The Engineering of Intent, Chapter 21: Working With Teammates (Human and Agentic)

This is Part 21 of a series walking through my book The Engineering of Intent. In the previous chapter, we covered the weekly cadence that keeps daily practice honest. Chapter 21 steps out from the individual to the team — and to the blurred line between human and agent colleagues.

Vibe Coding Is Only Solo in the Narrowest Sense

Every piece of code you ship has reviewers, downstream consumers, and future maintainers. Some of those are agents; most are humans. Chapter 21 is about working well with all of them — the review contract, pair Vibe Coding, agent handoffs, and the mentorship that has quietly become a different craft.

The Review Contract

When you ask a colleague to review your PR, you are asking them to spend fifteen to thirty minutes of their finite attention. Honor it. Write a PR description that costs you five minutes and saves them ten. State: what this changes, why, what to pay attention to, what you’re uncertain about, and what the test coverage is.

In return: when you review someone else’s PR, read carefully enough that your comments are worth more than the cost of your time. Drive-by cosmetic comments are worse than no review. If you’re too busy to review thoughtfully, defer — don’t drive-by.

Pair Vibe Coding

Two humans, one agent. One drives, one navigates; rotate every thirty minutes. Output quality is materially better than solo. Speed is about the same. The navigator catches errors the driver would have spent hours fixing.

Handing Off to Agents (Without Walking Off a Cliff)

“When you hand a task to an agent and walk away, you are still responsible for the outcome. ‘The agent did it’ is not a defense; you chose to hand it off.”

Concretely: hand off tasks where the success criteria are checkable without you. If the task requires human taste to validate, do not walk away — supervise or delay. If the task involves irreversible action (production deploys, data migrations, external API calls with real effects), require human confirmation even from otherwise trusted agents.

Mentorship in the AI-Native Era

Mentorship has changed. The junior engineer no longer learns by typing boilerplate — the agent types it for them. Something real is missing, and it is widely mourned.

The mentorship that works now is different: the senior engineer shows the junior how to read agent output critically; how to structure a plan; how to frame the Context Pack; how to say no to a plausible-but-wrong suggestion. These are meta-skills that do not transfer by osmosis — they require explicit teaching.

Concrete practices: review the junior’s session transcripts, not just their final code (the transcripts are where the judgment is visible). Sit with them during the planning phase of a hard task; show them how you push back on a plan. Give them tasks that are slightly too hard; let them fail safely and discuss the failure carefully. The juniors who end up excellent in two years are the juniors whose seniors treated the mentorship explicitly.

Next up — Chapter 22: A Day in the Life — A Narrated Session. Every chapter up to this point has been theory or practice. Chapter 22 is the narrated walk-through of a real medium-complex day, hour by hour, to show what the practices actually feel like in use.