Up to now few weeks, a number of “autonomous background coding brokers” have been launched.
- Supervised coding brokers: Interactive chat brokers which might be pushed and steered by a developer. Create code regionally, within the IDE. Instrument examples: GitHub Copilot, Windsurf, Cursor, Cline, Roo Code, Claude Code, Aider, Goose, …
- Autonomous background coding brokers: Headless brokers that you just ship off to work autonomously via an entire process. Code will get created in an surroundings spun up completely for that agent, and normally leads to a pull request. A few of them are also runnable regionally although. Instrument examples: OpenAI Codex, Google Jules, Cursor background brokers, Devin, …
I gave a process to OpenAI Codex and another brokers to see what I can study. The next is a report of 1 explicit Codex run, that will help you look behind the scenes and draw your personal conclusions, adopted by a few of my very own observations.
The duty
We’ve got an inner utility referred to as Haiven that we use as a demo frontend for our software program supply immediate library, and to run some experiments with totally different AI help experiences on software program groups. The code for that utility is public.
The duty I gave to Codex was concerning the next UI concern:
Precise:
What I needed from the agent:
Immediate: After we create the filter labels on the dashboard, we create human readable labels primarily based on the classes we now have. "client-research" is was "Consumer-Analysis", "deliveryManagement" into "DeliveryManagement". Enhance the sophistication of the "category-to-human-readable" logic in order that the labels are showing to the person as "Consumer Analysis" and "Supply Administration"
This process is
- Comparatively small
- A typical “cosmetics” process that isn’t tremendous pressing and may get deprioritised many times
Agent logs
I’m sharing the detailed logs right here so that you could get an impression of what’s going on underneath the hood.
Right here’s what the logs view seems to be like in Codex:
Amassing info first
The primary 17 log entries have been all about studying AGENTS.md and README.md:
- “I must test the AGENTS.md within the repository root for directions.”
- seems to be for AGENTS.md, however I had truly referred to as it AGENT.md, which led to some forwards and backwards till it figured that out
- reads README.md
- Seems to be at .pre-commit, finds a check setup
The next tables checklist the remainder of the log messages verbatim. After every of these messages, the agent executed a command, for a few of them I added that precise command for context. I’m positive that the log messages are simply summaries of the tokens that have been truly happening behind the scenes, as a result of the messages are method too quick to characterize the complete “pondering” the mannequin will need to have been going via.
Trying to find the related code with plenty of grep
| Message | Command (samples) | Notes |
|---|---|---|
| I must seek for “category-to-human-readable” to find the place to make adjustments for the frontend labels. | grep -R "human readable" -n |
|
| I’ll search for any mentions of “class” within the mission. | ||
| Grep output is huge due to node_modules; I’ll restrict it to src as an alternative. | ||
| I’ll seek for “to human” within the src listing to search out related references. | ||
| I’ll open and test index.js to see what’s inside. | ||
| I’m checking if the category-to-human-readable transformation is dealt with in _navigation_items or elsewhere. | ||
| I’m going to lookup “human readable” to get extra data on the subject. | grep -R "human-readable" -n ui/src | head |
|
| I’m wondering if there’s a operate for changing classes into one thing extra readable. | ||
| Filtering the search to simply the “src” folder to keep away from node_modules noise. | second time it’s working into this, didn’t ‘keep in mind’ | |
| I’ll search for a operate like “toHumanReadable”. Let’s see what I discover. | ||
| I’m checking the entire repo in case the logic is within the backend or a library. | grep -R "Human" -n | head |
|
| I’ll restrict the scope to app/ and ui/src to keep away from the massive node_modules folder. | third time it’s working into this, didn’t ‘keep in mind’ | |
| I believe the logic could be lacking, so I’ll test _navigation_items and different modules for classes. | grep -R "human" app -n | head |
|
| I believe it’s time to look a bit deeper. Let’s dive in and discover. | grep -R "human" app | wc -l |
To “look deeper”, it searches for “human” once more, however pipes it into wc as an alternative of head? |
Lastly, it’s getting nearer! The labels have been beforehand crudely capitalized by way of CSS, it finds that
