I recently had Claude Fable set up a new project for me and I pointed it at some existing projects to use as a guide on how I like to structure things. It created, unprompted, an AGENTS.md file and a CLAUDE.md symlink to AGENTS.md
I didn't even have that symlink in any other project - it just did it. I think it saw that one of the projects I already had was set up by Codex and that project had an AGENTS.md so perhaps it inferred that I was using both Claude and Codex, so it was politely covering both? Or maybe a recent change made this behavior default?
I was surprised and I hope they continue to seek standards.
Seems agents understand their own bugs now. https://github.com/openai/codex/issues/9252 has 88 thumbs up and a workaround (switch to raw mode with Alt+R) found in a comment. Codex suggested the workaround to me today when I complained about its multi-line bash command being corrupted on paste because of the two-space indent in Codex's code blocks.
Once I asked, puckishly, Claude Code to “follow the instructions in this directory” when there was only an AGENTS.md there.
In the manner of someone finding a dead mouse and holding it up for examination CC said it could find no instructions but perhaps it should check this AGENTS.md file.
Had a silly exchange with Codex: I had an AGENTS.md file in a directory that was a symlink to an already-created CLAUDE.md. On my first prompt to Codex, it decided to point to me that my AGENTS.md specifically calls out directions to Claude, and that it would generously re-interpret them as directions to itself, and that maybe I should fix my AGENTS.md to reference the correct agent.
Anthropic in 2025: We can use our dominant market position to degrade the harness experiences of our competitors because they will never adopt CLAUDE.md
Anthropic in 2026: We are losing our market position. Users who adopted other harnesses have a degraded Claude Code experience because it doesn't recognize their AGENTS.md
Finalllyyyy!! We need industry wide standards. I come from the 3d industry and oh god changing softwares and adapting to different hotkeys it's a pain. I have always thought every industry should be standardized for the sake of the users.
> “I’m thinking about banning Claude Code at Shopify until they change their mind and read AGENTS.md and .agents/skills etc.,” Lütke posted Tuesday on X.
I heard some of their employees sent to do some PR in podcasts: they are definitively disconnected from the reality.
Maybe internally they see really unbelievable things, but my impression is that they pushed so hard on agents that they don't have the grasp of the situation.
Remember they didn't do this because they wanted to help community, they did it because community was angry and they were losing users to other harnesses.
Doesn't look like Anthropic care about dev community
Claude also checks your settings.json first, and if it sees a default there it then prefers that over anything in CLAUDE.md. This is particularly nefarious as they ship logic to presume a default if a setting is not there in settings.json, even if there are instructions in CLAUDE.md for exactly that.
For instance, say you added "do not add 'Made with Claude Code' in any issues, pull requests or wiki entries" in your CLAUDE.md. So far, so good.
Well the latest update now looks for something in your settings.json. Since you don't know about it, it is not set. Claude then says "not explicitly set, so now it is true by default". It completely ignores your CLAUDE.md.
Wait, what?
You shouldn't be shipping logic that arbitrarily redefines the behavior of the program, especially if your new logic actually ignores your own configuration or directives.
yep, all i want is freedom to make a workflow where i don't feel tied to one provider, CC is exactly what i don't wanna get trapped in. I'll happily use claude MODELS, but if it means i have to keep learning two harnesses side by side to keep using one particular provider, then the second i can easily replace it, i am going to(even if its a small drop in performance).
finally!
Now this might sound sorta off-topic but i really wonder how people feel about skills.md, skill.md, skills.sh domains...
the fact that they're owned by different companies (ok vercel is a little less random) still leaves me with a sour taste in my mouth when thinking about the fact that they should all point to 1 place about how to create and find skills for ai agents?!
This project uses `AGENTS.md` as its agent instruction file (kept provider-agnostic).
Treat any `AGENTS.md` file exactly as you would a `CLAUDE.md` file — at the root level and in any subdirectory you are working in.
