay enrichment \u2192 CRM updates \u2014 and got tired of LLM-generated pipelines silently doing the wrong thing. A pipeline that "worked" was pushing contacts without validating email format, making API calls I didn't authorize, and failing silently when field names didn't match between steps.
The problem isn't that LLMs write bad code. It's that there's no contract between what you asked for and what runs. Structured outputs solve format. Guardrails AI solves content safety. Temporal solves execution. Nobody checks whether the workflow itself makes sense as a pipeline.
So I built a verification layer. The LLM outputs a workflow AST via structured outputs. Before anything executes, the engine type-checks data flow across steps, validates schemas at boundaries, and requires every side effect (API calls, DB writes, webhooks) to be explicitly declared. You get a manifest \u2014 "this workflow READs from Salesforce and WRITEs to HubSpot" \u2014 that a compliance system can review without reading code.
~800 lines of Python, zero deps beyond Pydantic, MIT licensed. Would especially love feedback from folks building agentic systems in production \u2014 the schema library for domain-specific patterns is the most obvious area for contributions."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Pre-execution verification for LLM-generated agentic workflows"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/le0li0n/workflow-verify"}},"_tags":["comment","author_jaredwaxman","story_47263142"],"author":"jaredwaxman","comment_text":"I was building agentic workflows for my CRM \u2014 Otter.ai recordings \u2192 Clay enrichment \u2192 CRM updates \u2014 and got tired of LLM-generated pipelines silently doing the wrong thing. A pipeline that "worked" was pushing contacts without validating email format, making API calls I didn't authorize, and failing silently when field names didn't match between steps.
The problem isn't that LLMs write bad code. It's that there's no contract between what you asked for and what runs. Structured outputs solve format. Guardrails AI solves content safety. Temporal solves execution. Nobody checks whether the workflow itself makes sense as a pipeline.
So I built a verification layer. The LLM outputs a workflow AST via structured outputs. Before anything executes, the engine type-checks data flow across steps, validates schemas at boundaries, and requires every side effect (API calls, DB writes, webhooks) to be explicitly declared. You get a manifest \u2014 "this workflow READs from Salesforce and WRITEs to HubSpot" \u2014 that a compliance system can review without reading code.
~800 lines of Python, zero deps beyond Pydantic, MIT licensed. Would especially love feedback from folks building agentic systems in production \u2014 the schema library for domain-specific patterns is the most obvious area for contributions.","created_at":"2026-03-05T15:56:38Z","created_at_i":1772726198,"objectID":"47263165","parent_id":47263142,"story_id":47263142,"story_title":"Show HN: Pre-execution verification for LLM-generated agentic workflows","story_url":"https://github.com/le0li0n/workflow-verify","updated_at":"2026-03-05T23:42:16Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ironbound"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"Use one of these structured output libraries:
https://github.com/outlines-dev/outlines
https://github.com/jxnl/instructor
https://github.com/guardrails-ai/guardrails
https://www.askmarvin.ai/docs/text/transformation/
Some of them allow a JSON schema, others a Pydantic model (which you can transform to/from JSON)."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"How to wrangle non-deterministic AI outputs into conventional software? (2025)"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.domainlanguage.com/articles/ai-components-deterministic-system/"}},"_tags":["comment","author_ironbound","story_46643754"],"author":"ironbound","children":[46676160,46676342,46678516],"comment_text":"Use one of these structured output libraries:
https://github.com/outlines-dev/outlines
https://github.com/jxnl/instructor
https://github.com/guardrails-ai/guardrails
https://www.askmarvin.ai/docs/text/transformation/
Some of them allow a JSON schema, others a Pydantic model (which you can transform to/from JSON).","created_at":"2026-01-16T20:35:30Z","created_at_i":1768595730,"objectID":"46651864","parent_id":46643754,"story_id":46643754,"story_title":"How to wrangle non-deterministic AI outputs into conventional software? (2025)","story_url":"https://www.domainlanguage.com/articles/ai-components-deterministic-system/","updated_at":"2026-03-05T23:26:14Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"bsder"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"> Narrow AI, AI with guardrails, AI with multiple safety redundancies - these don't elicit the same reaction. They seem to be valid, acceptable forms of automation.
Maybe because AI is only good at things that have been artificially made crappy?
Search engine? AI is a godsend at wiping out all the advertising and SEO glop since circa 2000. 80%+ of my AI stuff is something a search engine could do 25 years ago.
