Research Insights Made Simple #29: The AI-Native SDLC Playbook with Anton Kosterin (Rubric #AI4SDLC)

Coding agents can already write code much faster. But if planning, architecture review, security, testing, and deployment remain at “human speed,” does delivery actually accelerate—or does the queue simply move to the next stage?

On August 27 at 14:00, I will host Research Insights Made Simple live with Anton Kosterin, a principal engineer at T-Bank’s R&D center. We will examine Anthropic’s new AI-Native SDLC Playbook. Anton and I have already discussed architecture governance, RFCs and ADRs, and architecture reviews. This time the question is broader: how do we redesign not only code generation but the entire SDLC around agents?

The playbook’s thesis is that once Build accelerates, the bottleneck moves to Plan, Test/Review, and Deploy. Anthropic proposes turning the six stages from Plan through Maintain into a loop. intent.md, spec.md, plan.md, the diff and its tests, review findings, and incident records form a chain of version-controlled artifacts that also serves as an audit trail. Humans remain accountable, but their attention moves to the points where judgment is genuinely required.

We will discuss:

• Whether code has actually stopped being the bottleneck and where the queue moves next;
• Why markdown artifacts are needed when a company already has Jira, wikis, and architecture processes;
• How to turn institutional knowledge into CLAUDE.md, skills, and verifiable policies without preserving mistakes;
• How continuous evals, hooks, agent and human review, and production monitoring fit together;
• Where to start transforming the SDLC and which metrics can demonstrate an effect.

This is a practical guide from the Claude team, not a research paper with a comparative experiment. We will therefore test which parts are transferable engineering practice and which parts remain a vendor’s vision—especially for a large regulated organization.

Join us on August 27 at 14:00 and bring your cases and questions, especially if coding agents have already accelerated while delivery somehow has not.

#AI4SDLC #AI #Agents #Engineering #Architecture #DevOps #Research