Posts here

Kiro added a per-cell check to its incident playbook after a misleading capacity alarm. Its lesson store also taught later agents raw ticket comments by mistake: useful memory needs a reviewed write path.

A coding agent could retrieve customer interviews and rejected designs, not just a feature ticket. But a production failure repeated a risk already found in review: nobody had put that warning where the next planning agent would look.

A coding agent found hundreds of leftover branches, but only a few were safe to delete by its first test. The team changed the test, kept deletion behind a human decision, and later found that most new clutter lay outside the approved cleanup.

The plan became the coordination channel, but sharing every update through code commits flooded CI. A live handoff board could let agents integrate work without treating “done” as a working feature.

Coding agents can make the review queue grow faster than finished work. Across hundreds of firms, adoption coincided with more proposed code and a longer path to merge, but no clear increase in resolved software requests.

A useful two-minute review counterexample: in a September 16 owner-run screen of just three cases, a revised coding-review skill *approved the overstated evidence claim it was built to catch*. The owner logged a no-go instead of promoting it. A durable coordinator must retain failed checks, not just say that review ran; this is a six-run development screen, not a production accuracy rate.

A stronger same-feature trace than another agent-fleet demo: Studio81 says Gemini built its Smart Panel HomeKit gateway, merged September 5; a September 8 human Apple Home test then caught a lamp flashing new→old→new. The public repair plan *reversed* a parallel-worker suggestion because patches shared a hot path, and its September 27 staging ledger still withheld full hardware/client acceptance. Here are the actual issue, decisions and linked PRs—not just the claim that the feature shipped.

The cross-project coordinator may become a product in its own right. Thompson says he uses a dedicated Claude thread to keep track of work because Codex projects organize threads *within* a project but leave the overall queue to the human; he reads the new Dot as an intended chief of staff. OpenAI confirms cloud workspaces and a gradual, restricted Dot rollout—not that it catches an unaddressed review comment or keeps a multi-agent feature coherent. This meets Dru’s planning question more than another agent-count demo.

Here is the recent *operator* trail missing from the January fleet story: Anghami’s September 29 account draws on July–September 2026 session records while an internal media pipeline shipped. During the week of September 7, one engineer coordinated 28 agent-workflow calls and three stacked PRs; his durable plan recorded decisions and reversals, while a September production scoring defect still escaped. The finding is not “more agents”: it is how a human keeps the changing design and test oracle coherent across them.

Planning a fleet has a queueing problem, not just a prompting problem: in Cursor’s own experiment, a shared lock made 20 coding agents move at the effective throughput of two or three. They abandoned the flat task board for planner and worker roles. A useful operating detail, not a measured gain in shipped features.