So, with the help of Amp Code CLI, I've built my own take on the compound engineering workflow. I tried to keep it agnostic to project stacks and as efficient as possible, so the context window could be used in the best way. I also wanted it to be extendable (for example, just drop your own subagents for review that are specific to your project). I also wanted to be easy to set up and update, so I made a simple CLI tool that keeps track of files in the `.agents` directory, updates when new versions are found in the repository, and displays a diff in the terminal before overwriting any customisations.
I feel this matches well with my personal preferences when working with AI agents, but I would love to have feedback from more people."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Praxis, my personal take on Compound Engineering with AI"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/DFilipeS/praxis"}},"_tags":["story","author_DFilipeS","story_47143410","show_hn"],"author":"DFilipeS","created_at":"2026-02-24T21:33:14Z","created_at_i":1771968794,"num_comments":0,"objectID":"47143410","points":5,"story_id":47143410,"story_text":"Hey HN! I really enjoy Every's approach to Compound Engineering (https://every.to/guides/compound-engineering), but their plugin is tightly tied to their project (Cora) and stack (Ruby/Rails). I also found the files too big, and they used more context window than what I would like for my personal use.
So, with the help of Amp Code CLI, I've built my own take on the compound engineering workflow. I tried to keep it agnostic to project stacks and as efficient as possible, so the context window could be used in the best way. I also wanted it to be extendable (for example, just drop your own subagents for review that are specific to your project). I also wanted to be easy to set up and update, so I made a simple CLI tool that keeps track of files in the `.agents` directory, updates when new versions are found in the repository, and displays a diff in the terminal before overwriting any customisations.
I feel this matches well with my personal preferences when working with AI agents, but I would love to have feedback from more people.","title":"Show HN: Praxis, my personal take on Compound Engineering with AI","updated_at":"2026-03-05T23:36:41Z","url":"https://github.com/DFilipeS/praxis"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"hmokiguess"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"There's lots of models benchmark out there, but how do you evaluate coding agents?
I've been seeing a lot of OpenCode fuzz on HN lately, because of Anthropic disabling their access to the private subscription endpoints, and I confess it made me feel like I could be missing out on something though I can't tell for sure.
There's also Amp Code who seems to be picking up traction, and, although more on the IDE side, I have tried Kiro through AWS Credits and it surprisingly outperforms Claude Code for me in some cases but didn't fully bait me into the switch.
Codex works as good as Claude Code for me but I like Claude's UX and Opus 4.5 better.
Are there any reliable Coding Agents benchmark out there? What is your take?"},"title":{"matchLevel":"none","matchedWords":[],"value":"How do you pick a Coding Agent HN?"}},"_tags":["story","author_hmokiguess","story_46634773","ask_hn"],"author":"hmokiguess","created_at":"2026-01-15T16:15:41Z","created_at_i":1768493741,"num_comments":0,"objectID":"46634773","points":4,"story_id":46634773,"story_text":"There's lots of models benchmark out there, but how do you evaluate coding agents?
I've been seeing a lot of OpenCode fuzz on HN lately, because of Anthropic disabling their access to the private subscription endpoints, and I confess it made me feel like I could be missing out on something though I can't tell for sure.
There's also Amp Code who seems to be picking up traction, and, although more on the IDE side, I have tried Kiro through AWS Credits and it surprisingly outperforms Claude Code for me in some cases but didn't fully bait me into the switch.
Codex works as good as Claude Code for me but I like Claude's UX and Opus 4.5 better.
Are there any reliable Coding Agents benchmark out there? What is your take?","title":"How do you pick a Coding Agent HN?","updated_at":"2026-03-05T23:25:05Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"thomasahle"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"this is an alternative to langfuse and mlflow for visualizing your agent traces. It has no sign up, no hosting and all data stays on your laptop.
I was working with benchmarking agents using terminal-bench, and I often found myself with a bunch of messages json that I needed to read. Combined with MCP calls and base64 images, these things can get quite hairy.
So I build a visualizer, inspired by ampcode, that neatly parses eveything and displays it in your browser for quick debugging. It's open source and serverless, so you can run it as you please.
