With Claude Code, the goal is clearly to take a slice of Cursor and its competitors' market share. I expected this to happen eventually.
The app layer has barely any moat, so any successful app with the potential to generate significant revenue will eventually be absorbed by foundation model companies in their quest for growth and profits."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude","code"],"value":"Claude 3.7 Sonnet and Claude Code"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"https://www.anthropic.com/news/claude-3-7-sonnet"}},"_tags":["comment","author_fsndz","story_43163011"],"author":"fsndz","children":[43163777,43163840,43165178],"comment_text":"Anthropic is back and cementing its place as the creator of the best coding models\u2014bravo!
With Claude Code, the goal is clearly to take a slice of Cursor and its competitors' market share. I expected this to happen eventually.
The app layer has barely any moat, so any successful app with the potential to generate significant revenue will eventually be absorbed by foundation model companies in their quest for growth and profits.","created_at":"2025-02-24T19:15:05Z","created_at_i":1740424505,"objectID":"43163642","parent_id":43163488,"story_id":43163011,"story_title":"Claude 3.7 Sonnet and Claude Code","story_url":"https://www.anthropic.com/news/claude-3-7-sonnet","updated_at":"2025-03-03T01:29:25Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"stillpointlab"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"My experience is that I am bottle-necked on specs. The agent loop is less of a thing for me now.
If I can get a clear understanding of what I want to build, communicate that to Claude Code in planning mode with the goal to write an actionable spec (not code, plan to write the spec) then I tend to get very good results once the agent goes to implement.
But this strategy, while effective, puts a big load on me to write the specs. The agent tends to knock each one out of the park (usually 2 to 3 follow ups based on code review) but then I'm back at the stage that requires the spec.
Another issue for me is that when I step away, if the agent finishes a task and could technically start on an existing spec (no overlap on files so no conflict possible) it doesn't know it can just create a new branch and start. Before I go to bed I'll often say "do task X and once done and pushed start on task Y". But I haven't had luck beyond that. Often I find that it starts on Y and has a question and then the agent is idle the rest of the time.
The final issue is dependency coupled with the above. For example, today I was writing a background job processor. Obviously, the jobs that are in subsequent tasks require the system. That happens with some frequency. Even the specs need to be refreshed after the implementation to take any details that were resolved at coding time into account.
But I am just on the cusp of wanting the outer loop. The gate is almost entirely on spec creation and PR review. In places where those gates don't matter, I want the agent to keep chugging away.
As an aside, I strongly believe we need to start using tools that are better for LLMs even if they are worse for us. For example, Rust is annoying because the compiler is so strict. Bad for me, great for LLMs."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"The Coming Loop"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://lucumr.pocoo.org/2026/6/23/the-coming-loop/"}},"_tags":["comment","author_stillpointlab","story_48643180"],"author":"stillpointlab","children":[48652118,48652209,48652229,48654025,48654260,48656926,48658062],"comment_text":"My experience is that I am bottle-necked on specs. The agent loop is less of a thing for me now.
If I can get a clear understanding of what I want to build, communicate that to Claude Code in planning mode with the goal to write an actionable spec (not code, plan to write the spec) then I tend to get very good results once the agent goes to implement.
But this strategy, while effective, puts a big load on me to write the specs. The agent tends to knock each one out of the park (usually 2 to 3 follow ups based on code review) but then I'm back at the stage that requires the spec.
Another issue for me is that when I step away, if the agent finishes a task and could technically start on an existing spec (no overlap on files so no conflict possible) it doesn't know it can just create a new branch and start. Before I go to bed I'll often say "do task X and once done and pushed start on task Y". But I haven't had luck beyond that. Often I find that it starts on Y and has a question and then the agent is idle the rest of the time.
The final issue is dependency coupled with the above. For example, today I was writing a background job processor. Obviously, the jobs that are in subsequent tasks require the system. That happens with some frequency. Even the specs need to be refreshed after the implementation to take any details that were resolved at coding time into account.
But I am just on the cusp of wanting the outer loop. The gate is almost entirely on spec creation and PR review. In places where those gates don't matter, I want the agent to keep chugging away.
As an aside, I strongly believe we need to start using tools that are better for LLMs even if they are worse for us. For example, Rust is annoying because the compiler is so strict. Bad for me, great for LLMs.","created_at":"2026-06-23T21:46:22Z","created_at_i":1782251182,"objectID":"48651919","parent_id":48643180,"story_id":48643180,"story_title":"The Coming Loop","story_url":"https://lucumr.pocoo.org/2026/6/23/the-coming-loop/","updated_at":"2026-06-25T16:30:22Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"helloplanets"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"With the new Ultracode modes in Claude Code and Codex it's been taken to the next level. I mean, multi agent systems have been available for a long time, but the newest models seem to be much more RL'd for them.
Both GPT-5.6 and Fable will happily run 3+ hours straight off a relatively simple goal in that mode, burn through millions of tokens in the process.
Not saying it's great bang for your buck, but just that those modes are there and being pushed by the companies."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"How much energy do data centers and artificial intelligence use?"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://ourworldindata.org/how-much-energy-do-data-centers-and-artificial-intelligence-use"}},"_tags":["comment","author_helloplanets","story_49022412"],"author":"helloplanets","comment_text":"With the new Ultracode modes in Claude Code and Codex it's been taken to the next level. I mean, multi agent systems have been available for a long time, but the newest models seem to be much more RL'd for them.
Both GPT-5.6 and Fable will happily run 3+ hours straight off a relatively simple goal in that mode, burn through millions of tokens in the process.
