The validation loop I mentioned at the end of the first writeup finally happened — I demoed the ADA agent to the senior developer whose 50-hour estimate is what started this whole thing. He liked it. He also had a suggestion I hadn't considered: have the agent generate a patch file for every file it touches, formatted for SVN.
Patch files were new to me
I'd heard of patch files before, but I'd never actually created or applied one, and SVN in particular wasn't something I'd worked with directly before this. The obvious question was whether the agent could handle something I didn't fully understand myself yet. So I tested it: told the agent to review every file in a folder, apply the ADA fixes it already knew how to make, and generate an SVN patch file for each one it touched. It could.
The "no patch needed" bug
It wasn't clean on the first pass. The agent would sometimes generate a patch file labeled "No patch needed" for files that actually did need ADA changes — technically following the instruction to produce a file per input, but wrong about whether a fix was warranted. I revised the agent until it consistently applied the correct ADA standards before deciding whether a patch was needed at all, instead of defaulting to "nothing to do" when it should have kept looking.
What's next
I reported the progress back to the senior dev, and the next step is demoing the whole thing — the review, the WCAG citations, the impact/feasibility scoring, and now the patch files themselves — live in the team meeting. That's the actual test of the question the first writeup ended on: how much of this stays something only I run, versus something the rest of the team can pick up. A working patch-file output is the difference between "here's a tool I built" and "here's something you can apply to your own branch."
The numbers make the case better than I can. ADA compliance work is estimated at 50 to 60 hours per system, and we have more than 50 systems that need it — a manual backlog measured in the thousands of hours. In that folder-scoped test, the agent generated patch files in under an hour. That gap is the whole argument for why this belongs in a team meeting instead of staying something I quietly run on my own.
Update: the per-file loop itself turned out to be the next bottleneck — see the follow-up on letting the agent scan an entire codebase on its own.