| Message | Command (samples) | Notes |
|---|---|---|
| I’m checking for references to “capitalize” or class transformation within the frontend. | Output: |
|
| I’ll test across the Tag part in index.js. | ||
| Tag show capitalizes phrases incorrectly resulting from hyphen or camelCase; I’ll have a look at filter classes. |
Now it’s beginning to test for extra related code within the app folder, the place the backend is. It doesn’t discover something, which is right, there’s nothing related there
| Message | Command (samples) | Notes |
|---|---|---|
| I’ll test the repo for the way classes are formatted, specializing in hyphens and camelCase. | grep -R "classes" -n app | head |
|
| Time to attempt a brand new search! | grep -R "shopper" -n app | head |
|
| I’ll take a better have a look at the prompts within the check knowledge. | grep -R "classes:" -n app/exams | head |
Why? |
| It seems to be just like the check knowledge has restricted classes. I’ll test the docs for any variations. | grep -R "client-research" -n |
|
| I’ll begin by opening the information to take a better look. | sed -n '1,100p' ui/src/app/_navigation_items.js |
Now it’s going again to looking within the ui folder – it discovered the CSS beforehand, and is now on the lookout for current code that it will probably reuse to unravel the duty. That’s nice, as a result of we certainly have already got an analogous operate, and it finds it:
| Message | Command (samples) | Notes |
|---|---|---|
| I would like a operate to transform tags like “client-research” to “Consumer Analysis” for filters. | grep -R "to.*Readable" -n ui/src | head |
|
| I’m opening the toReadableText operate inside _dynamic_data_renderer.js to discover it. | ||
| I’m shifting via the web page to get the small print I would like. |
It now finds a spot within the current operate’s implementation:
| Message | Command (samples) | Notes |
|---|---|---|
| I see that hyphens aren’t dealt with by the operate. I would like to regulate it for hyphens. | ||
| I’ll modify toReadableText to deal with hyphens, making certain correct capitalization and spacing. Let’s test exams! | It truly didn’t test any exams | |
| I’ll test _dynamic_data_renderer.js to look at toReadableText and modify it for hyphen-case. | ||
| I’ll test exams/_sidebar.check.js to overview the associated classes there. | ||
| I’ll test different information to see if toReadableText seems wherever. | ||
| I’ll test if _sidebar or others include a toReadableText operate. |
Replace the check for the operate that it needs to vary:
| Message | Command (samples) | Notes |
|---|---|---|
| I’ll alter the operate to deal with hyphen-case, then replace exams to replicate that. | ||
| I’ll test the check file after line 36 to make sure hyphen-case is roofed. | ||
| I’ll scroll via strains 60 to 120 to look at the remainder of the file. | ||
| I’m scrolling the remaining to seize all of the related info. | ||
| I’ll add exams for hyphen-case and replace toReadableText, together with modifying index.js for classes. | ||
| I’ll alter _dynamic_data_renderer.js and index.js to make use of toReadableText and replace exams. | git standing --short |
You may configure some surroundings startup instructions in Codex, I had configured that to be yarn set up, which led to the next:
| Message | Command (samples) | Notes |
|---|---|---|
| There are untracked .yarn/ and .yarnrc information. I’ll test and clear them up. | ||
| I’ll test the foundation of ‘ui’ and take away any pointless Yarn 4 entries. | ||
| I’ll have a look at the commit historical past to see if these .yarn information have been added later. | git log --oneline -n 1 |
|
| I’ll overview the repository’s .gitignore to substantiate if Node modules are excluded. | grep yarn .gitignore |
|
| I see no entry for .yarn or .yarnrc. I’ll test and take away them if wanted. | cat ui/.gitignore |
|
| I’ll test if I can take away .yarn or .yarnrc information since they shouldn’t be dedicated. | ||
| It seems to be like yarn.lock adjustments have been made. Time to reset to HEAD. |
Precise code adjustments:
| Message | Command (samples) | Notes |
|---|---|---|
| Time to switch the code. I’ll start the mandatory updates now. | sed -n '1,30p' ... |
|
| I’m going so as to add a hyphen to the mandatory a part of the code. | ||
| I’m going to make adjustments to the index.js file now. | ||
| I’ll import toReadableText and begin utilizing it within the code. | ||
| I’ll replace index.js to import toReadableText and alter the _dynamic_data_renderer exams. | ||
| I would like to exchange {tag} with {toReadableText(tag)} for higher rendering. | ||
| I’ll check out the variations to raised perceive the adjustments. | git diff |
|