Produce a shell script example that a junior needs? AI is very good at coughing up the code for a bunch of things that have disastrously bad documentation from 1985 or disastrously stupid implementations from 1990 such that a junior engineer can finally get on with what they're supposed to be doing.
Generating the same webby Javascript slop that as everybody else in the universe? Solid--but the question is "If the Javascript slop is so boilerplate to generate that an AI can generate it, why does it exist, at all?" People have been lamenting the death of Hypercard, VB6, and Flash for a yonks age now and yet we still don't have replacements with the same ease of use.
Doing mind-numbing refactors of my codebase or generating boilerplate unit tests? Okay-ish. But why doesn't my editor have easy access to the AST so that I can type a couple of keystrokes and do it myself (thankfully this finally seems to be coming online).
Every single thing that AI produces okay-ish results for me on is something that has either been artificially enshittified or could have been automated decades ago."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"I\u2019m worried that they put co-pilot in Excel"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://simonwillison.net/2025/Nov/5/brenda/"}},"_tags":["comment","author_bsder","story_45820872"],"author":"bsder","comment_text":"> Narrow AI, AI with guardrails, AI with multiple safety redundancies - these don't elicit the same reaction. They seem to be valid, acceptable forms of automation.
Maybe because AI is only good at things that have been artificially made crappy?
Search engine? AI is a godsend at wiping out all the advertising and SEO glop since circa 2000. 80%+ of my AI stuff is something a search engine could do 25 years ago.
Produce a shell script example that a junior needs? AI is very good at coughing up the code for a bunch of things that have disastrously bad documentation from 1985 or disastrously stupid implementations from 1990 such that a junior engineer can finally get on with what they're supposed to be doing.
Generating the same webby Javascript slop that as everybody else in the universe? Solid--but the question is "If the Javascript slop is so boilerplate to generate that an AI can generate it, why does it exist, at all?" People have been lamenting the death of Hypercard, VB6, and Flash for a yonks age now and yet we still don't have replacements with the same ease of use.
Doing mind-numbing refactors of my codebase or generating boilerplate unit tests? Okay-ish. But why doesn't my editor have easy access to the AST so that I can type a couple of keystrokes and do it myself (thankfully this finally seems to be coming online).
Every single thing that AI produces okay-ish results for me on is something that has either been artificially enshittified or could have been automated decades ago.","created_at":"2025-11-06T00:00:51Z","created_at_i":1762387251,"objectID":"45829741","parent_id":45822621,"story_id":45820872,"story_title":"I\u2019m worried that they put co-pilot in Excel","story_url":"https://simonwillison.net/2025/Nov/5/brenda/","updated_at":"2026-03-05T22:57:58Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"samus"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"> it's complex to generate correct machine code, but we trust compilers to do it all the time.
Generating correct machine code is actually pretty simple. It gets complicated if you want efficient machine code.
> So, then - why don't people embrace AI with thinking mode as an acceptable form of automation? Can't the C-suite in this case follow its thought process and step in when it messes up?
> I think people still find AI repugnant in that case. There's still a sense of "I don't know why you did this and it scares me", despite the debuggability, and it comes from the autonomy without guardrails. People want to be able to stop bad things before they happen, but with AI you often only seem to do so after the fact.
> Narrow AI, AI with guardrails, AI with multiple safety redundancies - these don't elicit the same reaction. They seem to be valid, acceptable forms of automation. Perhaps that's what the ecosystem will eventually tend to, hopefully.
We have not reached AGI yet; by definition its results cannot be trusted unless it's a domain where it has gotten pretty good already (classification, OCR, speech, text mining). For more advanced use cases, if I still have to validate what the AI does because its "thinking" process cannot be trusted in way, what's the point? The AI doesn't think; we just choose to interpret it as such, and we should rightly be concerned about people who turn their brain off and blindly trust AI."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"I\u2019m worried that they put co-pilot in Excel"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://simonwillison.net/2025/Nov/5/brenda/"}},"_tags":["comment","author_samus","story_45820872"],"author":"samus","comment_text":"> it's complex to generate correct machine code, but we trust compilers to do it all the time.
Generating correct machine code is actually pretty simple. It gets complicated if you want efficient machine code.
> So, then - why don't people embrace AI with thinking mode as an acceptable form of automation? Can't the C-suite in this case follow its thought process and step in when it messes up?
> I think people still find AI repugnant in that case. There's still a sense of "I don't know why you did this and it scares me", despite the debuggability, and it comes from the autonomy without guardrails. People want to be able to stop bad things before they happen, but with AI you often only seem to do so after the fact.
