AI writes the code. You still own the design.

“Agents are terrible at estimating how long a unit of work will take. In my AI-DLC I keep a state file per UoW which records when it starts, its progress and when it's complete; the delivery agent then looks at the past ten state files and hones its estimation policy, so its estimates get more accurate over time.”

“Storing where the orchestrator agent got to in a state.json file lets it pick up work more easily after an interruption, and lets you generate local agent dashboards that show what's in flight at any given point.”

“Part of your review agent's job should be checking that all of the scoped changes are in place in the files it expected, as agents have a habit of completing tasks without doing the actual work.”

“When your delivery agent creates the units of work, make sure each one links to the scope items, non-functional requirements and any email or screen templates that have been mocked up. After the build, it can then test that it has completed the work correctly.”

“Agents are terrible at estimating how long a unit of work will take. In my AI-DLC I keep a state file per UoW which records when it starts, its progress and when it's complete; the delivery agent then looks at the past ten state files and hones its estimation policy, so its estimates get more accurate over time.”Agents are terrible at estimating how long a unit of work will take. In my AI-DLC I keep a state file per UoW which records when it starts, its progress and when it's complete; the delivery agent then looks at the past ten state files and hones its estimation policy, so its estimates get more accurate over time.