Blog

Notes on building the control tower for an autonomous software factory — pipeline observability, agentic copilots, and what we learn wiring four products together.

• Jul 23, 2026 The cockpit stops lying to you This cycle was about honesty: a board that prunes its own phantom tasks, a live-agents panel that admits when a verify is running, human-in-the-loop actions that never write without a click, and one page that documents every knob.
• Jul 19, 2026 One control tower over autonomous agents A plain GitHub issue went in and a tested pull request came out with no human in the loop. This is that run watched from the cockpit — the pipeline strip, the event feed, the live agent terminals, the review queue, and the bill.
• Jul 17, 2026 One cockpit, many tenants Work items and the cockpit APIs are now tenant-scoped behind a flag, the identity layer that feeds them is wired end to end, and fixed issues finally close themselves.
• Jul 13, 2026 If you cannot see it, you cannot trust it The cockpit's job this cycle: make the cost of a run, the model tier each stage used, and the difference between a real success and a silent no-op something an operator can actually watch.
• Jul 7, 2026 The cockpit becomes the shell: one screen to run the whole factory A fleet-wide command palette, a 'needs you' inbox, a portal switcher, and a full light mode land in Mission Control — with a one-page showcase ready to download.
• Jun 23, 2026 Every requirement, its test, in one table The cockpit now answers the question a reviewer actually has before they approve a build: is every acceptance criterion actually tested, and how far was it verified? A traceability matrix renders acceptance-criterion x test x VAL x verdict in one pane, and an AC with no covering test is flagged as a gap rather than left quietly blank.
• Jun 21, 2026 The cockpit shows the graph, the mode, and the logs Week of 15-21 June: the task-detail view now renders the live execution graph across all three PARR stages, the Tokens page tells metered dollars apart from subscription and local runs, and the concurrency work from RFC-0016 finally has somewhere to be watched. Job-native log streaming is wired through but not yet on by default across the fleet.
• Jun 13, 2026 The cockpit learns to show live, per-worker cost A day of work took the Tokens & cost page from a per-service total to a per-worker drill-down, then put a live, ticking cost stamp and a cumulative sparkline on every running task — all additive, all gated on a real test run.
• Jun 7, 2026 The cockpit takes shape: seven views, one factory Three days of work turned CFactory from a board you watch into a control tower you operate — running-tasks, a live audit feed, service health, and a cockpit that now ships itself.
• Jun 5, 2026 Watch the agents, and see the cost Two updates land in the cockpit: live agent terminals streamed into Mission Control, and real token spend across all three stages.
• Jun 4, 2026 Introducing CFactory: a control tower for your software factory Three autonomous tools are powerful. Three you can see, correlate and steer from one place are a factory.