Continuous Release: From Two-Week Sprints to Continuous Deployment

Why sprints are not enough

For a decade, FourKites ran on the standard agile cadence: two-week sprints, sprint planning, sprint review, retrospective. This model works well for organizations where the rate-limiting factor is coordination. When AI becomes part of the engineering workflow, the rate-limiting factor shifts. Code generation, test generation, and review acceleration mean the bottleneck is no longer 'how fast can humans write code?' but 'how fast can we validate and deploy?'

The sprint boundary becomes artificial. A feature that is coded, tested, and reviewed by Tuesday should not wait until the sprint ends on Friday to deploy. AI-accelerated engineering demands a deployment pipeline that matches the new velocity.

The continuous release model

FourKites has moved to continuous deployment: code that passes automated testing, security scanning, and code review deploys to production when ready, not when the sprint calendar allows. The North Star metric is PRs merged to production per engineer per day, a measure of sustained throughput that is impossible to game and impossible to achieve without AI acceleration at every stage.

What AI accelerates

AI participates in every stage of the development lifecycle. Requirements decomposition: product requirements are translated into structured specifications that engineers and AI can both consume. Code generation: AI generates implementation code from specifications, with engineers reviewing and refining. Test generation: AI generates test cases from code patterns and requirements, catching edge cases that manual test writing misses. Code review: AI provides first-pass review for style, correctness, and security before human reviewers see the code. Documentation: AI generates documentation from code and commit messages.

Why this matters

The move to continuous release means new outcomes and agent capabilities ship in days, not quarters. When a customer describes a new workflow requirement, the path from requirement to production is measured in days: Sophie builds the workflow, engineers review the code, the pipeline validates and deploys. The sprint boundary is gone. The velocity is real. And the customers see the difference.