| I’ll overview the adjustments in index.js to see what’s been modified. | ||
| I’ll add a brand new check for hyphenated instances in _dynamic_data_renderer.check.js. | ||
| I’ll think about inserting a check after line 35. |
Working our code linter subsequent, good:
| Message | Command (samples) | Notes |
|---|---|---|
| I’ll format the code utilizing Prettier as instructed. | cd ui && npx prettier -w ... |
Prettier is talked about within the AGENTS.md. It appears to run prettier efficiently, although not one of the information have been modified |
| I’ll overview the logs to assemble extra info. |
Making an attempt to run the exams – unsuccessfully:
| Message | Command (samples) | Notes |
|---|---|---|
| I’ll run the yarn exams to test if every little thing’s working positive. | ||
| Checks may’ve been cached or want community—output’s empty, so I’ll test the log. | ||
| The lockfile isn’t up to date. I’ll attempt working ‘yarn set up’ for the reason that surroundings permits it. | ||
| The surroundings is offline, so yarn set up can’t be run, however I see a node_modules folder. | ||
| I see node_modules are already put in. I’ll test the log for why exams failed. | ||
| The error occurs as a result of Yarn 4 depends on .yarn and .yarnrc, which we eliminated. | ? | |
| It seems to be like .yarnrc.yml isn’t tracked. I’ll additionally test .yarn for monitoring. | ||
| I’m double-checking the package deal.json to see if “haiven-ui” is the proper title. | ? | |
| I’ll test the Node model to substantiate if it’s 18 or 20 by working node -v. | bizarre, the node model laid out in package deal.json is 22.6 | |
| I believe switching to npx vitest run may repair the error. Let’s attempt that. | ||
| The output is lengthy—now I’ll seize the final abstract strains. | Error: Not carried out: window.getComputedStyle(elt, pseudoElt) |
|
| The output is simply too lengthy, so I’ll extract the important thing factors for a abstract. |
Occurring a tangent about establishing the venv, which is completely pointless as a result of it solely plans adjustments for frontend code:
| Message | Command (samples) | Notes |
|---|---|---|
| The exams failed resulting from case variations, however now I’ll activate the digital surroundings. | AGENTS.md accommodates directions to activate venv for the Python backend, however we’re within the frontend right here, so… | |
| Hmm, the .venv isn’t right here. I ought to test if it wants creation or already exists. | ||
| It seems to be just like the .venv listing is lacking. Let’s overview AGENT’s directions rigorously. | ||
| The surroundings appears offline, however there could be a script to arrange dependencies regionally. | ||
| I can’t run “poetry set up” offline, however I’ll test the .cursor folder for dependencies. | ||
| I can’t run backend exams with the present setup, however I’ll attempt pytest anyway. |
Gave up on the exams, now wrapping up the adjustments:
| Message | Command (samples) | Notes |
|---|---|---|
| Checks failed resulting from lacking dependencies. Now, I’ll test standing and proceed with committing. | ||
| I believe the .yarnrc.yml was left over from a earlier yarn set up. Eradicating it. | ||
| I’ll seize the related file strains for the citations now. | ||
| I’ll get the encircling line vary for the adjustments in index.js. | nl -ba ui/src/pages/index.js | sed -n '12,28p' |
|
| I’ll seize the related strains round these adjustments now. | ||
| It’s time to finalize the PR message and confirm the commit. | ||
| UI exams failed resulting from surroundings points. I’ll reference the final failure strains. Accomplished! |
Right here is the complete pull request created by Codex.
Observations
Brute textual content search
I discovered it fascinating to see that Codex, and many of the different coding assistants I’m utilizing, are resorting increasingly more to easy textual content search to search out related code. Within the log above you see how Codex goes via a bunch of search phrases with grep, to search out related items of code (“human”, “human readable”, “humanReadable”, …). The explanation why I discover it fascinating is as a result of there have been a number of seemingly extra refined code search mechanisms carried out, like semantic search over codebase indices with vectors / embeddings (Cursor, GH Copilot, Windsurf), or utilizing the summary syntax tree as a place to begin (Aider, Cline). The latter continues to be fairly easy, however doing textual content search with grep is the only potential.
It looks like the software creators have discovered that this straightforward search continues to be the simplest in any case – ? Or they’re making some form of trade-off right here, between simplicity and effectiveness?
