The Agentic SDLC

Agents need a delivery process

Generating code is one part of software delivery. Discovery, architecture, test planning, review, testing, and documentation still need to happen, and the work needs to survive the handoff between them.

Used in enterprise consulting, this session walks through an operating model in which an orchestrator coordinates specialist agents across that lifecycle. Each specialist has a focused task and produces a written artifact. The next stage starts with an explicit decision about whether the work is ready to proceed.

The central idea: agents can staff the SDLC. Engineering discipline still defines it.

What the session covers

• Focused agents and context. Why discovery, implementation, and review benefit from separate responsibilities and fresh working context.
• Written handoffs. How discovery summaries, architecture decisions, test plans, and review reports make the process inspectable and resumable.
• Parallel work with one decision. How an orchestrator combines specialist findings, resolves conflicts, and produces a single verdict.
• Gates and bounded correction. Where builds, tests, review findings, and explicit stop conditions belong, including when a failure needs human judgment.
• A process that fits the product. How reviewer responsibilities change with the codebase, infrastructure, and security risks, and how lessons feed back into the workflow.

What the audience takes away

A concrete way to reason about agentic delivery beyond a single coding assistant: who owns each step, what evidence they leave behind, and what must be true before work moves forward.

The material supports both a technical walkthrough and a leadership discussion about introducing agentic workflows into an existing engineering organization: how responsibilities change, how quality is assessed, and where people retain ownership of delivery decisions.