Not saying it's great bang for your buck, but just that those modes are there and being pushed by the companies.","created_at":"2026-07-23T16:57:00Z","created_at_i":1784825820,"objectID":"49024738","parent_id":49024364,"story_id":49022412,"story_title":"How much energy do data centers and artificial intelligence use?","story_url":"https://ourworldindata.org/how-much-energy-do-data-centers-and-artificial-intelligence-use","updated_at":"2026-07-23T17:06:50Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"david_d8912"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"GPT-5.5 + Opus-4.7 here. Codex for pure coding task with clear goal, claude code for the rest. Also combined with opencode to experiment new models.
fyi: I didn't have much lock on Deepseek v4 pro, with opencode + openrouter it's incredibly slow. How did op did it?"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["mode"],"value":"Ask HN: What LLM models are you using and why?"}},"_tags":["comment","author_david_d8912","story_48166147"],"author":"david_d8912","comment_text":"GPT-5.5 + Opus-4.7 here. Codex for pure coding task with clear goal, claude code for the rest. Also combined with opencode to experiment new models.
fyi: I didn't have much lock on Deepseek v4 pro, with opencode + openrouter it's incredibly slow. How did op did it?","created_at":"2026-05-17T05:17:42Z","created_at_i":1778995062,"objectID":"48166219","parent_id":48166147,"story_id":48166147,"story_title":"Ask HN: What LLM models are you using and why?","updated_at":"2026-05-17T20:17:22Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"Topfi"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"Additional information may be found here: https://developers.openai.com/codex/ide
Am seemingly one of very few people to somewhat reliably get usable results from GPT-5 and have been relying on Codex Cloud more and more, especially as an alternative to having to remote into my computer when prompting for minor changes on the phone. This new IDE plugin, naturally, takes very strongly after Cursor, Windsurf, etc. but also does better than some in unifying both Cloud and CLI changes, something I have seen others separate more strictly in their UI.
Last time I tested a none CLI LLM coding assistance solutions was when Amazon released KIRO and that was mainly focused on verifying whether there is something I could/should adapt from their way of doing requirements/planning implementations, so I am no expert in the UI of alternatives like Cursor, Windsurf, etc. but in my limited testing, Codex makes diffs very pleasant to read. Would like the added ability to expand individual files over only having them in a collapsable list via "View changes" though. I also very much enjoy the Todo list implementation I have seen on some tasks, it is a bit more informative than the more broad goals Claude Code or CLI Codex tend to provide, while retaining the full, step-by-step work log as an added source of information. Have not found a way to reliably trigger this feature though.
Also, some aria-labels are overflowing/covered by other UI elements making them unreadable and I'd prefer this to be placed in a positionable panel, rather than the primary side bar, cause as it stands, one cannot have the Codex panel and e.g. version control, file explorer or search open. Lastly, trepidations of unrestricted local LLM use aside, I don't think the "Agent (full access)" button should expand like that: https://imgur.com/a/T1slbxk
Am also testing the new Code Review implementation they have added, was surprised to see the model communicate via emojis/reactions of all things, but haven't yet tried any LLM code review solutions so again don't know how these usually communicate state/provide confirmation of actions: https://developers.openai.com/codex/cloud"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"OpenAI Codex for VSCode has been released"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://marketplace.visualstudio.com/items?itemName=openai.chatgpthttps://marketplace.visualstudio.com/items?itemName=openai.chatgpt"}},"_tags":["comment","author_Topfi","story_45045015"],"author":"Topfi","comment_text":"Additional information may be found here: https://developers.openai.com/codex/ide
Am seemingly one of very few people to somewhat reliably get usable results from GPT-5 and have been relying on Codex Cloud more and more, especially as an alternative to having to remote into my computer when prompting for minor changes on the phone. This new IDE plugin, naturally, takes very strongly after Cursor, Windsurf, etc. but also does better than some in unifying both Cloud and CLI changes, something I have seen others separate more strictly in their UI.
Last time I tested a none CLI LLM coding assistance solutions was when Amazon released KIRO and that was mainly focused on verifying whether there is something I could/should adapt from their way of doing requirements/planning implementations, so I am no expert in the UI of alternatives like Cursor, Windsurf, etc. but in my limited testing, Codex makes diffs very pleasant to read. Would like the added ability to expand individual files over only having them in a collapsable list via "View changes" though. I also very much enjoy the Todo list implementation I have seen on some tasks, it is a bit more informative than the more broad goals Claude Code or CLI Codex tend to provide, while retaining the full, step-by-step work log as an added source of information. Have not found a way to reliably trigger this feature though.
Also, some aria-labels are overflowing/covered by other UI elements making them unreadable and I'd prefer this to be placed in a positionable panel, rather than the primary side bar, cause as it stands, one cannot have the Codex panel and e.g. version control, file explorer or search open. Lastly, trepidations of unrestricted local LLM use aside, I don't think the "Agent (full access)" button should expand like that: https://imgur.com/a/T1slbxk
Am also testing the new Code Review implementation they have added, was surprised to see the model communicate via emojis/reactions of all things, but haven't yet tried any LLM code review solutions so again don't know how these usually communicate state/provide confirmation of actions: https://developers.openai.com/codex/cloud","created_at":"2025-08-27T20:50:53Z","created_at_i":1756327853,"objectID":"45045080","parent_id":45045015,"story_id":45045015,"story_title":"OpenAI Codex for VSCode has been released","story_url":"https://marketplace.visualstudio.com/items?itemName=openai.chatgpthttps://marketplace.visualstudio.com/items?itemName=openai.chatgpt","updated_at":"2026-03-05T22:32:37Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"ghshephard"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"I use cursor 8+ hours/day at work, and have full (and effectively unlimited) access to Claude Code and Codex - tools which I also use personally. I suspect that your "constant popups" were when you were using the editor - a mode that I'll confess I haven't touched in 3+ months.