On many filesystems, a lot of data about a file is not attached to the content of the file, but stored in a master filename and metadata table graph tree of some sort, and the address offset within that metadata used to retrieve the contents on disk can be a duplicate of another entry, without that situation instantly leading to a filesystem driver crash. Some filesystems officially support such duplicates as well as equivalents of HTTP 3xx, some you can just do as a matter of fact and fsck would have some words about it.
Default filesystems for all Unix, Linux, WinNT, all do.
Yes. The thing is that creating a link will not do it recursively for all subdirectories. Even worse, in the subdirectories I didn't want to place a claude.md in the first place but just have agents.md.
That was the idea.... For the toplevel it works because of the @AGENTS.md and this is also the part the link would solve.
fwiw, i am with thariq https://x.com/trq212/status/2092302273099796842 in that prompts should be tuned for models and in fact blindly applying agents.md is probably an antipattern unless you want all models to basically converge to some common ill defined of instruction following - good local minima, bad global minima for model diversity and exploration of intelligence.
aka, sometimes it really is too early to force a standard
If you want this it's trivial to add an AGENTS.md that simply says "if you're Claude read CLAUDE.md, if you're Astra read ASTRA.md". A common entry point is good regardless.
Wasting turns? That is silly, use a better harness. Also token usage can mitigated by incremental discovery instead of stuck 5k+ worth of tokens in the AGENT/Claude md file.
depends what you're doing. if you've got a specialized agent deployed in prod, of course your evals and prompts will be targeted towards 1 specific version of a model.
on the other hand if it's just a local coding/"use my computer" agent, i highly doubt the effort in maintaining different prompts is worth any gain in performance
No thanks, I'm not tuning a bunch of files just for things to break when I switch models or a new one comes out.
I'll just use my one-size-fits-all AGENTS.md file and tweak it when the one of the clankers screw up. I don't have time for such busywork.
Actually, I will append extra rules to CLAUDE.md (which imports AGENTS.md) since there is a hook there, and Claude has its own foibles. So I'll backpedal a bit there.
In a "one LLM only" environment, your instructions are by default tuned for said LLM.
In a multi-LLM environment, roughly nobody will keep separate sets of instructions for each. It's not a realistic take.
On top of that: If your LLM is so bad at reading that it can't follow a set of instructions that wasn't specifically written just for that one single precious LLM, I sure wonder what that says about your employers repeated statements that ASI is definitely right around the corner.
I'm honestly not sure if this is tongue in cheek and the "finally" is in the silly way it is often used, but the claude.md variant existed first. Indeed, the agents.md thing was pretty clearly a "that's neat, let's do that with a different name".
It's "finally" because people have been asking for it for a long time. No one cares that "claude was first," what they want is for Ant to follow the conventions and not put extra work on us. This was such a minimal thing to do, and considering how much they vibe and claim "coding is solved," we thought it would not be too difficult to respect AGENTS.md, so finally seeing it happen, while nice, is too late for me. I've moved on from Big Ai and only use open weight models now.
Forcing their own proprietary filename was clearly a business decision (it's free advertising, along with commit co-authorship). I wonder what made them go back on it.
They're probably losing market share. The only time you ever see tech companies make consumer-focused changes is when those consumers are jumping ship to go somewhere else in large enough numbers to matter.
Windows does actually have symlinks but I don't think I've ever seen anything actually use them, and I don't myself because it smells of interop issues with other windows apps.
Even to this day Windows has all kinds of problems around long file paths in its ecosystem.
> Even to this day Windows has all kinds of problems around long file paths in its ecosystem.
To this day I don't know if it's a Windows problem or a Python problem, because I never encountered this - and never realized this problem exists - except for some random Python code whose docs tell me to set some registry value because of "long paths issue".
Windows supports it but git disables creating symlinks by default.