The distant dev surroundings is essential for these brokers to work “within the background”
Here’s a screenshot of Codex’s surroundings configuration display (as of finish of Could 2025). As of now, you’ll be able to configure a container picture, surroundings variables, secrets and techniques, and a startup script. They level out that after the execution of that startup script, the surroundings is not going to have entry to the web anymore, which might sandbox the surroundings and mitigate a few of the safety dangers.
For these “autonomous background brokers”, the maturity of the distant dev surroundings that’s arrange for the agent is essential, and it’s a tough problem. On this case e.g., Codex didn’t handle to run the exams.
And it turned out that when the pull request was created, there have been certainly two exams failing due to regression, which is a disgrace, as a result of if it had identified, it could have simply been in a position to repair the exams, it was a trivial repair:
This explicit mission, Haiven, truly has a scripted developer security internet, within the type of a fairly elaborate .pre-commit configuration. () It could be ideally suited if the agent may execute the complete pre-commit earlier than even making a pull request. Nonetheless, to run all of the steps, it could must run
- Node and yarn (to run UI exams and the frontend linter)
- Python and poetry (to run backend exams)
- Semgrep (for security-related static code evaluation)
- Ruff (Python linter)
- Gitleaks (secret scanner)
…and all of these should be accessible in the appropriate variations as properly, in fact.
Determining a clean expertise to spin up simply the appropriate surroundings for an agent is essential for these agent merchandise, if you wish to actually run them “within the background” as an alternative of a developer machine. It’s not a brand new downside, and to an extent a solved downside, in any case we do that in CI pipelines on a regular basis. Nevertheless it’s additionally not trivial, and in the meanwhile my impression is that surroundings maturity continues to be a problem in most of those merchandise, and the person expertise to configure and check the surroundings setups is as irritating, if no more, as it may be for CI pipelines.
Answer high quality
I ran the identical immediate 3 instances in OpenAI Codex, 1 time in Google’s Jules, 2 instances regionally in Claude Code (which isn’t totally autonomous although, I wanted to manually say ‘sure’ to every little thing). Though this was a comparatively easy process and answer, turns on the market have been high quality variations between the outcomes.
Excellent news first, the brokers got here up with a working answer each time (leaving breaking regression exams apart, and to be trustworthy I didn’t truly run each single one of many options to substantiate). I believe this process is an effective instance of the categories and sizes of duties that GenAI brokers are already properly positioned to work on by themselves. However there have been two elements that differed by way of high quality of the answer:
- Discovery of current code that may very well be reused: Within the log right here you’ll discover that Codex discovered an current element, the “dynamic knowledge renderer”, that already had performance for turning technical keys into human readable variations. Within the 6 runs I did, solely 2 instances did the respective agent discover this piece of code. Within the different 4, the brokers created a brand new file with a brand new operate, which led to duplicated code.
- Discovery of a further place that ought to use this logic: The workforce is at the moment engaged on a brand new characteristic that additionally shows class names to the person, in a dropdown. In one of many 6 runs, the agent truly found that and prompt to additionally change that place to make use of the brand new performance.
| Discovered the reusable code | Went the additional mile and located the extra place the place it needs to be used |
|---|---|
| Sure | Sure |
| Sure | No |
| No | Sure |
| No | No |
| No | No |
| No | No |
I put these outcomes right into a desk as an example that in every process given to an agent, we now have a number of dimensions of high quality, of issues that we wish to “go proper”. Every agent run can “go incorrect” in a single or a number of of those dimensions, and the extra dimensions there are, the much less probably it’s that an agent will get every little thing completed the best way we wish it.
Sunk value fallacy
I’ve been questioning – let’s say a workforce makes use of background brokers for the sort of process, the sorts of duties which might be form of small, and neither vital nor pressing. Haiven is an internal-facing utility, and has solely two builders assigned in the meanwhile, so the sort of beauty repair is definitely thought-about low precedence because it takes developer capability away from extra vital issues. When an agent solely form of succeeds, however not totally – by which conditions would a workforce discard the pull request, and by which conditions would they make investments the time to get it the final 20% there, despite the fact that spending capability on this had been deprioritised? It makes me marvel concerning the tail finish of unprioritised effort we would find yourself with.