Workflow in Cursor is actually awesome - I'm a little outdated in how I use it - I still establish goals/objectives, rather than managing the loop which does so - but if you can think broadly enough - I find it's pretty efficient.
Key things I like about Cursor (and I recognize I'm dating myself a bit here):\n - Plan Mode is really solid - I shift-tab, have it go create the plan using whatever insanely expensive SOTA model is available - I will usually spend 5-10 minutes on the Plan - review it, maybe even tweak it a little. (though 90% of the time it's fine out of the gate)
- Ability to select any model for every task - I'll switch between Opus 4.8 High/xHigh/... I'll even switch to 1M context for the planning phase upfront. \n\n - It does an *excellent* job managing permissions and looping the agents and spinning up sub-agents for you - you set the goal, run the plan mode - and then let it churn for however long is required - pretty common to have a 30-45 minute run and come back to a fully created/tested product. \n\n Workflow in Cursor is actually awesome - I'm a little outdated in how I use it - I still establish goals/objectives, rather than managing the loop which does so - but if you can think broadly enough - I find it's pretty efficient.
Key things I like about Cursor (and I recognize I'm dating myself a bit here):\n - Plan Mode is really solid - I shift-tab, have it go create the plan using whatever insanely expensive SOTA model is available - I will usually spend 5-10 minutes on the Plan - review it, maybe even tweak it a little. (though 90% of the time it's fine out of the gate)
- Ability to select any model for every task - I'll switch between Opus 4.8 High/xHigh/... I'll even switch to 1M context for the planning phase upfront. \n\n - It does an *excellent* job managing permissions and looping the agents and spinning up sub-agents for you - you set the goal, run the plan mode - and then let it churn for however long is required - pretty common to have a 30-45 minute run and come back to a fully created/tested product. \n\n I think teams need to be able to write nested workflows that transition between code-led and agent-led, with either supporting human-in-the-loop checkpoints.
Been iterating on what this should look like at our startup (https://www.amika.dev/). Model labs are also improving capabilities here, such as Codex's `/goal` and Claude Code's dynamic workflows[1]
The points about API usage cost still stand, but model intelligence is getting cheaper every month! No need to use the frontier model for every part of the work.
[1]: https://code.claude.com/docs/en/workflows"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Backpressure is all you need"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.lucasfcosta.com/blog/backpressure-is-all-you-need"}},"_tags":["comment","author_dbmikus","story_48345090"],"author":"dbmikus","children":[48350661],"comment_text":"To stop agents from pausing for checkpointing, you can have a deterministic outer loop that re-runs until a stop condition is met.
I think teams need to be able to write nested workflows that transition between code-led and agent-led, with either supporting human-in-the-loop checkpoints.
Been iterating on what this should look like at our startup (https://www.amika.dev/). Model labs are also improving capabilities here, such as Codex's `/goal` and Claude Code's dynamic workflows[1]
The points about API usage cost still stand, but model intelligence is getting cheaper every month! No need to use the frontier model for every part of the work.
[1]: https://code.claude.com/docs/en/workflows","created_at":"2026-05-31T21:29:47Z","created_at_i":1780262987,"objectID":"48349943","parent_id":48345557,"story_id":48345090,"story_title":"Backpressure is all you need","story_url":"https://www.lucasfcosta.com/blog/backpressure-is-all-you-need","updated_at":"2026-05-31T23:07:16Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"gck1"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"Every proprietary harness is just proprietary junk without ability to extend it without polluting context. This includes claude-code, gemini-cli, codex etc. They have tools which hardcode the behavior that is impossible to modify, they add tools you may not need that pollute context, they inject an entire textbook's worth of words into the system prompt which pollutes context, they provide zero observability into what the agent is doing when it's launching a subagent as one example.
They don't provide easy way to use multiple models from multiple providers for varying tasks. One model may be the best thing on earth at one thing, but fail miserably for another. Try orchestrating multiple agents from claude, gemini and codex in any of these proprietary boxes.
They also... suck at TUI UX. I don't know if it was fixed already, but claude code had flickering issue that was unresolved for more than a year.
You need to take a very good care of what goes into your context. A black box of proprietary harnesses is not it. Check out pi [1] for example, which is a very minimal harness with really nice extension system. The idea is that you start with barebones and add things that you need for your own goals.
claude-code HAS to have all these bells and whistles that pollute context to support larger audiences that can't tinker with it. If you have the ability to only pull in only what you need and extend things in a way that works for your workflow, you'll always get the best experience. And claude-code may never be that without making it complicated for the masses. OSS will always win here.
[1] https://github.com/badlogic/pi-mono"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"How will OpenAI compete?"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://www.ben-evans.com/benedictevans/2026/2/19/how-will-openai-compete-nkg2x"}},"_tags":["comment","author_gck1","story_47158975"],"author":"gck1","children":[47166653],"comment_text":"Every proprietary harness is just proprietary junk without ability to extend it without polluting context. This includes claude-code, gemini-cli, codex etc. They have tools which hardcode the behavior that is impossible to modify, they add tools you may not need that pollute context, they inject an entire textbook's worth of words into the system prompt which pollutes context, they provide zero observability into what the agent is doing when it's launching a subagent as one example.
