Area 7

Practical workflow
& best practices

Where the mental models meet the day job: when to reach for AI, how to verify it, and how the role of the engineer is changing.

When to use AI — and when not

Great for drafts. Risky for safety-critical.

Lean in for prototyping, boilerplate, refactors with good test coverage, and exploration. Be cautious in regulated or safety-critical systems without rigorous verification.

REACH FOR IT

Prototyping & boilerplate

Refactors backed by tests, scaffolding, and exploration — where speed compounds and mistakes are cheap.

VERIFY RIGOROUSLY

Safety-critical systems

Medical devices, aviation, finance. The ACM's April 2026 TechBrief warned that fast, unchecked AI coding often skips over core engineering practices.

Verifying & reviewing output

Treat output as a junior's draft.

Read it, test it, review it. Tests are both your verification and your best spec — red/green TDD turns the agent loose to iterate until green.

Keeping a human in the loop

Confirm destructive ops.

Scale autonomy with risk. The MCP spec itself recommends a human be able to deny tool invocations before they run.

Where things stand in 2026

AI coding is already the norm.

The speed gains are real — but they only hold when review keeps pace with generation.

92%
of US devs use AI coding tools in and out of work
~41%
of global code reported AI-generated
3–5×
faster prototyping
~70%
rise in PR volume since late 2022, flat headcount