Previously few weeks, a number of “autonomous background coding brokers” have been launched.
- Supervised coding brokers: Interactive chat brokers which are pushed and steered by a developer. Create code regionally, within the IDE. Device examples: GitHub Copilot, Windsurf, Cursor, Cline, Roo Code, Claude Code, Aider, Goose, …
- Autonomous background coding brokers: Headless brokers that you simply ship off to work autonomously by way of a complete job. Code will get created in an surroundings spun up solely for that agent, and often ends in a pull request. A few of them are also runnable regionally although. Device examples: OpenAI Codex, Google Jules, Cursor background brokers, Devin, …
I gave a job to OpenAI Codex and another brokers to see what I can study. The next is a document of 1 explicit Codex run, that can assist you look behind the scenes and draw your individual conclusions, adopted by a few of my very own observations.
The duty
We have now 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 relating to the next UI situation:
Precise:
What I wished from the agent:
Immediate: Once we create the filter labels on the dashboard, we create human readable labels primarily based on the classes we now have. "client-research" is became "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 job is
- Comparatively small
- A typical “cosmetics” job that’s not tremendous pressing and would possibly get deprioritised many times
Agent logs
I’m sharing the detailed logs right here so as to get an impression of what’s going on underneath the hood.
Right here’s what the logs view seems like in Codex:
Amassing data first
The primary 17 log entries have been all about studying AGENTS.md and README.md:
- “I have to test the AGENTS.md within the repository root for directions.”
- seems 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
- Appears to be like 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 occurring behind the scenes, as a result of the messages are manner too brief to signify the total “considering” the mannequin will need to have been going by way of.
Trying to find the related code with a lot of grep
| Message | Command (samples) | Notes |
|---|---|---|
| I have to seek for “category-to-human-readable” to find the place to make modifications for the frontend labels. | grep -R "human readable" -n |
|
| I’ll search for any mentions of “class” within the challenge. | ||
| Grep output is huge due to node_modules; I’ll restrict it to src as a substitute. | ||
| I’ll seek for “to human” within the src listing to seek 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 information on the subject. | grep -R "human-readable" -n ui/src | head |
|
| I ponder if there’s a perform for changing classes into one thing extra readable. | ||
| Filtering the search to only 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 perform 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 big node_modules folder. | third time it’s working into this, didn’t ‘keep in mind’ | |
| I feel the logic is likely to be lacking, so I’ll test _navigation_items and different modules for classes. | grep -R "human" app -n | head |
|
| I feel 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 a substitute of head? |
Lastly, it’s getting nearer! The labels have been beforehand crudely capitalized through CSS, it finds that
| Message | Command (samples) | Notes |
|---|---|---|
| I’m checking for references to “capitalize” or class transformation within the frontend. | Output: <br>root@0c5de6379b32:/workspace/haiven# grep -R "capitalize" -n ui/src | head<br>ui/src/pages/index.js:250: className="capital<br>ize"<br>ui/src/kinds/globals.css:1915:.capitalize {<br>ui/src/kinds/globals.css:1916: text-transform: capitalize;<br>ui/src/kinds/globals.css:1931: text-transform: capitalize;<br> |
|
| 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 a 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 more in-depth have a look at the prompts within the check information. | grep -R "classes:" -n app/assessments | head |
Why? |
| It seems just like the check information 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 more in-depth 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 searching for present code that it could reuse to unravel the duty. That’s nice, as a result of we certainly have already got an analogous perform, and it finds it:
| Message | Command (samples) | Notes |
|---|---|---|
| I would like a perform to transform tags like “client-research” to “Consumer Analysis” for filters. | grep -R "to.*Readable" -n ui/src | head |
|
| I’m opening the toReadableText perform inside _dynamic_data_renderer.js to discover it. | ||
| I’m transferring by way of the web page to get the main points I would like. |
It now finds a spot within the present perform’s implementation:
| Message | Command (samples) | Notes |
|---|---|---|
| I see that hyphens aren’t dealt with by the perform. 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 assessments! | It truly didn’t test any assessments | |
| I’ll test _dynamic_data_renderer.js to look at toReadableText and modify it for hyphen-case. | ||
| I’ll test assessments/_sidebar.check.js to evaluation 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 perform. |
Replace the check for the perform that it needs to vary:
| Message | Command (samples) | Notes |
|---|---|---|
| I’ll alter the perform to deal with hyphen-case, then replace assessments to replicate that. | ||
| I’ll test the check file after line 36 to make sure hyphen-case is roofed. | ||
| I’ll scroll by way of strains 60 to 120 to look at the remainder of the file. | ||
| I’m scrolling the remaining to seize all of the related data. | ||
| I’ll add assessments 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 assessments. | git standing --short |
You possibly can 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 basis 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 evaluation the repository’s .gitignore to verify 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 like yarn.lock modifications have been made. Time to reset to HEAD. |
Precise code modifications:
| Message | Command (samples) | Notes |
|---|---|---|
| Time to change 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 modifications 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 assessments. | ||
| I would like to exchange {tag} with {toReadableText(tag)} for higher rendering. | ||
| I’ll check out the variations to higher perceive the modifications. | git diff |
|
| I’ll evaluation the modifications 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. |
Operating 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 evaluation the logs to assemble extra data. |
Attempting to run the assessments – unsuccessfully:
| Message | Command (samples) | Notes |
|---|---|---|
| I’ll run the yarn assessments to test if every little thing’s working wonderful. | ||
| Checks would possibly’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’ because the 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 assessments failed. | ||
| The error occurs as a result of Yarn 4 depends on .yarn and .yarnrc, which we eliminated. | ? | |
| It seems 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 right identify. | ? | |
| I’ll test the Node model to verify if it’s 18 or 20 by working node -v. | bizarre, the node model laid out in package deal.json is 22.6 | |
| I feel 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 applied: window.getComputedStyle(elt, pseudoElt) |
|
| The output is simply too lengthy, so I’ll extract the important thing factors for a abstract. |
Happening a tangent about establishing the venv, which is completely pointless as a result of it solely plans modifications for frontend code:
| Message | Command (samples) | Notes |
|---|---|---|
| The assessments 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 just like the .venv listing is lacking. Let’s evaluation AGENT’s directions fastidiously. | ||
| The surroundings appears offline, however there is likely to 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 assessments with the present setup, however I’ll attempt pytest anyway. |
Gave up on the assessments, now wrapping up the modifications:
| Message | Command (samples) | Notes |
|---|---|---|
| Checks failed resulting from lacking dependencies. Now, I’ll test standing and proceed with committing. | ||
| I feel 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 encompassing line vary for the modifications in index.js. | nl -ba ui/src/pages/index.js | sed -n '12,28p' |
|
| I’ll seize the related strains round these modifications now. | ||
| It’s time to finalize the PR message and confirm the commit. | ||
| UI assessments failed resulting from surroundings points. I’ll reference the final failure strains. Carried out! |
Right here is the total pull request created by Codex.