They don't provide easy way to use multiple models from multiple providers for varying tasks. One model may be the best thing on earth at one thing, but fail miserably for another. Try orchestrating multiple agents from claude, gemini and codex in any of these proprietary boxes.
They also... suck at TUI UX. I don't know if it was fixed already, but claude code had flickering issue that was unresolved for more than a year.
You need to take a very good care of what goes into your context. A black box of proprietary harnesses is not it. Check out pi [1] for example, which is a very minimal harness with really nice extension system. The idea is that you start with barebones and add things that you need for your own goals.
claude-code HAS to have all these bells and whistles that pollute context to support larger audiences that can't tinker with it. If you have the ability to only pull in only what you need and extend things in a way that works for your workflow, you'll always get the best experience. And claude-code may never be that without making it complicated for the masses. OSS will always win here.
[1] https://github.com/badlogic/pi-mono","created_at":"2026-02-26T10:24:43Z","created_at_i":1772101483,"objectID":"47164172","parent_id":47163889,"story_id":47158975,"story_title":"How will OpenAI compete?","story_url":"https://www.ben-evans.com/benedictevans/2026/2/19/how-will-openai-compete-nkg2x","updated_at":"2026-03-05T23:37:43Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"chrisischris"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"I was hitting Claude Code's rate limit pretty often while paying for their max subscription. Started thinking \u2013 I've got a decent GPU sitting at home doing nothing most of the day.
So I'm building a distributed AI inference platform where you can run models on your own hardware and access it from anywhere privately. Keep your data on infrastructure you control, but also leverage a credit system to tap into more powerful compute when you need it. Your idle GPU time can earn credits for accessing bigger models.\nThe goal is making it dead simple to use your home hardware from wherever you're working.
It's for anyone who wants infrastructure optionality: developers who don't want vendor lock-in, businesses with compliance requirements, or just people who don't want their data sent to third parties.
Get notified when we launch: https://sporeintel.com"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Ask HN: What Are You Working On? (Nov 2025)"}},"_tags":["comment","author_chrisischris","story_45869146"],"author":"chrisischris","children":[45879559,45880916,45934706],"comment_text":"I was hitting Claude Code's rate limit pretty often while paying for their max subscription. Started thinking \u2013 I've got a decent GPU sitting at home doing nothing most of the day.
So I'm building a distributed AI inference platform where you can run models on your own hardware and access it from anywhere privately. Keep your data on infrastructure you control, but also leverage a credit system to tap into more powerful compute when you need it. Your idle GPU time can earn credits for accessing bigger models.\nThe goal is making it dead simple to use your home hardware from wherever you're working.
It's for anyone who wants infrastructure optionality: developers who don't want vendor lock-in, businesses with compliance requirements, or just people who don't want their data sent to third parties.
Get notified when we launch: https://sporeintel.com","created_at":"2025-11-10T18:49:34Z","created_at_i":1762800574,"objectID":"45879351","parent_id":45869146,"story_id":45869146,"story_title":"Ask HN: What Are You Working On? (Nov 2025)","updated_at":"2026-03-05T23:01:12Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"pranshu54"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"After months of feedback from devs juggling multiple chat tools just to break big tasks into smaller steps, we re\u2011imagined our workflow as a Kanban\u2011style Phase Board right inside your favourite IDE. The new Phase mode turns any large task into a clean sequence of PR\u2011sized phases you can review and commit one by one.
How it works
1. Describe the goal (Task Query) \u2013 In Phase mode, type a concise description of what you want to build or change. Example: \u201cAdd rate\u2011limit middleware and expose a /metrics endpoint.\u201d Traycer treats this as the parent task.\n2. Clarify intent (AI follow\u2011up) \u2013 Traycer may ask one or two quick questions (constraints, coding style). Answer them so the scope is crystal\u2011clear.\n3. Auto\u2011generate the Phase Board \u2013 Traycer breaks the task into a sequential list of PR\u2011sized phases you can reorder, edit, or delete.\n4. Open a phase & generate its plan \u2013 get a detailed file\u2011level plan: which files, functions, symbols, and tests will be touched.\n5. Handoff to your coding agent \u2013 Hit Execute to send that plan straight to Cursor, Claude Code, or any agent you prefer.\n6. Verify the diff \u2013 When your agent finishes, Traycer compares the diff to the plan and checks compatibility with upcoming phases, flagging any mismatches.\n7. Review & commit (or tweak) \u2013 Approve and commit the phase, or adjust the plan and rerun. Then move on to the next phase.
Why it helps?
* True PR checkpoints \u2013 every phase is small enough to reason about and ship.\n* No runaway prompts \u2013 only the active phase is in context, so tokens stay low and results stay focused.\n* Tool-agnostic \u2013 Traycer plans and verifies; your coding agent writes code.\n* Fast course-correction \u2013 if something feels off, just edit that phase and re-run.
Try it out & share feedback
Install the Traycer extension (https://traycer.ai/installation), create a new task, and the Phase Board will appear. Add a few phases, run one through, and see how the PR\u2011sized checkpoints feel in practice.\nIf you have suggestions that could make the flow smoother, drop them in the comments - every bit of feedback helps."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Kanban-style Phase Board: plan \u2192 execute \u2192 verify \u2192 commit"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://traycer.ai/"}},"_tags":["story","author_pranshu54","story_44757409","show_hn"],"author":"pranshu54","created_at":"2025-08-01T14:36:00Z","created_at_i":1754058960,"num_comments":0,"objectID":"44757409","points":5,"story_id":44757409,"story_text":"After months of feedback from devs juggling multiple chat tools just to break big tasks into smaller steps, we re\u2011imagined our workflow as a Kanban\u2011style Phase Board right inside your favourite IDE. The new Phase mode turns any large task into a clean sequence of PR\u2011sized phases you can review and commit one by one.