I tested mostly with my own logs from ~/.claude/projects/, so it's possible you'll be able to break it. Let me know if you do!"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Trace.taxi \u2013 easy agent messages visualization"}},"_tags":["story","author_thomasahle","story_45921651","show_hn"],"author":"thomasahle","children":[45922016],"created_at":"2025-11-13T22:43:10Z","created_at_i":1763073790,"num_comments":1,"objectID":"45921651","points":3,"story_id":45921651,"story_text":"this is an alternative to langfuse and mlflow for visualizing your agent traces. It has no sign up, no hosting and all data stays on your laptop.
I was working with benchmarking agents using terminal-bench, and I often found myself with a bunch of messages json that I needed to read. Combined with MCP calls and base64 images, these things can get quite hairy.
So I build a visualizer, inspired by ampcode, that neatly parses eveything and displays it in your browser for quick debugging. It's open source and serverless, so you can run it as you please.
I tested mostly with my own logs from ~/.claude/projects/, so it's possible you'll be able to break it. Let me know if you do!","title":"Show HN: Trace.taxi \u2013 easy agent messages visualization","updated_at":"2026-03-05T22:58:53Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"_smertens_"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"Hi,
I built tlsctl, a small CLI for inspecting, testing, and debugging TLS connections:
https://github.com/catay/tlsctl
It aims to make TLS diagnostics more readable and structured than stitching together `openssl` and other ad-hoc commands. \nYou can inspect certificates and chains, check protocol support, and output in different formats.
Part of the motivation was practical (I got tired of parsing `openssl s_client` output), and part was educational.\nI wanted to build more hands-on experience with agentic engineering workflows. \nShout out to https://ampcode.com/ for their coding agent.
I\u2019d love feedback on usefulness, missing features, and whether this fits into real-world TLS troubleshooting workflows.
Thanks!
Steven"},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Tlsctl \u2013 A CLI for inspecting and troubleshooting TLS"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/catay/tlsctl"}},"_tags":["story","author__smertens_","story_47111796","show_hn"],"author":"_smertens_","created_at":"2026-02-22T15:32:31Z","created_at_i":1771774351,"num_comments":0,"objectID":"47111796","points":3,"story_id":47111796,"story_text":"Hi,
I built tlsctl, a small CLI for inspecting, testing, and debugging TLS connections:
https://github.com/catay/tlsctl
It aims to make TLS diagnostics more readable and structured than stitching together `openssl` and other ad-hoc commands. \nYou can inspect certificates and chains, check protocol support, and output in different formats.
Part of the motivation was practical (I got tired of parsing `openssl s_client` output), and part was educational.\nI wanted to build more hands-on experience with agentic engineering workflows. \nShout out to https://ampcode.com/ for their coding agent.
I\u2019d love feedback on usefulness, missing features, and whether this fits into real-world TLS troubleshooting workflows.
Thanks!
