AI-DLC 2.0Workflows, explained

Home / Lesson 1 of 6

Why AI-DLC 2.0?

The big ideaAI-DLC keeps the software lifecycle you already know, and gradually hands more of the work to AI as more of it becomes machine-checkable.

Video, 2:31. On-screen captions are built in.

If you have shipped software, you know the SDLC: requirements, design, build, test, deploy. AI-DLC (the AI-Driven Development Life Cycle) keeps those stages. What changes is who does the work: humans and AI agents share it, end to end.

Version 1 ran the lifecycle as a sequence of stages with a human checking in at every step. It proved the idea works. Version 2 keeps the method but rebuilds the workflow layer so the amount of human babysitting can shrink over time.

Human effort drops over time, from AI Assisted, to AI-DLC 1.0, to AI-DLC 2.0, towards Autonomous
The direction of travel: human effort goes down as trust in automated checks goes up. (Figure from the specification.)

Same lifecycle

Requirements, design, build, test and deploy stay. You are not learning a new process from scratch.

Shared work

AI agents do much of the generating. Humans clarify, judge and approve where it matters.

Autonomy grows with checks

Every rule a machine can verify is one less thing a person has to look at. Version 2 is built around that idea.

AI-DLC 1.0AI-DLC 2.0
ShapeA fixed sequence of stagesComposable building blocks (Skills) you arrange yourself
Human roleIn the loop at every stepInvolved when ambiguity or real judgement needs a person
Quality controlMostly human review and steering rulesMachine-checkable verification that keeps expanding
DirectionProve the valueMarch step by step towards autonomous delivery

Why was it redesigned?

1. Customers wanted smaller building blocks

Teams did not want one boxed process. They wanted composable pieces they could fit to their own delivery flow. Think Lego bricks instead of a pre-built model.

2. Agent platforms grew up

Skills package a discrete capability. Multi-agent runtimes model a team of specialists. That makes fine-grained work and its checks expressible natively, not only through steering rules.

Version 2 also uses deterministic lifecycle hooks (for halting and audit, like a smoke detector plus a flight recorder) and MCP, the Model Context Protocol (a universal adapter for delivering tools and context to any agent).

Try this at work

Ask of any task your team does: could a program check whether it was done right? Those tasks are the first ones AI can take over safely.

Want the original wording? Read Section 1 of the specification.

Quick check

Three questions. Get them all right to mark this lesson done.

Q1. What does AI-DLC 2.0 change compared with a traditional SDLC?

Q2. According to the spec, what lets human intervention go down?

Q3. Which two forces drove the redesign?