How it works
1. Describe the goal (Task Query) \u2013 In Phase mode, type a concise description of what you want to build or change. Example: \u201cAdd rate\u2011limit middleware and expose a /metrics endpoint.\u201d Traycer treats this as the parent task.\n2. Clarify intent (AI follow\u2011up) \u2013 Traycer may ask one or two quick questions (constraints, coding style). Answer them so the scope is crystal\u2011clear.\n3. Auto\u2011generate the Phase Board \u2013 Traycer breaks the task into a sequential list of PR\u2011sized phases you can reorder, edit, or delete.\n4. Open a phase & generate its plan \u2013 get a detailed file\u2011level plan: which files, functions, symbols, and tests will be touched.\n5. Handoff to your coding agent \u2013 Hit Execute to send that plan straight to Cursor, Claude Code, or any agent you prefer.\n6. Verify the diff \u2013 When your agent finishes, Traycer compares the diff to the plan and checks compatibility with upcoming phases, flagging any mismatches.\n7. Review & commit (or tweak) \u2013 Approve and commit the phase, or adjust the plan and rerun. Then move on to the next phase.
Why it helps?
* True PR checkpoints \u2013 every phase is small enough to reason about and ship.\n* No runaway prompts \u2013 only the active phase is in context, so tokens stay low and results stay focused.\n* Tool-agnostic \u2013 Traycer plans and verifies; your coding agent writes code.\n* Fast course-correction \u2013 if something feels off, just edit that phase and re-run.
Try it out & share feedback
Install the Traycer extension (https://traycer.ai/installation), create a new task, and the Phase Board will appear. Add a few phases, run one through, and see how the PR\u2011sized checkpoints feel in practice.\nIf you have suggestions that could make the flow smoother, drop them in the comments - every bit of feedback helps.","title":"Show HN: Kanban-style Phase Board: plan \u2192 execute \u2192 verify \u2192 commit","updated_at":"2025-08-01T23:54:35Z","url":"https://traycer.ai/"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"allnew_llc"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"LLM coding agents (Claude Code, Codex, etc.) have structural weaknesses: blind retry loops, volatile learning, silent\n spec drift, and gate gaming. These aren't bugs in specific models \u2014 they're properties of goal-optimizing systems.
ANDON applies Toyota Production System principles to address this:\n\n - Jidoka (autonomation): auto-detect failures and block forward-progress commands (git push, deploy)\n - Kaizen (continuous improvement): force Five Whys root cause analysis, then standardize prevention rules\n - Meta-ANDON: detect when the agent is stuck in a whack-a-mole debugging loop\n\n Install: pip install andon-for-llm-agents\n\n Works with any agent that supports hooks/callbacks. Apache-2.0 licensed. ANDON applies Toyota Production System principles to address this:\n\n - Jidoka (autonomation): auto-detect failures and block forward-progress commands (git push, deploy)\n - Kaizen (continuous improvement): force Five Whys root cause analysis, then standardize prevention rules\n - Meta-ANDON: detect when the agent is stuck in a whack-a-mole debugging loop\n\n Install: pip install andon-for-llm-agents\n\n Works with any agent that supports hooks/callbacks. Apache-2.0 licensed.It provides ~70 functions across four packages:
Sync and Async -- map, filter, detect, reduce, every, some, reject, groupBy -- for both slices and maps. Every async function runs iteratees in goroutines with panic recovery.\nConcurrency limiting -- all async functions have a Limit variant that caps goroutine count via a semaphore.\nWorker pool -- a generic work queue with configurable buffer, concurrency, timeouts, and graceful shutdown.\nControl flow -- waterfall-style sequential execution with type-safe context passing.
Disclaimer: Parts of the code were generated using Claude Code. One of my goals with this project was to experiment with the latest models and see what they're capable of in a real codebase -- from writing implementations to tests to docs.
go get github.com/skatiyar/goutils@v1.0.0
Would love feedback on the API design and whether the function naming feels natural for Go."},"title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Goutils \u2013 70 type-safe generic functions for async/functional Go"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/skatiyar/goutils"}},"_tags":["story","author_skatiyar","story_47022476","show_hn"],"author":"skatiyar","created_at":"2026-02-15T09:56:23Z","created_at_i":1771149383,"num_comments":0,"objectID":"47022476","points":3,"story_id":47022476,"story_text":"I built a Go utility library inspired by the JavaScript async library, implemented idiomatically with Go 1.18+ generics. Avoids using reflection and interface{} wherever possible.
It provides ~70 functions across four packages:
Sync and Async -- map, filter, detect, reduce, every, some, reject, groupBy -- for both slices and maps. Every async function runs iteratees in goroutines with panic recovery.\nConcurrency limiting -- all async functions have a Limit variant that caps goroutine count via a semaphore.\nWorker pool -- a generic work queue with configurable buffer, concurrency, timeouts, and graceful shutdown.\nControl flow -- waterfall-style sequential execution with type-safe context passing.