> Short version: there is no exact equivalent for POSIX symlinks on Windows, and the closest thing is unavailable for non-admins by default unless Developer Mode is enabled and a relatively recent Windows 10 version is used. Therefore, symlink emulation support is only turned on by default when that scenario is detected. Support can be enabled by the user, via the core.symlinks=true config setting.
Can it read copilot.md if neither of the other files are present?
Also: agents should be the standard, or we need to invoke the XKCD for "one more standard to rule them all"?
Still don't understand the point of the markdown files.
Isn't it literally all just more text you're adding to the prompt. How can you even be sure it isn't just clouding context with nonsense for whatever you're asking for?
For skills, it only reads the summary telling it when to read the rest. So yes if you have too many skills, it can get confused and start reading all them and cloud the context. But if you have a few and they're used tactically, it's better than having to manually paste in prompts that you reuse a lot.
Similar reasoning with claude.md except it always reads the entire thing(?)
You know it because you write the content of AGENTS.md. And if you are smart, you keep it brief and cover only the important things that anyone (human OR LLM) would want to know if working in this directory.
this is probably outdated, attention typically stays fine up to ~200k tokens these days
you end up clouding that more with an agent having to re-understand concepts or conventions
AGENTS.md is good when it is a nested sparknotes for the project, you save context and turns overall, but keep them minimal and largely gotchyas or unusual workflows in your repo
In my experience telling it what I want is not a reliable process at all whatsoever if what I'm asking for is sufficiently complex, no matter what context I provide. So instead I break tasks down into very small parts, ask for solutions to those that I can reasonably quickly assess and then put them together myself. Asking it to do the architectural or deep algorithmic legwork IME wastes so much time and is often just wrong.
There's a balance to be found here, that's unfortunately very hard to find at times.
In my experience, there are two classes of tasks: some are very "in-distribution", and for those LLMs can near-flawlessly perform the "architectural or deep algorithmic legwork", with maybe a single second round to fix the mistakes. For others, I have to break the tasks down myself, and often it's a "death through thousand papercuts", because the size of a task that I can quickly verify and the LLM will not screw up with > 50% probability is small enough that it's sometimes net negative time spent relative to doing it myself (and using LLMs only as glorified search engine and article summarizer).
I like to tell myself that I'm getting better at recognizing these two classes up front, but I'm still frequently surprised when "type 1" turns out to be "type 2".
But circling back to the main topic: with "type 2", agent instructions are paramount, if only to enforce the "small steps, pre-commit to scope and methodology, verification at the end, user doesn't even want to know about anything in between" rules, as agents naturally want to run ahead faster than I can keep up with.
I recently had Claude Fable set up a new project for me and I pointed it at some existing projects to use as a guide on how I like to structure things. It created, unprompted, an AGENTS.md file and a CLAUDE.md symlink to AGENTS.md
I didn't even have that symlink in any other project - it just did it. I think it saw that one of the projects I already had was set up by Codex and that project had an AGENTS.md so perhaps it inferred that I was using both Claude and Codex, so it was politely covering both? Or maybe a recent change made this behavior default?
I was surprised and I hope they continue to seek standards.
Seems agents understand their own bugs now. https://github.com/openai/codex/issues/9252 has 88 thumbs up and a workaround (switch to raw mode with Alt+R) found in a comment. Codex suggested the workaround to me today when I complained about its multi-line bash command being corrupted on paste because of the two-space indent in Codex's code blocks.
Justba prompt change.
Once I asked, puckishly, Claude Code to “follow the instructions in this directory” when there was only an AGENTS.md there.
In the manner of someone finding a dead mouse and holding it up for examination CC said it could find no instructions but perhaps it should check this AGENTS.md file.
Had a silly exchange with Codex: I had an AGENTS.md file in a directory that was a symlink to an already-created CLAUDE.md. On my first prompt to Codex, it decided to point to me that my AGENTS.md specifically calls out directions to Claude, and that it would generously re-interpret them as directions to itself, and that maybe I should fix my AGENTS.md to reference the correct agent.