Steven","title":"Show HN: Tlsctl \u2013 A CLI for inspecting and troubleshooting TLS","updated_at":"2026-03-05T23:34:54Z","url":"https://github.com/catay/tlsctl"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"hnhn34"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"Nope,they use Agent SDK now https://ampcode.com/news/use-your-claude-plan"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"OpenCode 2.0"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://anoma.ly/notes/opencode-2-0/"}},"_tags":["comment","author_hnhn34","story_50094994"],"author":"hnhn34","comment_text":"Nope,they use Agent SDK now https://ampcode.com/news/use-your-claude-plan","created_at":"2026-10-15T09:07:39Z","created_at_i":1792055259,"objectID":"50101117","parent_id":50099514,"story_id":50094994,"story_title":"OpenCode 2.0","story_url":"https://anoma.ly/notes/opencode-2-0/","updated_at":"2026-10-15T09:12:15Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"pugz"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"Amp recently achieved it. Apparently it's permitted as long as you use the Claude Agent SDK. https://ampcode.com/news/use-your-claude-plan"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Codex CLI installs a persistent, self-updating background daemon by default"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/openai/codex/issues/48195"}},"_tags":["comment","author_pugz","story_50060694"],"author":"pugz","children":[50069544],"comment_text":"Amp recently achieved it. Apparently it's permitted as long as you use the Claude Agent SDK. https://ampcode.com/news/use-your-claude-plan","created_at":"2026-10-12T22:43:51Z","created_at_i":1791845031,"objectID":"50064110","parent_id":50061347,"story_id":50060694,"story_title":"Codex CLI installs a persistent, self-updating background daemon by default","story_url":"https://github.com/openai/codex/issues/48195","updated_at":"2026-10-13T10:34:25Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"impulser_"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"https://ampcode.com/what-are-orbs
Agents running in a remote vm."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Berthd"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://berthd.app/"}},"_tags":["comment","author_impulser_","story_49982735"],"author":"impulser_","children":[49986877],"comment_text":"https://ampcode.com/what-are-orbs
Agents running in a remote vm.","created_at":"2026-10-06T22:56:39Z","created_at_i":1791327399,"objectID":"49985349","parent_id":49983856,"story_id":49982735,"story_title":"Berthd","story_url":"https://berthd.app/","updated_at":"2026-10-07T01:41:18Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"impulser_"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"The more time goes on the more I think there is basically no moat in AI products because everyone is literally building the same shit. There is zero creativity in AI product right now. Yeah they might be different designs, but they are all the same exact product under the skin.
Cursor, Devin, Factory, Amp Code, Codex, Claude Codex, Antigravity, Conductor, OpenCode, T3 Code, Orca, Emdash, and hundreds of more of the same exact shit.
Yeah they might market it same fancy revolutionary technology like Orbs, but it not it just an agent harness local or remote. You can do orbs literally with Claude Code with a single prompt it not special.
We have AI that lets you build things freely and yet here we are all building the same exact thing.
This is just coding harness. The personal assistant space is the exact same."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Berthd"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://berthd.app/"}},"_tags":["comment","author_impulser_","story_49982735"],"author":"impulser_","children":[49983856,49983906],"comment_text":"The more time goes on the more I think there is basically no moat in AI products because everyone is literally building the same shit. There is zero creativity in AI product right now. Yeah they might be different designs, but they are all the same exact product under the skin.
Cursor, Devin, Factory, Amp Code, Codex, Claude Codex, Antigravity, Conductor, OpenCode, T3 Code, Orca, Emdash, and hundreds of more of the same exact shit.
Yeah they might market it same fancy revolutionary technology like Orbs, but it not it just an agent harness local or remote. You can do orbs literally with Claude Code with a single prompt it not special.
We have AI that lets you build things freely and yet here we are all building the same exact thing.
This is just coding harness. The personal assistant space is the exact same.","created_at":"2026-10-06T20:39:37Z","created_at_i":1791319177,"objectID":"49983772","parent_id":49982735,"story_id":49982735,"story_title":"Berthd","story_url":"https://berthd.app/","updated_at":"2026-10-07T21:09:50Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cavoirom"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"I dont know about Claude Code, but I'm using Amp Code since November 2025 and it don't need the mode at all, just conversation."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Plan mode is dead"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.aymannadeem.com/artificial/intelligence,/developer/tools/2026/09/24/plan-mode-is-dead.html"}},"_tags":["comment","author_cavoirom","story_49840054"],"author":"cavoirom","comment_text":"I dont know about Claude Code, but I'm using Amp Code since November 2025 and it don't need the mode at all, just conversation.","created_at":"2026-09-26T09:36:25Z","created_at_i":1790415385,"objectID":"49854848","parent_id":49840054,"story_id":49840054,"story_title":"Plan mode is dead","story_url":"https://www.aymannadeem.com/artificial/intelligence,/developer/tools/2026/09/24/plan-mode-is-dead.html","updated_at":"2026-09-26T09:37:43Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"fishtoaster"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"Can't say I've tried the CLI. I've mostly focused on the cloud agents, which I was explicitly looking for. I compared against cursor's cloud agents, ampcode, and hoplite, and came out surprisingly enjoying devin.