Disclaimer: Parts of the code were generated using Claude Code. One of my goals with this project was to experiment with the latest models and see what they're capable of in a real codebase -- from writing implementations to tests to docs.
go get github.com/skatiyar/goutils@v1.0.0
Would love feedback on the API design and whether the function naming feels natural for Go.","title":"Show HN: Goutils \u2013 70 type-safe generic functions for async/functional Go","updated_at":"2026-03-05T23:33:01Z","url":"https://github.com/skatiyar/goutils"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kirumachi"},"story_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"Hey HN,
I have been using Claude Code and Cursor lately and as we all know, they write code incredibly fast but a few times i have noticed they can introduce the same security flaws. For example, you ask the LLM to build a file upload feature, you will get working code in minutes, but it would almost always miss magic-byte validation or leaves you vulnerable to SVG XSS. The LLM optimizes for code that compiles not code that is secure.
To fix this for my own workflow, I made a set of 8 security-focused AI agents (AppSec, GRC, Cloud/Platform, etc) that you can drop into any MCP-compatible tool (Cursor, Windsurf) or use with Claude Code.
To clarify, the goal here is not to say that LLM/AI replaces AppSec or the Secure Software Development Cycle, instead the goal is to provide a series of structured prompts and concrete security artifacts (like STRIDE based threat models and ASVS mapped functional requirements) for developers who are already using AI to write code. The aim is to force the LLM to pause and sort of put on a security hat during specific phases of the SDLC.
What It Actually Is
It is an MIT licenced repo containing the agent prompts, document templates and an MCP server. You can install via Claude marketplace or globally via npm, which gives you a CLI to scaffold git hooks, CodeQL CI Gates and the MCP config. Also included are 3 full walkthroughs in the repo showing how the agents catch things.
I am an Application Security Engineer, and I am really curious to get feedback and critique. Please try it out without any signups using the URL. I will be around to answer any questions"},"title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude"],"value":"Show HN: Secure SDLC Agents for Claude and Cursor (MCP)"},"url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/Kaademos/secure-sdlc-agents"}},"_tags":["story","author_kirumachi","story_47672458","show_hn"],"author":"kirumachi","children":[47677559],"created_at":"2026-04-07T09:00:01Z","created_at_i":1775552401,"num_comments":0,"objectID":"47672458","points":1,"story_id":47672458,"story_text":"Hey HN,
I have been using Claude Code and Cursor lately and as we all know, they write code incredibly fast but a few times i have noticed they can introduce the same security flaws. For example, you ask the LLM to build a file upload feature, you will get working code in minutes, but it would almost always miss magic-byte validation or leaves you vulnerable to SVG XSS. The LLM optimizes for code that compiles not code that is secure.
To fix this for my own workflow, I made a set of 8 security-focused AI agents (AppSec, GRC, Cloud/Platform, etc) that you can drop into any MCP-compatible tool (Cursor, Windsurf) or use with Claude Code.
To clarify, the goal here is not to say that LLM/AI replaces AppSec or the Secure Software Development Cycle, instead the goal is to provide a series of structured prompts and concrete security artifacts (like STRIDE based threat models and ASVS mapped functional requirements) for developers who are already using AI to write code. The aim is to force the LLM to pause and sort of put on a security hat during specific phases of the SDLC.
What It Actually Is
It is an MIT licenced repo containing the agent prompts, document templates and an MCP server. You can install via Claude marketplace or globally via npm, which gives you a CLI to scaffold git hooks, CodeQL CI Gates and the MCP config. Also included are 3 full walkthroughs in the repo showing how the agents catch things.
I am an Application Security Engineer, and I am really curious to get feedback and critique. Please try it out without any signups using the URL. I will be around to answer any questions","title":"Show HN: Secure SDLC Agents for Claude and Cursor (MCP)","updated_at":"2026-04-12T19:06:15Z","url":"https://github.com/Kaademos/secure-sdlc-agents"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"cedws"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"
ANTI_DISTILLATION_CC\n \n This is Anthropic's anti-distillation defence baked into Claude Code. When enabled, it injects anti_distillation: ['fake_tools'] into every API request, which causes the server to silently slip decoy tool definitions into the model's system prompt. The goal: if someone is scraping Claude Code's API traffic to train a competing model, the poisoned training data makes that distillation attempt less useful. ANTI_DISTILLATION_CC\n \n This is Anthropic's anti-distillation defence baked into Claude Code. When enabled, it injects anti_distillation: ['fake_tools'] into every API request, which causes the server to silently slip decoy tool definitions into the model's system prompt. The goal: if someone is scraping Claude Code's API traffic to train a competing model, the poisoned training data makes that distillation attempt less useful.- Base Claude Code (released)
- Extensive, self-orchestrated, local specs & documentation; ie waterfall for many features/longer term project goals (summer)
- Base Claude Code (today)
Claude Code is getting better at orchestrating it's own subagents for divide/conquer type work.
My problem with these extensive self-orchestrated multi-agent / spec modes is the type of drift and rot of all the changes and then integrated parts of an application that a lot of the time end up in merge conflicts. Aside from my own decision cognitive space, it's also a lot to just generally orchestrate and review. I spent a ton of type enforcing Claude to use the system I put in place including documentation updates and continuous logging of work.
I feel extremely productive with a single Claude Code for a project. Maybe for minor features, I'll launch Claude Code in the web so that it can operate in an isolated space to knock them out and create a PR.
I will plan and annotate extensively for large features, but not many features or broad project specs all at the same time. Annotation and better planning UX, I think, are going to be increasingly important for now. The only augment of Claude Code I have is a hook for plan mode review: https://github.com/backnotprop/plannotator"},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Parallel coding agents with tmux and Markdown specs"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://schipper.ai/posts/parallel-coding-agents/"}},"_tags":["comment","author_ramoz","story_47218318"],"author":"ramoz","children":[47222888],"comment_text":"I did a sort of bell curve with this type of workflow over summer.
