Part 0 · 1 chapters · ~8 min
What AI-Native Means
The delivery loop rebuilt around agents: specification, scaffolding, implementation, testing, review and release, what the agent does in each and what the engineer owns, and how the skill mix shifts toward precise intent and rigorous verification.
1
What AI-native means
the loop, rebuilt around agents
- Specification: the engineer writes the intent; the agent challenges it.
- Scaffolding is cheap to produce and cheap to throw away.
- Implementation happens in small steps, against the spec and the tests.
- Testing: agents generate tests, and humans check those tests actually test something.
- Review: agents critique, and humans approve design, security and money handling.
- Release is gated as before. The loop is only as strong as its two ends.
| skill | before | AI-native |
|---|---|---|
| writing code | most of the time | less of the time; still essential for steering and for the hard parts |
| writing specifications | tickets, often thin | the main input: precise intent, constraints, edge cases, verification |
| reading and verifying code | in review | constantly: every agent output is a diff to verify |
| testing | written after or alongside | generated, then judged; tests become the contract for the agent |
| architecture and boundaries | important | more important: clear boundaries are what make agent work local and safe |
| knowing what not to delegate | implicit | explicit policy (part 5) |
the honest version
Agents make a strong engineer much faster on well-specified work in a well-structured codebase. They make a weak specification or a tangled codebase fail faster and more confidently. Everything else in this course follows from that.
THE AI-NATIVE LOOP
seven stages of delivery, what the agent does in each, and what the engineer still owns
swipe the figure sideways, or tap expand for full screen
1/6
specification
Specification: the engineer writes intent as a spec an agent can act on: the goal, constraints, interfaces, edge cases, and how to verify. The agent can help draft and challenge the spec ("what happens if the provider times out?"), but the decisions in it are the engineer's.