I will say that the lack of parity between Devin cloud and Devin desktop is downright embarrassing. It's very clear that the latter is a thinly-reskinned Windsurf. A visually similar UI with vastly different capabilities. Definitely a black mark on the whole thing."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://cognition.com/blog/swe-2"}},"_tags":["comment","author_fishtoaster","story_49645443"],"author":"fishtoaster","comment_text":"Can't say I've tried the CLI. I've mostly focused on the cloud agents, which I was explicitly looking for. I compared against cursor's cloud agents, ampcode, and hoplite, and came out surprisingly enjoying devin.
I will say that the lack of parity between Devin cloud and Devin desktop is downright embarrassing. It's very clear that the latter is a thinly-reskinned Windsurf. A visually similar UI with vastly different capabilities. Definitely a black mark on the whole thing.","created_at":"2026-09-11T04:23:45Z","created_at_i":1789100625,"objectID":"49653534","parent_id":49650669,"story_id":49645443,"story_title":"Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra","story_url":"https://cognition.com/blog/swe-2","updated_at":"2026-09-11T04:25:40Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cavoirom"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"AmpCode is not popular here but I recommend it, give it I try and you will not go back, except the price may a bit expensive."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Our decision on Cursor following its acquisition by SpaceX"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/"}},"_tags":["comment","author_cavoirom","story_49486172"],"author":"cavoirom","comment_text":"AmpCode is not popular here but I recommend it, give it I try and you will not go back, except the price may a bit expensive.","created_at":"2026-08-29T06:56:18Z","created_at_i":1787986578,"objectID":"49487590","parent_id":49487167,"story_id":49486172,"story_title":"Our decision on Cursor following its acquisition by SpaceX","story_url":"https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/","updated_at":"2026-08-29T07:22:15Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"yomismoaqui"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"Ok, human produced wall of text incoming, hope you read it later ;)
The core of an agent is an agentic loop, that is, you provide some tools to the LLM and you have a for loop that sends the user request to the LLM and if it wants to invoke some tools you do it and send the results aback to the LLM (and loop again). Also if the LLM returns some text you show it to the user. Then when the LLM has returned the last answer and there are no more tools to call you just exit the loop. This is all there is, I learned about it from this article https://ampcode.com/notes/how-to-build-an-agent.
Right now I provide the agent some core tools that are always available from the start:
- spawn_agent: The agent can create a subagent to do some task using another context, this way you can for example launch new agent to do some long thing and get just the result without filling the root agent context with junk.
- load_skill: the agent can load a skill by its id. The lists of available skills is provided on the system prompt this way the agent can dynamically load something like "websearch" if it needs it to provide an answer to the user's request.
In short: skills are just markdown with a description on their frontmatter section, this same description is what is used on the system prompt to give the LLM info about that this skill does. Also the skill has some tools that are enabled when that script is loaded. For example some tools on the skill that uses accessibility service to control the phone:
- search_app_package: returns the package id from some text (e.g. search for "Whatsapp" get "com.whatsapp")
- launch_app: launch app using package name
- read_screen_content: returns a textual representation of the accessibility tree on screen (that is, you can read all things on screen and see which you can interact with)
- find_element: finds element by text with optional type and flag for exact match.
- click_element: clicks on element by the id returned by find_element (or an id from read_screen_content)
- etc...
Then as you can imagine using the LLM to call these kind of tools is slow, non-deterministic and it costs money, so the next step is allowing the agent to create scripts in JS that can invoke these tools. The typical use case is asking the agent to do something like sending a message using the whatsapp application and seeing it fumble around while it opens the app, taps things and writes text on the right input text (hopefully). Then in the same chat you ask the agent to create a script to automatize this and it does something like this (simplified JS pseudocode):
result = await ("device.search_app_package", {text: "whatsapp"})\n await call("device.launch_app", {package: result.package_name})\n sleep(2000)\n result = await call("device.read_screen_content")\n ...\nThere are more details about the agent but this is the gist of it, ask about anything else if you want."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"The A.I. In Google's New Pixel 11 Is Not Helpful"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.nytimes.com/2026/08/19/technology/personaltech/google-pixel-11-review.html"}},"_tags":["comment","author_yomismoaqui","story_49365383"],"author":"yomismoaqui","children":[49389287],"comment_text":"Ok, human produced wall of text incoming, hope you read it later ;)
The core of an agent is an agentic loop, that is, you provide some tools to the LLM and you have a for loop that sends the user request to the LLM and if it wants to invoke some tools you do it and send the results aback to the LLM (and loop again). Also if the LLM returns some text you show it to the user. Then when the LLM has returned the last answer and there are no more tools to call you just exit the loop. This is all there is, I learned about it from this article https://ampcode.com/notes/how-to-build-an-agent.