- Base Claude Code (released)
- Extensive, self-orchestrated, local specs & documentation; ie waterfall for many features/longer term project goals (summer)
- Base Claude Code (today)
Claude Code is getting better at orchestrating it's own subagents for divide/conquer type work.
My problem with these extensive self-orchestrated multi-agent / spec modes is the type of drift and rot of all the changes and then integrated parts of an application that a lot of the time end up in merge conflicts. Aside from my own decision cognitive space, it's also a lot to just generally orchestrate and review. I spent a ton of type enforcing Claude to use the system I put in place including documentation updates and continuous logging of work.
I feel extremely productive with a single Claude Code for a project. Maybe for minor features, I'll launch Claude Code in the web so that it can operate in an isolated space to knock them out and create a PR.
I will plan and annotate extensively for large features, but not many features or broad project specs all at the same time. Annotation and better planning UX, I think, are going to be increasingly important for now. The only augment of Claude Code I have is a hook for plan mode review: https://github.com/backnotprop/plannotator","created_at":"2026-03-02T19:02:15Z","created_at_i":1772478135,"objectID":"47222472","parent_id":47218318,"story_id":47218318,"story_title":"Parallel coding agents with tmux and Markdown specs","story_url":"https://schipper.ai/posts/parallel-coding-agents/","updated_at":"2026-03-05T23:39:59Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"SamInTheShell"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"I'm inclined to isolate the chat processes only if I keep the bash tool. I'm undecided if I'm keeping it in. Implementing an MCP server in python is dead easy and so is making a bash tool.
Right now I'm more interested in getting ACP working for gemini-cli and claude-code to serve it their models. My first goal is just to make the manually operated tool that just gets out of the way or whatever. Sane permissions out of the box.
If anyone wants to just come along and add an optional feature today, I'm happy to merge under the same license. Otherewise, I will eventually add this feature, I'm just not sure if it will be sooner or later."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Show HN: Vibe coded an AI chat app with features I wanted, Poe"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/SamInTheShell/poe"}},"_tags":["comment","author_SamInTheShell","story_46014362"],"author":"SamInTheShell","comment_text":"I'm inclined to isolate the chat processes only if I keep the bash tool. I'm undecided if I'm keeping it in. Implementing an MCP server in python is dead easy and so is making a bash tool.
Right now I'm more interested in getting ACP working for gemini-cli and claude-code to serve it their models. My first goal is just to make the manually operated tool that just gets out of the way or whatever. Sane permissions out of the box.
If anyone wants to just come along and add an optional feature today, I'm happy to merge under the same license. Otherewise, I will eventually add this feature, I'm just not sure if it will be sooner or later.","created_at":"2025-11-23T04:24:30Z","created_at_i":1763871870,"objectID":"46020738","parent_id":46019771,"story_id":46014362,"story_title":"Show HN: Vibe coded an AI chat app with features I wanted, Poe","story_url":"https://github.com/SamInTheShell/poe","updated_at":"2026-03-05T23:05:36Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"mccoyb"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"Same experience here, probably in a slightly different way of work (PhD student). Was extremely skeptical of LLMs, Claude Code has completely transformed the way I work.
It doesn't take away the requirements of _curation_ - that remains firmly in my camp (partially what a PhD is supposed to teach you! to be precise and reflective about why you are doing X, what do you hope to show with Y, etc -- breakdown every single step, explain those steps to someone else -- this is a tremendous soft skill, and it's even more important now because these agents do not have persistent world models / immediately forget the goal of a sequence of interactions, even with clever compaction).
If I'm on my game with precise communication, I can use CC to organize computation in a way which has never been possible before.
It's not easier than programming (if you care about quality!), but it is different, and it comes with different idioms."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"Everything around LLMs is still magical and wishful thinking"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://dmitriid.com/everything-around-llms-is-still-magical-and-wishful-thinking"}},"_tags":["comment","author_mccoyb","story_44467949"],"author":"mccoyb","comment_text":"Same experience here, probably in a slightly different way of work (PhD student). Was extremely skeptical of LLMs, Claude Code has completely transformed the way I work.
It doesn't take away the requirements of _curation_ - that remains firmly in my camp (partially what a PhD is supposed to teach you! to be precise and reflective about why you are doing X, what do you hope to show with Y, etc -- breakdown every single step, explain those steps to someone else -- this is a tremendous soft skill, and it's even more important now because these agents do not have persistent world models / immediately forget the goal of a sequence of interactions, even with clever compaction).
If I'm on my game with precise communication, I can use CC to organize computation in a way which has never been possible before.
