Six months into the AI coding tool rollout, the picture has clarified enough to say what worked, what did not, and what the next six months should look like. The tool the developer has been using, the tool the developer has been abandoning, the tool the executive team has been paying for. The honest framing matters here, because the AI coding tool rollout the executive team greenlit based on the demo the executive team watched has produced a six month track record the executive team has not yet read carefully.
What follows runs as the working version of the retrospective. The shorter version is what the engineering leader and the developer both actually have time to read.
What actually shipped faster
Three things, in roughly that order of how much each one shipped. The first runs as the boilerplate and the test, where the boilerplate the developer used to write by hand, the boilerplate the AI now writes, the test the AI now generates the test cases for, the test the developer used to skip when the developer was under deadline, the boilerplate and the test that the AI shipped in the time the developer used to spend on the boilerplate alone. The second runs as the documentation, where the documentation the developer used to skip, the documentation the AI now writes, the documentation that the on boarding developer now has, the documentation that the AI shipped faster than the developer used to skip it. The third runs as the small refactor, where the refactor the developer used to defer, the refactor the AI now proposes, the refactor the developer can apply with the diff the AI produces, the small refactor that the AI shipped at a rate the developer could not match by hand.
What got slower or worse
Three things, in roughly that order of how much each one hurt. The first runs as the review burden, where the review the developer used to do on the small change, the review the developer now has to do on the AI generated change, the review that takes longer than the change would have taken by hand, the review that has started to consume the senior developer’s calendar. The second runs as the architecture drift, where the drift the AI introduced into the codebase (the inconsistent pattern, the duplicated utility, the missing error handling), the drift the developer now has to clean up, the architecture debt the AI is accumulating faster than the developer can pay down. The third runs as the security regression, where the regression the AI introduced (the SQL injection, the hardcoded secret, the missing auth check), the regression the developer caught in the review, the regression the developer would not have written by hand, the security regression the AI is generating at a rate the code review process is not designed to catch.
What the next six months should fix
Three moves if you are the engineering leader who wants the next six months to look better than the last six months. Tune the tool to the codebase, because the tool tuned to the codebase (the custom rules, the example patterns, the test conventions the AI has been told to follow), the tool that fits the architecture the team has built, the tool that ships the refactor the team would have shipped by hand, the tuning that the platform team can do in a quarter. Add the review automation, because the review automation (the static analysis, the security scanner, the architecture linter) the AI generated code has to clear before the developer reviews the code, the automation that catches the regression the developer would not have caught in the manual review, the automation that costs a sprint to set up and saves the bug the AI would have shipped. Measure the right thing, because the line of code the AI produced the executive team has been celebrating sits as the line of code the executive team should be ignoring, the right metric the team should measure (the cycle time, the bug rate, the security regression rate), the metric that the next six month retro should report. The leader who tunes, automates, and measures serves as the leader who has turned the AI coding tool into the productivity gain the demo promised.
The bottom line
AI coding at the six month mark sits as the tool that shipped the boilerplate, the documentation, the small refactor faster, and the tool that added the review burden, the architecture drift, the security regression. The tuning, the review automation, the right metric, those three are what the next six months should fix. The leader who does the three holds the productivity. The leader who keeps the executive demo as the only metric does not.
Sources & Further Reading
All claims in this article are sourced from primary documentation, vendor advisories, and reputable security researchers.
Spotted an error? Email the editor. Corrections are issued with a visible correction note.
Editorial standards. Every article on humanrequired.org is reviewed by a human editor before publication. AI may assist with drafting or research; final editorial control is human. Read the full standards.
