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 domestically, 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 by means of a complete job. Code will get created in an setting spun up completely for that agent, and normally leads to a pull request. A few of them are also runnable domestically although. Instrument examples: OpenAI Codex, Google Jules, Cursor background brokers, Devin, …
I gave a job to OpenAI Codex and another brokers to see what I can be taught. 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 have now an inside software 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 software is public.
The duty I gave to Codex was concerning 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 changed into "Shopper-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 "Shopper Analysis" and "Supply Administration"
This job is
- Comparatively small
- A typical “cosmetics” job that’s not tremendous pressing and would possibly get deprioritised repeatedly
Agent logs
I’m sharing the detailed logs right here in an effort to get an impression of what’s going on beneath 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 must examine the AGENTS.md within the repository root for directions.”
- seems for AGENTS.md, however I had really referred to as it AGENT.md, which led to some backwards and forwards 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 really occurring behind the scenes, as a result of the messages are approach too brief to signify the complete “considering” the mannequin will need to have been going by means of.
Trying to find the related code with numerous grep
| Message | Command (samples) | Notes |
|---|---|---|
| I must 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 undertaking. | ||
| Grep output is very large due to node_modules; I’ll restrict it to src as an alternative. | ||
| I’ll seek for “to human” within the src listing to seek out related references. | ||
| I’ll open and examine 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 ponder if there’s a perform 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 operating 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 massive node_modules folder. | third time it’s operating into this, didn’t ‘keep in mind’ | |
| I believe the logic is likely to be lacking, so I’ll examine _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 examine 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 examine 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 examine 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 examine 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 out within the ui folder – it discovered the CSS beforehand, and is now searching for current code that it could actually reuse to resolve the duty. That’s nice, as a result of we certainly have already got the same perform, and it finds it:
| Message | Command (samples) | Notes |
|---|---|---|
| I want a perform to transform tags like “client-research” to “Shopper 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 shifting by means of the web page to get the small print I want. |
It now finds a niche within the current perform’s implementation:
| Message | Command (samples) | Notes |
|---|---|---|
| I see that hyphens aren’t dealt with by the perform. I want to regulate it for hyphens. | ||
| I’ll modify toReadableText to deal with hyphens, making certain correct capitalization and spacing. Let’s examine assessments! | It really didn’t examine any assessments | |
| I’ll examine _dynamic_data_renderer.js to look at toReadableText and modify it for hyphen-case. | ||
| I’ll examine assessments/_sidebar.check.js to evaluate the associated classes there. | ||
| I’ll examine different information to see if toReadableText seems anyplace. | ||
| I’ll examine if _sidebar or others include a toReadableText perform. |
Replace the check for the perform that it desires to vary:
| Message | Command (samples) | Notes |
|---|---|---|
| I’ll regulate the perform to deal with hyphen-case, then replace assessments to mirror that. | ||
| I’ll examine the check file after line 36 to make sure hyphen-case is roofed. | ||
| I’ll scroll by means of strains 60 to 120 to look at the remainder of the file. | ||
| I’m scrolling the remainder 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 regulate _dynamic_data_renderer.js and index.js to make use of toReadableText and replace assessments. | git standing --short |
You may configure some setting 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 examine and clear them up. | ||
| I’ll examine 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 evaluate the repository’s .gitignore to verify if Node modules are excluded. | grep yarn .gitignore |
|
| I see no entry for .yarn or .yarnrc. I’ll examine and take away them if wanted. | cat ui/.gitignore |
|
| I’ll examine 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 regulate the _dynamic_data_renderer assessments. | ||
| I want to exchange {tag} with {toReadableText(tag)} for higher rendering. | ||
| I’ll check out the variations to raised perceive the modifications. | git diff |
|
| I’ll evaluate 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 contemplate 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 evaluate the logs to assemble extra data. |
Attempting to run the assessments – unsuccessfully:
| Message | Command (samples) | Notes |
|---|---|---|
| I’ll run the yarn assessments to examine if the whole lot’s working nice. | ||
| Checks would possibly’ve been cached or want community—output’s empty, so I’ll examine the log. | ||
| The lockfile isn’t up to date. I’ll attempt operating ‘yarn set up’ for the reason that setting permits it. | ||
| The setting 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 examine 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 examine .yarn for monitoring. | ||
| I’m double-checking the bundle.json to see if “haiven-ui” is the right title. | ? | |
| I’ll examine the Node model to verify if it’s 18 or 20 by operating node -v. | bizarre, the node model laid out in bundle.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 organising 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 setting. | 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 examine if it wants creation or already exists. | ||
| It seems just like the .venv listing is lacking. Let’s evaluate AGENT’s directions fastidiously. | ||
| The setting appears offline, however there is likely to be a script to arrange dependencies domestically. | ||
| I can’t run “poetry set up” offline, however I’ll examine 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 examine 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 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 setting 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 attention-grabbing to see that Codex, and a lot 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 means 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 a whole lot of seemingly extra subtle 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 remains to be fairly easy, however doing textual content search with grep is the only potential.
It looks like the device creators have discovered that this easy search remains to be the simplest in spite of everything – ? Or they’re making some type of trade-off right here, between simplicity and effectiveness?
The distant dev setting is vital for these brokers to work “within the background”
Here’s a screenshot of Codex’s setting configuration display screen (as of finish of Could 2025). As of now, you’ll be able to configure a container picture, setting variables, secrets and techniques, and a startup script. They level out that after the execution of that startup script, the setting won’t have entry to the web anymore, which might sandbox the setting and mitigate a number of the safety dangers.
For these “autonomous background brokers”, the maturity of the distant dev setting 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 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 recognized, it will have simply been in a position to repair the assessments, it was a trivial repair:
This explicit undertaking, Haiven, really has a scripted developer security internet, within the type of a fairly elaborate .pre-commit configuration. () It could be ultimate if the agent may execute the complete pre-commit earlier than even making a pull request. Nevertheless, to run all of the steps, it will must 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 should be accessible in the correct variations as effectively, after all.
Determining a easy expertise to spin up simply the correct setting for an agent is vital 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 spite of everything we do that in CI pipelines on a regular basis. But it surely’s additionally not trivial, and in the intervening time my impression is that setting maturity remains to be a problem in most of those merchandise, and the person expertise to configure and check the setting 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 domestically in Claude Code (which isn’t totally autonomous although, I wanted to manually say ‘sure’ to the whole lot). Despite the fact that this was a comparatively easy job 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 assessments apart, and to be trustworthy I didn’t really run each single one of many options to verify). I believe this job is an efficient instance of the kinds and sizes of duties that GenAI brokers are already effectively positioned to work on by themselves. However there have been two facets that differed by way of high quality of the answer:
- Discovery of current code that might be reused: Within the log right here you’ll discover that Codex discovered an current element, 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 presently 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 really found that and instructed 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 need to “go proper”. Every agent run can “go improper” 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 the whole lot carried out the best way we would like it.
Sunk price fallacy
I’ve been questioning – let’s say a workforce makes use of background brokers for one of these job, the sorts of duties which might be type of small, and neither essential nor pressing. Haiven is an internal-facing software, and has solely two builders assigned in the intervening time, so one of these beauty repair is definitely thought-about low precedence because it takes developer capability away from extra essential issues. When an agent solely type of succeeds, however not totally – wherein conditions would a workforce discard the pull request, and wherein 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 in regards to the tail finish of unprioritised effort we would find yourself with.