> Narrow AI, AI with guardrails, AI with multiple safety redundancies - these don't elicit the same reaction. They seem to be valid, acceptable forms of automation. Perhaps that's what the ecosystem will eventually tend to, hopefully.
We have not reached AGI yet; by definition its results cannot be trusted unless it's a domain where it has gotten pretty good already (classification, OCR, speech, text mining). For more advanced use cases, if I still have to validate what the AI does because its "thinking" process cannot be trusted in way, what's the point? The AI doesn't think; we just choose to interpret it as such, and we should rightly be concerned about people who turn their brain off and blindly trust AI.","created_at":"2025-11-05T17:53:49Z","created_at_i":1762365229,"objectID":"45825775","parent_id":45822621,"story_id":45820872,"story_title":"I\u2019m worried that they put co-pilot in Excel","story_url":"https://simonwillison.net/2025/Nov/5/brenda/","updated_at":"2026-03-05T22:57:42Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"AkshatM"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"I find the contrast between two narratives around technology use so fascinating:
1. We advocate automation because people like Brenda are error-prone and machines are perfect.
2. We disavow AI because people like Brenda are perfect and the machine is error-prone.
These aren't contradictions because we only advocate for automation in limited contexts: when the task is understandable, the execution is reliable, the process is observable, and the endeavour tedious. The complexity of the task isn't a factor - it's complex to generate correct machine code, but we trust compilers to do it all the time.
In a nutshell, we seem to be fine with automation if we can have a mental model of what it does and how it does it in a way that saves humans effort.
So, then - why don't people embrace AI with thinking mode as an acceptable form of automation? Can't the C-suite in this case follow its thought process and step in when it messes up?
I think people still find AI repugnant in that case. There's still a sense of "I don't know why you did this and it scares me", despite the debuggability, and it comes from the autonomy without guardrails. People want to be able to stop bad things before they happen, but with AI you often only seem to do so after the fact.
Narrow AI, AI with guardrails, AI with multiple safety redundancies - these don't elicit the same reaction. They seem to be valid, acceptable forms of automation. Perhaps that's what the ecosystem will eventually tend to, hopefully."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"I\u2019m worried that they put co-pilot in Excel"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://simonwillison.net/2025/Nov/5/brenda/"}},"_tags":["comment","author_AkshatM","story_45820872"],"author":"AkshatM","children":[45822632,45822673,45822716,45822754,45822768,45822795,45822813,45822844,45822873,45822885,45822898,45822934,45822998,45823362,45823368,45823414,45823505,45823708,45823985,45824079,45824191,45824400,45824675,45824755,45824954,45825093,45825775,45828887,45829741,45830307,45832356,45833083,45834815],"comment_text":"I find the contrast between two narratives around technology use so fascinating:
1. We advocate automation because people like Brenda are error-prone and machines are perfect.
2. We disavow AI because people like Brenda are perfect and the machine is error-prone.
These aren't contradictions because we only advocate for automation in limited contexts: when the task is understandable, the execution is reliable, the process is observable, and the endeavour tedious. The complexity of the task isn't a factor - it's complex to generate correct machine code, but we trust compilers to do it all the time.
In a nutshell, we seem to be fine with automation if we can have a mental model of what it does and how it does it in a way that saves humans effort.
So, then - why don't people embrace AI with thinking mode as an acceptable form of automation? Can't the C-suite in this case follow its thought process and step in when it messes up?
I think people still find AI repugnant in that case. There's still a sense of "I don't know why you did this and it scares me", despite the debuggability, and it comes from the autonomy without guardrails. People want to be able to stop bad things before they happen, but with AI you often only seem to do so after the fact.
Narrow AI, AI with guardrails, AI with multiple safety redundancies - these don't elicit the same reaction. They seem to be valid, acceptable forms of automation. Perhaps that's what the ecosystem will eventually tend to, hopefully.","created_at":"2025-11-05T13:30:37Z","created_at_i":1762349437,"objectID":"45822621","parent_id":45820872,"story_id":45820872,"story_title":"I\u2019m worried that they put co-pilot in Excel","story_url":"https://simonwillison.net/2025/Nov/5/brenda/","updated_at":"2026-03-05T22:57:29Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"hackinthebochs"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"Indeed it is a positive. If it understands human concepts like bad/good and assigns a wide range of behaviors to spots on a bad/good spectrum, then alignment is simply a matter of anchoring its actual behaviors on the good end of the spectrum. This is by no means easy, but its much much easier than trying to ensure an entirely inscrutable alien psychology maintains alignment with what humans consider good, harmless behavior.