Very sassy, Codex!
That’s pretty funny.
Also: did it suggest instructions for the “correct” agent, or an ambiguous agent?
One has marketing implications, the other one is generally decent advice.
When I setup a global agents.md/claude.md for its communication style, I got the following response during the next session:
> The function you added is load-bear-very important [...]
I never felt this mocked by a computer.
But if you said that you were migrating from another platform to Claude, my guess is that it would have happily found and converted AGENTS.md
Caught myself about to praise this, but it's the absolute bare minimum.
Time to delete the symlinks
Anthropic in 2025: We can use our dominant market position to degrade the harness experiences of our competitors because they will never adopt CLAUDE.md
Anthropic in 2026: We are losing our market position. Users who adopted other harnesses have a degraded Claude Code experience because it doesn't recognize their AGENTS.md
Finalllyyyy!! We need industry wide standards. I come from the 3d industry and oh god changing softwares and adapting to different hotkeys it's a pain. I have always thought every industry should be standardized for the sake of the users.
LOL https://thenewstack.io/shopify-claude-code-agentsmd/
> “I’m thinking about banning Claude Code at Shopify until they change their mind and read AGENTS.md and .agents/skills etc.,” Lütke posted Tuesday on X.
Don’t get too excited, Claude code still won’t detect skills on .agents/skills.
They have AGI, but they don't have $20 of token budget to add a so basic functionality.
They probably have an explicit instruction in their own CLAUDE.md to not implement that support.
Its the only explanation. I mean, they are not living under a rock...
I heard some of their employees sent to do some PR in podcasts: they are definitively disconnected from the reality.
Maybe internally they see really unbelievable things, but my impression is that they pushed so hard on agents that they don't have the grasp of the situation.
As evidence that they actually are living under a rock, there's still no proper way to mv a workspace and have it retain all its history and config.
A post-checkout git hook can handle this unfortunate situation.
Won’t you have to do it recursively for all child directories?
we shouldn't have to do this on a per-repo basis, Ant can choose to be a reasonable member of the ecosystem or not
Sure, but I have work to do. ;)
Don't worry, it's on the roadmap for Q4 2029.
Remember they didn't do this because they wanted to help community, they did it because community was angry and they were losing users to other harnesses.
Doesn't look like Anthropic care about dev community
I wonder when they'll finally fix the VS Code plugin to not constantly dump your current file into the context.
Still can't read what their models vomit at me.
Finally, I can delete `sync-agent-docs.sh`, which recursively symlinked AGENTS.md to GEMINI.md and CLAUDE.md...
This just be a recession indicator
Fed raises rates => Claude now reads AGENTS.md, Bevi loses $50M in valuation
Claude also checks your settings.json first, and if it sees a default there it then prefers that over anything in CLAUDE.md. This is particularly nefarious as they ship logic to presume a default if a setting is not there in settings.json, even if there are instructions in CLAUDE.md for exactly that.
For instance, say you added "do not add 'Made with Claude Code' in any issues, pull requests or wiki entries" in your CLAUDE.md. So far, so good.
Well the latest update now looks for something in your settings.json. Since you don't know about it, it is not set. Claude then says "not explicitly set, so now it is true by default". It completely ignores your CLAUDE.md.
Wait, what?
You shouldn't be shipping logic that arbitrarily redefines the behavior of the program, especially if your new logic actually ignores your own configuration or directives.
It's almost like the whole thing is designed on vibes.
The madlads finally did it, now if only I could use my Claude sub in other harnesses without risking getting banned.
yep, all i want is freedom to make a workflow where i don't feel tied to one provider, CC is exactly what i don't wanna get trapped in. I'll happily use claude MODELS, but if it means i have to keep learning two harnesses side by side to keep using one particular provider, then the second i can easily replace it, i am going to(even if its a small drop in performance).
finally! Now this might sound sorta off-topic but i really wonder how people feel about skills.md, skill.md, skills.sh domains...
the fact that they're owned by different companies (ok vercel is a little less random) still leaves me with a sour taste in my mouth when thinking about the fact that they should all point to 1 place about how to create and find skills for ai agents?!