Right now I provide the agent some core tools that are always available from the start:
- spawn_agent: The agent can create a subagent to do some task using another context, this way you can for example launch new agent to do some long thing and get just the result without filling the root agent context with junk.
- load_skill: the agent can load a skill by its id. The lists of available skills is provided on the system prompt this way the agent can dynamically load something like "websearch" if it needs it to provide an answer to the user's request.
In short: skills are just markdown with a description on their frontmatter section, this same description is what is used on the system prompt to give the LLM info about that this skill does. Also the skill has some tools that are enabled when that script is loaded. For example some tools on the skill that uses accessibility service to control the phone:
- search_app_package: returns the package id from some text (e.g. search for "Whatsapp" get "com.whatsapp")
- launch_app: launch app using package name
- read_screen_content: returns a textual representation of the accessibility tree on screen (that is, you can read all things on screen and see which you can interact with)
- find_element: finds element by text with optional type and flag for exact match.
- click_element: clicks on element by the id returned by find_element (or an id from read_screen_content)
- etc...
Then as you can imagine using the LLM to call these kind of tools is slow, non-deterministic and it costs money, so the next step is allowing the agent to create scripts in JS that can invoke these tools. The typical use case is asking the agent to do something like sending a message using the whatsapp application and seeing it fumble around while it opens the app, taps things and writes text on the right input text (hopefully). Then in the same chat you ask the agent to create a script to automatize this and it does something like this (simplified JS pseudocode):
result = await ("device.search_app_package", {text: "whatsapp"})\n await call("device.launch_app", {package: result.package_name})\n sleep(2000)\n result = await call("device.read_screen_content")\n ...\nThere are more details about the agent but this is the gist of it, ask about anything else if you want.","created_at":"2026-08-19T20:23:46Z","created_at_i":1787171026,"objectID":"49366721","parent_id":49366042,"story_id":49365383,"story_title":"The A.I. In Google's New Pixel 11 Is Not Helpful","story_url":"https://www.nytimes.com/2026/08/19/technology/personaltech/google-pixel-11-review.html","updated_at":"2026-08-21T15:08:17Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"6ty6thhDJEHDE"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"This is easy for Amp Code to do but not so easy for other people with different types of businesses.
I started at a business doing millions of customer transactions worth 100s of millions of dollars and their whole process was a complete YOLO Playground. As a result they lost money, people (it is high stress if everything breaks all the time) and most importantly: customer confidence.
What brought back some sanity? The basics for a mature business: deployment reviews and approvals, removing prod access, code reviews, QA in the loop. Release notes. All these things come up in audits. In SOC1 because the customer wants assurances, in SOC2 because we need to show prospects that we are serious and mature enough to get their business.
Please don't think it is just theatre. If you get away with bending the rules to the bare minimum to get a checkmark then good for you and all the power to you.
But if you are running a business where you are dealing with critical systems, PII, customer data, money, etc. Well maybe being very strict about having good change management policies in place will actually help you mature.
That maturity will also allow you to get customers. Because guess what, most customers who are going to trust you with their business have some expectations about how you run your ship.
I wish my company was as "simple" as Amp. No disrespect, I love what they do. But it doesn't really look like the kind of business where you get in serious trouble for using a minimal process with high trust in people.