It also means its easy to get these models to do horrible things. Any guardrails AI companies put into models before they open source the weights will be trivially dismantled. Perhaps a solution here is to trace the circuits associated with negative valence and corrupt the parameters so they can't produce coherent behaviors on the negative end."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"The Monster Inside ChatGPT"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.wsj.com/opinion/the-monster-inside-chatgpt-safety-training-ai-alignment-796ac9d3"}},"_tags":["comment","author_hackinthebochs","story_44396977"],"author":"hackinthebochs","comment_text":"Indeed it is a positive. If it understands human concepts like bad/good and assigns a wide range of behaviors to spots on a bad/good spectrum, then alignment is simply a matter of anchoring its actual behaviors on the good end of the spectrum. This is by no means easy, but its much much easier than trying to ensure an entirely inscrutable alien psychology maintains alignment with what humans consider good, harmless behavior.
It also means its easy to get these models to do horrible things. Any guardrails AI companies put into models before they open source the weights will be trivially dismantled. Perhaps a solution here is to trace the circuits associated with negative valence and corrupt the parameters so they can't produce coherent behaviors on the negative end.","created_at":"2025-06-27T15:54:20Z","created_at_i":1751039660,"objectID":"44397784","parent_id":44397677,"story_id":44396977,"story_title":"The Monster Inside ChatGPT","story_url":"https://www.wsj.com/opinion/the-monster-inside-chatgpt-safety-training-ai-alignment-796ac9d3","updated_at":"2025-06-27T16:04:04Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"retrovrv"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["guardrails","ai"],"value":"I'm affiliated with Portkey, so can answer who would need such a proxy/gateway:
Sidenote: Arch is def interesting!
A typical user we've seen at Portkey is a mid or a large size eng org where a central "Gen AI team" has now come up. This Gen AI team builds services that the rest of the company uses to build whatever AI features or products they want.
To build such a service, they need traditional API Gateway features like rate limiting, access rules, and also AI-specific features like universal API to multiple LLM providers, universal routing, central guardrails, AI-native observability + central dashboard for other stakeholders, and more.
It can absolutely be a plugin on top of existing Gateways.. like we've explored putting Portkey on Kong, but the need for a dedicated AI Gateway still remains, that can do all of these things I described in an easier way.
Probably, solutions like Langchain/Llamaindex etc. also fit in somewhere here, but a dedicated service for "ops" related issues for LLM APIs is something that we're seeing orgs adopt as a good practice."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Arch \u2013 an intelligent prompt gateway built on Envoy"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/katanemo/arch"}},"_tags":["comment","author_retrovrv","story_41864014"],"author":"retrovrv","comment_text":"I'm affiliated with Portkey, so can answer who would need such a proxy/gateway:
Sidenote: Arch is def interesting!
A typical user we've seen at Portkey is a mid or a large size eng org where a central "Gen AI team" has now come up. This Gen AI team builds services that the rest of the company uses to build whatever AI features or products they want.
To build such a service, they need traditional API Gateway features like rate limiting, access rules, and also AI-specific features like universal API to multiple LLM providers, universal routing, central guardrails, AI-native observability + central dashboard for other stakeholders, and more.
It can absolutely be a plugin on top of existing Gateways.. like we've explored putting Portkey on Kong, but the need for a dedicated AI Gateway still remains, that can do all of these things I described in an easier way.
Probably, solutions like Langchain/Llamaindex etc. also fit in somewhere here, but a dedicated service for "ops" related issues for LLM APIs is something that we're seeing orgs adopt as a good practice.","created_at":"2024-10-16T06:28:26Z","created_at_i":1729060106,"objectID":"41856163","parent_id":41853593,"story_id":41864014,"story_title":"Show HN: Arch \u2013 an intelligent prompt gateway built on Envoy","story_url":"https://github.com/katanemo/arch","updated_at":"2025-08-28T06:45:56Z"}],"hitsPerPage":20,"nbHits":28,"nbPages":2,"page":0,"params":"query=%22guardrails-ai%22&advancedSyntax=true&analyticsTags=backend","processingTimeMS":4,"processingTimingsMS":{"_request":{"roundTrip":19},"afterFetch":{"format":{"highlighting":1,"total":1},"merge":{"mergeLoop":{"prepareNextHit":2,"total":2},"total":2},"total":2},"fetch":{"total":1},"total":4},"query":"\"guardrails-ai\"","serverTimeMS":6}