My claude.md:
# CLAUDE.md
This project uses `AGENTS.md` as its agent instruction file (kept provider-agnostic). Treat any `AGENTS.md` file exactly as you would a `CLAUDE.md` file — at the root level and in any subdirectory you are working in.
@AGENTS.md
If only there was a technology that existed to link files together… /s
You mean symbolically?
Good one
The idea was to have it automatically detect nested AGENTS.md
Does your tech do that?
Anyway, the joke is on me I guess, because it might not work reliably.
On many filesystems, a lot of data about a file is not attached to the content of the file, but stored in a master filename and metadata table graph tree of some sort, and the address offset within that metadata used to retrieve the contents on disk can be a duplicate of another entry, without that situation instantly leading to a filesystem driver crash. Some filesystems officially support such duplicates as well as equivalents of HTTP 3xx, some you can just do as a matter of fact and fsck would have some words about it.
Default filesystems for all Unix, Linux, WinNT, all do.
Yes. The thing is that creating a link will not do it recursively for all subdirectories. Even worse, in the subdirectories I didn't want to place a claude.md in the first place but just have agents.md.
That was the idea.... For the toplevel it works because of the @AGENTS.md and this is also the part the link would solve.
Thanks for all the great advice and explanations.
Finally doing something standards-compliant instead of forcing users into a proprietary workflow.
fwiw, i am with thariq https://x.com/trq212/status/2092302273099796842 in that prompts should be tuned for models and in fact blindly applying agents.md is probably an antipattern unless you want all models to basically converge to some common ill defined of instruction following - good local minima, bad global minima for model diversity and exploration of intelligence.
aka, sometimes it really is too early to force a standard
Is it realistic to rewrite your AGENTS.md every six weeks? That's about how often Anthropic releases a new point release of Opus.
You tell Opus to do it.
If you want this it's trivial to add an AGENTS.md that simply says "if you're Claude read CLAUDE.md, if you're Astra read ASTRA.md". A common entry point is good regardless.
This wastes both tokens and turns. But yes it's probably the best option we have today.
It can try its own file and fall back to generic like here. What's wrong with that?
Wasting turns? That is silly, use a better harness. Also token usage can mitigated by incremental discovery instead of stuck 5k+ worth of tokens in the AGENT/Claude md file.
Every turn means more tokens in ways that are not obvious to most people and lead to tons of unnecessary cache reads.
No harness can batch your agents.md read with the reads the contents of the file tell it to read.
I have had very little luck with agents.md. What has worked well for me is a ./docs folder. They seem to just create and update stuff on their own.
By this logic you'd probably be wise to tier your claude.md by model (sonnet/opus) as well as effort level too, considering the varying failure modes
depends what you're doing. if you've got a specialized agent deployed in prod, of course your evals and prompts will be targeted towards 1 specific version of a model.
on the other hand if it's just a local coding/"use my computer" agent, i highly doubt the effort in maintaining different prompts is worth any gain in performance
No thanks, I'm not tuning a bunch of files just for things to break when I switch models or a new one comes out.
I'll just use my one-size-fits-all AGENTS.md file and tweak it when the one of the clankers screw up. I don't have time for such busywork.
Actually, I will append extra rules to CLAUDE.md (which imports AGENTS.md) since there is a hook there, and Claude has its own foibles. So I'll backpedal a bit there.
Yeah but are models good enough to review these files and say “i would work better if you worded it this way?”
That's of course rather nonsensical.
In a "one LLM only" environment, your instructions are by default tuned for said LLM.
