AI-Native Delivery
What happens to a delivery system when writing code stops being the expensive part. Transaction cost, batch economics, and why verification becomes the constraint you have to engineer.
Move faster because the flow physics say to, not in spite of them.
Foundational
2The Transaction Cost Collapse
AI-assisted engineering collapses the cost of producing a change. It does not collapse the cost of verifying one. This is the economic argument for what that does to batch size, and where the constraint moves next.
9 min readai deliverybatch sizeWhat AI Does Not Change
The corrective piece. Little's Law, dependency structure, Conway's Law, approval queues and funding models are all untouched. Coordination cost and conceptual integrity get harder at volume, not easier.
9 min readflowconstraints
Practitioner
3Code Review When Code Is Cheap
Review practice under high change volume. What a reviewer should look for when the author did not type the code, why pull request size discipline matters more than before, and who owns a defect nobody wrote by hand.
9 min readcode reviewverificationPutting AI In The Delivery Pipeline
Where assistants belong in the lifecycle beyond code completion — test generation, large-scale migration, decision capture, triage, review assistance — which of these pay back immediately, which are traps, and what provenance a regulated estate has to evidence.
9 min readsdlcautomationVerification Is The New Bottleneck
When producing a change is cheap, the constraint becomes establishing that it is correct. What that means operationally for review capacity, test confidence, and the shift from writing code to specifying and checking it.
9 min readverificationcode review
Advanced
2Forecasting When Throughput Changes
Historical throughput predicts the future only while the system is stable. What to do when capability is shifting under you — re-baselining, estimation practice, which DORA and flow metrics survive, and how to report it upward honestly.
10 min readforecastingmetricsRunning A Fleet Of Coding Agents
Several coding agents working a codebase in parallel. Decomposition so they do not collide, worktree isolation, why review capacity becomes the ceiling almost immediately, and the failure modes that are easy to miss.
10 min readai deliverycoding agents
Other domains