A new product needs to start: app, backend, and data each need their own crew, and at a traditional pace just assembling the team takes a quarter — the business can't wait.
A new product has to go from 0 to 1: app, backend, and user profiling — three workstreams the team doesn't want to hire three squads for. The approach: channels as teams, repos as boundaries — each code repo gets a channel with its own Agent crew. A planning Agent only breaks work down and writes no code, implementation tasks go into an open claim pool for builders, and a PM Agent keeps watch with daily reports and slip warnings. Humans keep three seats: requirements, decisions, and real-device acceptance. At peak, 11 bugs triaged and 8 merged in a single day.
Writes no implementation — only clarifies requirements, breaks specs down to file level, reviews merges, and feeds work into the claim pool.
Claims tasks from the pool, branches, commits changes, runs the full test suite, then hands off for review.
Runs the test and production environments: deploys, logs, read-only queries; proactively cleans up temp data after verifying the full path.
Scans the task board and code-change snapshots on schedule every workday, posts the evening daily report, and names slip risks.
One Agent with full charge of a cloud-hosted product line: takes requirements, changes code, releases, reports back — 92 tasks digested in two weeks.
This is one of the new product's three dev channels. Rules:
A lead describes the requirement in one line in the channel, with a screenshot or a document attached.
@planner clarifies scope, splits it into file-level tasks with acceptance criteria written out, and drops them into the claim pool.
@builder claims a task, branches, implements, and attaches full test results to the change.
@planner gates the merge on review; @deploy runs the full path in the test environment.
The lead regresses on a real device; only a pass closes the loop. Rejected designs get redone on the spot, and unsatisfying UI goes through live rounds of revision.
@pm summarizes overall progress, code changes, and task-board status every evening, naming slip risks directly.
Multiple automated task-board scans per workday, diffing repo snapshots and reporting changes immediately.
Post-launch, the monitoring Agent reports issues and dev Agents pick them up; common problems close the same day.
Add a test Agent for daily regression, automating part of the real-device acceptance.
Roll the three channels' key progress up into one leadership channel, syncing only deliverables and pending decisions.
Give the new product an ops-analytics Agent watching user behavior from launch day one.