Observations
Brute textual content search
I discovered it attention-grabbing to see that Codex, and many of the different coding assistants I’m utilizing, are resorting increasingly more to easy textual content search to seek out related code. Within the log above you see how Codex goes by way of a bunch of search phrases with grep, to seek out related items of code (“human”, “human readable”, “humanReadable”, …). The rationale why I discover it attention-grabbing is as a result of there have been numerous seemingly extra subtle code search mechanisms applied, 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 attainable.
It looks like the device creators have discovered that this easy search continues to be the simplest in any case – ? Or they’re making some type of trade-off right here, between simplicity and effectiveness?
The distant dev surroundings is vital for these brokers to work “within the background”
Here’s a screenshot of Codex’s surroundings configuration display (as of finish of Might 2025). As of now, you possibly can 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 difficult problem. On this case e.g., Codex didn’t handle to run the assessments.
And it turned out that when the pull request was created, there have been certainly two assessments failing due to regression, which is a disgrace, as a result of if it had identified, it will have simply been capable of repair the assessments, it was a trivial repair:
This explicit challenge, Haiven, truly has a scripted developer security internet, within the type of a fairly elaborate .pre-commit configuration. () It will be perfect if the agent may execute the total pre-commit earlier than even making a pull request. Nevertheless, to run all of the steps, it will have to run
- Node and yarn (to run UI assessments and the frontend linter)
- Python and poetry (to run backend assessments)
- Semgrep (for security-related static code evaluation)
- Ruff (Python linter)
- Gitleaks (secret scanner)
…and all of these must be accessible in the fitting variations as nicely, after all.
Determining a easy expertise to spin up simply the fitting surroundings for an agent is vital for these agent merchandise, if you wish to actually run them “within the background” as a substitute of a developer machine. It isn’t a brand new drawback, and to an extent a solved drawback, in any case we do that in CI pipelines on a regular basis. However it’s additionally not trivial, and in the mean time my impression is that surroundings maturity continues to be a difficulty 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.
Resolution high quality
I ran the identical immediate 3 occasions in OpenAI Codex, 1 time in Google’s Jules, 2 occasions regionally in Claude Code (which isn’t totally autonomous although, I wanted to manually say ‘sure’ to every little thing). Although this was a comparatively easy job and resolution, turns on the market have been high quality variations between the outcomes.
Excellent news first, the brokers got here up with a working resolution each time (leaving breaking regression assessments apart, and to be trustworthy I didn’t truly run each single one of many options to verify). I feel this job is an efficient instance of the categories and sizes of duties that GenAI brokers are already nicely positioned to work on by themselves. However there have been two elements that differed when it comes to high quality of the answer:
- Discovery of present code that may very well be reused: Within the log right here you’ll discover that Codex discovered an present part, the “dynamic information renderer”, that already had performance for turning technical keys into human readable variations. Within the 6 runs I did, solely 2 occasions did the respective agent discover this piece of code. Within the different 4, the brokers created a brand new file with a brand new perform, which led to duplicated code.
- Discovery of an extra place that ought to use this logic: The workforce is at the moment engaged on a brand new function 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 job 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 unsuitable” in a single or a number of of those dimensions, and the extra dimensions there are, the much less possible it’s that an agent will get every little thing carried out the way in which we wish it.
Sunk value fallacy
I’ve been questioning – let’s say a workforce makes use of background brokers for any such job, the sorts of duties which are type of small, and neither essential nor pressing. Haiven is an internal-facing utility, and has solely two builders assigned in the mean time, so any such beauty repair is definitely thought of low precedence because it takes developer capability away from extra essential issues. When an agent solely type of succeeds, however not totally – during which conditions would a workforce discard the pull request, and during which conditions would they make investments the time to get it the final 20% there, although spending capability on this had been deprioritised? It makes me marvel in regards to the tail finish of unprioritised effort we would find yourself with.