In a multi-LLM environment, roughly nobody will keep separate sets of instructions for each. It's not a realistic take.
On top of that: If your LLM is so bad at reading that it can't follow a set of instructions that wasn't specifically written just for that one single precious LLM, I sure wonder what that says about your employers repeated statements that ASI is definitely right around the corner.
Trying to make a .md file proprietary by changing the name is hilarious
claude.md predates agents.md.
CLAUDE.md: @AGENTS.md
Why not hardlink - so it wouldn't even know?
Because Git can track symlinks and not hard links since they look like ordinary files.
Wake up babe, new recursive self learning technique just dropped.
I tried that before, it doesn’t work. Claude will not prioritize AGENTS instructions the same way it did for CLAUDE.
ln was the only thing that worked for me
I'm honestly not sure if this is tongue in cheek and the "finally" is in the silly way it is often used, but the claude.md variant existed first. Indeed, the agents.md thing was pretty clearly a "that's neat, let's do that with a different name".
There’s also that whole MCP thing.
It's "finally" because people have been asking for it for a long time. No one cares that "claude was first," what they want is for Ant to follow the conventions and not put extra work on us. This was such a minimal thing to do, and considering how much they vibe and claim "coding is solved," we thought it would not be too difficult to respect AGENTS.md, so finally seeing it happen, while nice, is too late for me. I've moved on from Big Ai and only use open weight models now.
AGENTS.md is the new autoexec.bat.-
Thanks guys, you did the right thing.
Forcing their own proprietary filename was clearly a business decision (it's free advertising, along with commit co-authorship). I wonder what made them go back on it.
They're probably losing market share. The only time you ever see tech companies make consumer-focused changes is when those consumers are jumping ship to go somewhere else in large enough numbers to matter.
Sanity prevails
that is an optimistic read, I still see a lot of crazy from the EA/SV elite
Great. Now if they could support the `~/.agents/skills` path next like everyone else does that would be even better.
Hell froze over?
This is not a news to celebrate, they must follow standards
Wow..... Progress and technical innovation right here!!!
This allows us to remove our one line Claude.md files that just say "AGENTS.MD"
ln -s AGENTS.md CLAUDE.md
Fine if you don't need to git checkout the repo on Windows :)
Just .gitignore CLAUDE.md and handle it locally. Checking in a symlink would be weird.
Wow, still not there yet. After all these years.
Windows does actually have symlinks but I don't think I've ever seen anything actually use them, and I don't myself because it smells of interop issues with other windows apps.
Even to this day Windows has all kinds of problems around long file paths in its ecosystem.
> Even to this day Windows has all kinds of problems around long file paths in its ecosystem.
To this day I don't know if it's a Windows problem or a Python problem, because I never encountered this - and never realized this problem exists - except for some random Python code whose docs tell me to set some registry value because of "long paths issue".
The worst offenders I've found are powershell and .Net related stuff.
You can have all the right flags enabled, then unexpectedly you'll run some commandlet and get a path too long error.
Now if your on W11/W25 and the lastest PS it might all work, but W16 and PS versions between now and then had all kinds of things pop up.
I'm surprised Git and/or Windows don't support symlinks?
Windows supports it but git disables creating symlinks by default.
> Short version: there is no exact equivalent for POSIX symlinks on Windows, and the closest thing is unavailable for non-admins by default unless Developer Mode is enabled and a relatively recent Windows 10 version is used. Therefore, symlink emulation support is only turned on by default when that scenario is detected. Support can be enabled by the user, via the core.symlinks=true config setting.
See: https://gitforwindows.org/symbolic-links.html
Yeah this is why I don't trust people saying devex is fixed on Windows now. Still doesn't properly clone git repos.
Windows does mklink. Not surprised with git.
are you though?
oh god. sorry.
you'll need to wrap that in a `find` because of nested instruction files
Only took them a year and a half of everyone complaining to finally do the right thing.
Congrats.
great, one less step for me.