It's not easier than programming (if you care about quality!), but it is different, and it comes with different idioms.","created_at":"2025-07-04T22:08:29Z","created_at_i":1751666909,"objectID":"44468269","parent_id":44468153,"story_id":44467949,"story_title":"Everything around LLMs is still magical and wishful thinking","story_url":"https://dmitriid.com/everything-around-llms-is-still-magical-and-wishful-thinking","updated_at":"2025-07-05T23:39:04Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"tpicks"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"> Claude Code uses Haiku to read through the transcript and decide if the goal has been completed.
feels kinda odd to use a less capable model to determine if the goal is fully complete. Especially if the user is expecting /goal to thoroughly complete the task. A less capable model would be more likely to misclassify `isComplete?`"},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["goal"],"value":"Fable 5 vs. GPT-5.6 Sol on an NP-Hard Problem: Does /goal help?"},"story_url":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["goal"],"value":"https://charlesazam.com/blog/fable-5-gpt-5-6-sol-goal/"}},"_tags":["comment","author_tpicks","story_48956879"],"author":"tpicks","children":[48965008],"comment_text":"> Claude Code uses Haiku to read through the transcript and decide if the goal has been completed.
feels kinda odd to use a less capable model to determine if the goal is fully complete. Especially if the user is expecting /goal to thoroughly complete the task. A less capable model would be more likely to misclassify `isComplete?`","created_at":"2026-07-18T22:27:43Z","created_at_i":1784413663,"objectID":"48963084","parent_id":48959819,"story_id":48956879,"story_title":"Fable 5 vs. GPT-5.6 Sol on an NP-Hard Problem: Does /goal help?","story_url":"https://charlesazam.com/blog/fable-5-gpt-5-6-sol-goal/","updated_at":"2026-07-19T04:35:48Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"kakugawa"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"How much does /goal actually help? In auto mode, I've tried using and not using /goal and I haven't felt a difference.
https://code.claude.com/docs/en/goal#how-evaluation-works
> /goal is a wrapper around a session-scoped prompt-based Stop hook. Each time Claude finishes a turn, the condition and the conversation so far are sent to your configured small fast model, which defaults to Haiku. The model returns a yes-or-no decision and a short reason. A \u201cno\u201d tells Claude to keep working and includes the reason as guidance for the next turn. A \u201cyes\u201d clears the goal and records an achieved entry in the transcript.
> The evaluator runs on whichever provider your session is configured for. It does not call tools, so it can only judge what Claude has already surfaced in the conversation.
Apparently, it uses Haiku (by default) to evaluate every turn to determine if the goal has been achieved. However, it only relies on the transcript itself (including the reasoning of the main model). It can't independently verify if the goal has been achieved. So, if the main model thinks the goal is or isn't done, how often does Haiku disagree (in a productive way)? That's not clear to me."},"story_title":{"matchLevel":"none","matchedWords":[],"value":"The Coming Loop"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://lucumr.pocoo.org/2026/6/23/the-coming-loop/"}},"_tags":["comment","author_kakugawa","story_48643180"],"author":"kakugawa","children":[48649989],"comment_text":"How much does /goal actually help? In auto mode, I've tried using and not using /goal and I haven't felt a difference.
https://code.claude.com/docs/en/goal#how-evaluation-works
> /goal is a wrapper around a session-scoped prompt-based Stop hook. Each time Claude finishes a turn, the condition and the conversation so far are sent to your configured small fast model, which defaults to Haiku. The model returns a yes-or-no decision and a short reason. A \u201cno\u201d tells Claude to keep working and includes the reason as guidance for the next turn. A \u201cyes\u201d clears the goal and records an achieved entry in the transcript.
> The evaluator runs on whichever provider your session is configured for. It does not call tools, so it can only judge what Claude has already surfaced in the conversation.
Apparently, it uses Haiku (by default) to evaluate every turn to determine if the goal has been achieved. However, it only relies on the transcript itself (including the reasoning of the main model). It can't independently verify if the goal has been achieved. So, if the main model thinks the goal is or isn't done, how often does Haiku disagree (in a productive way)? That's not clear to me.","created_at":"2026-06-23T18:48:56Z","created_at_i":1782240536,"objectID":"48649559","parent_id":48648900,"story_id":48643180,"story_title":"The Coming Loop","story_url":"https://lucumr.pocoo.org/2026/6/23/the-coming-loop/","updated_at":"2026-06-23T19:22:00Z"},{"_highlightResult":{"author":{"matchLevel":"none","matchedWords":[],"value":"zkmon"},"comment_text":{"fullyHighlighted":false,"matchLevel":"full","matchedWords":["claude","code","goal","mode"],"value":"Next claude news (trump style): Recent versions of Claude code no longer allow talking to other models, or helping with any code that has the goal of moving away from anthropic models."},"story_title":{"fullyHighlighted":false,"matchLevel":"partial","matchedWords":["claude","code"],"value":"DeepClaude \u2013 Claude Code agent loop with DeepSeek V4 Pro"},"story_url":{"matchLevel":"none","matchedWords":[],"value":"https://github.com/aattaran/deepclaude"}},"_tags":["comment","author_zkmon","story_48002136"],"author":"zkmon","comment_text":"Next claude news (trump style): Recent versions of Claude code no longer allow talking to other models, or helping with any code that has the goal of moving away from anthropic models.","created_at":"2026-05-04T10:08:54Z","created_at_i":1777889334,"objectID":"48006658","parent_id":48002136,"story_id":48002136,"story_title":"DeepClaude \u2013 Claude Code agent loop with DeepSeek V4 Pro","story_url":"https://github.com/aattaran/deepclaude","updated_at":"2026-05-05T07:36:06Z"}],"hitsPerPage":20,"nbHits":388,"nbPages":20,"page":0,"params":"query=claude+code+goal+mode&advancedSyntax=true&analyticsTags=backend","processingTimeMS":25,"processingTimingsMS":{"_request":{"roundTrip":16},"afterFetch":{"format":{"highlighting":2,"total":3},"merge":{"mergeLoop":{"prepareNextHit":1,"total":1},"total":1},"total":1},"fetch":{"query":8,"scanning":14,"total":23},"total":25},"query":"claude code goal mode","serverTimeMS":29}