Think about that when you read this. What could go wrong in your business. Don't project Amp's process on yours if your business is different."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"\"That's not SOC 2 compliant\""},"story_url":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"https://ampcode.com/notes/thats-not-soc-2-compliant"}},"_tags":["comment","author_6ty6thhDJEHDE","story_49308073"],"author":"6ty6thhDJEHDE","comment_text":"This is easy for Amp Code to do but not so easy for other people with different types of businesses.
I started at a business doing millions of customer transactions worth 100s of millions of dollars and their whole process was a complete YOLO Playground. As a result they lost money, people (it is high stress if everything breaks all the time) and most importantly: customer confidence.
What brought back some sanity? The basics for a mature business: deployment reviews and approvals, removing prod access, code reviews, QA in the loop. Release notes. All these things come up in audits. In SOC1 because the customer wants assurances, in SOC2 because we need to show prospects that we are serious and mature enough to get their business.
Please don't think it is just theatre. If you get away with bending the rules to the bare minimum to get a checkmark then good for you and all the power to you.
But if you are running a business where you are dealing with critical systems, PII, customer data, money, etc. Well maybe being very strict about having good change management policies in place will actually help you mature.
That maturity will also allow you to get customers. Because guess what, most customers who are going to trust you with their business have some expectations about how you run your ship.
I wish my company was as "simple" as Amp. No disrespect, I love what they do. But it doesn't really look like the kind of business where you get in serious trouble for using a minimal process with high trust in people.
Think about that when you read this. What could go wrong in your business. Don't project Amp's process on yours if your business is different.","created_at":"2026-08-15T12:48:52Z","created_at_i":1786798132,"objectID":"49310143","parent_id":49308073,"story_id":49308073,"story_title":"\"That's not SOC 2 compliant\"","story_url":"https://ampcode.com/notes/thats-not-soc-2-compliant","updated_at":"2026-08-17T15:31:18Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"navs"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"Ampcode used a handoff feature for a while that I found genuinely useful [1] and then they removed it. Anecdotally, I felt it worked better than compaction.
[1] https://ampcode.com/news/handoff"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"How Compaction Works in Pi"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://earendil.com/posts/compaction-in-pi/"}},"_tags":["comment","author_navs","story_49289654"],"author":"navs","children":[49296605],"comment_text":"Ampcode used a handoff feature for a while that I found genuinely useful [1] and then they removed it. Anecdotally, I felt it worked better than compaction.
[1] https://ampcode.com/news/handoff","created_at":"2026-08-14T03:30:31Z","created_at_i":1786678231,"objectID":"49294467","parent_id":49289654,"story_id":49289654,"story_title":"How Compaction Works in Pi","story_url":"https://earendil.com/posts/compaction-in-pi/","updated_at":"2026-08-14T09:53:05Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"r5Khe"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["ampcode"],"value":"Looks neat! I've been using Amp (https://ampcode.com/) for a while (Which seems to be doing something very similar), and I really appreciate this type of workflow. One thread = one VM feels like a solid model going forward."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Launch HN: Hoplite (YC S26) \u2013 Effortlessly deploy cloud coding agents"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://hoplite.sh"}},"_tags":["comment","author_r5Khe","story_49157997"],"author":"r5Khe","children":[49158959,49162201],"comment_text":"Looks neat! I've been using Amp (https://ampcode.com/) for a while (Which seems to be doing something very similar), and I really appreciate this type of workflow. One thread = one VM feels like a solid model going forward.","created_at":"2026-08-03T17:29:57Z","created_at_i":1785778197,"objectID":"49158876","parent_id":49157997,"story_id":49157997,"story_title":"Launch HN: Hoplite (YC S26) \u2013 Effortlessly deploy cloud coding agents","story_url":"https://hoplite.sh","updated_at":"2026-08-03T22:18:59Z"}],"hitsPerPage":20,"nbHits":4284,"nbPages":50,"page":0,"params":"query=ampcode&advancedSyntax=true&analyticsTags=backend","processingTimeMS":18,"processingTimingsMS":{"_request":{"roundTrip":20},"afterFetch":{"format":{"total":1},"merge":{"total":1},"total":1},"fetch":{"query":12,"scanning":3,"total":16},"total":18},"query":"ampcode","serverTimeMS":20}