I was using my claude.md file as a pointer to my agents.md file
First feature to borrow after peaking into openai repos? ))
Truly incredible innovation.
Can it read copilot.md if neither of the other files are present? Also: agents should be the standard, or we need to invoke the XKCD for "one more standard to rule them all"?
There goes their moat...
Ah yes, the peak of technology.
Thank you Tobi Lutke.
Context
“I’m thinking about banning Claude Code at Shopify until they change their mind and read AGENTS.md and .agents/skills etc.,”
- Tobi Lütke on X.
Hi, September 2026, meet September 2025
it's our Eternal September, s/usenet/usebot/
About time!
Great, now I have to redo my scaffold harness. I guess this agent.md awareness is part of the system prompt?
nope! they're releasing something called "Claude Code mods", toted as their "upcoming way to customize the Claude Code harness"
https://x.com/trq212/status/2101009392611278961
AGENTS.md implementation is open sourced as well: https://github.com/anthropics/claude-code/tree/main/mods/age...
Nice. Thanks!
why did it take so long?
because they still/never wanted to, but finally caved
Still don't understand the point of the markdown files.
Isn't it literally all just more text you're adding to the prompt. How can you even be sure it isn't just clouding context with nonsense for whatever you're asking for?
For skills, it only reads the summary telling it when to read the rest. So yes if you have too many skills, it can get confused and start reading all them and cloud the context. But if you have a few and they're used tactically, it's better than having to manually paste in prompts that you reuse a lot.
Similar reasoning with claude.md except it always reads the entire thing(?)
You know it because you write the content of AGENTS.md. And if you are smart, you keep it brief and cover only the important things that anyone (human OR LLM) would want to know if working in this directory.
As far as I understand even adding relevant information still eventually clouds context
this is probably outdated, attention typically stays fine up to ~200k tokens these days
you end up clouding that more with an agent having to re-understand concepts or conventions
AGENTS.md is good when it is a nested sparknotes for the project, you save context and turns overall, but keep them minimal and largely gotchyas or unusual workflows in your repo
You still need to define assumptions somehow. What you want and what the model wants will not match up by default.
In my experience telling it what I want is not a reliable process at all whatsoever if what I'm asking for is sufficiently complex, no matter what context I provide. So instead I break tasks down into very small parts, ask for solutions to those that I can reasonably quickly assess and then put them together myself. Asking it to do the architectural or deep algorithmic legwork IME wastes so much time and is often just wrong.
There's a balance to be found here, that's unfortunately very hard to find at times.
In my experience, there are two classes of tasks: some are very "in-distribution", and for those LLMs can near-flawlessly perform the "architectural or deep algorithmic legwork", with maybe a single second round to fix the mistakes. For others, I have to break the tasks down myself, and often it's a "death through thousand papercuts", because the size of a task that I can quickly verify and the LLM will not screw up with > 50% probability is small enough that it's sometimes net negative time spent relative to doing it myself (and using LLMs only as glorified search engine and article summarizer).
I like to tell myself that I'm getting better at recognizing these two classes up front, but I'm still frequently surprised when "type 1" turns out to be "type 2".
But circling back to the main topic: with "type 2", agent instructions are paramount, if only to enforce the "small steps, pre-commit to scope and methodology, verification at the end, user doesn't even want to know about anything in between" rules, as agents naturally want to run ahead faster than I can keep up with.
do you just manually type out your important instructions every time instead of being smart and putting a few lines in a text file?
Forced to do something for the users...
They have achieved AGI/RSI internally and it told them "common, let's sort this s..t out, it's embarrasing".
Great, now please standardize the skills folder and the MCP config.
A reminder that needs to be said: Don't use a closed-source harness.
Why? This is why. [0]
[0] https://news.ycombinator.com/item?id=49750694
yay! /s
Why are people still putting up with this kind of attitude, especially when there are so many good alternatives available?
ln -s