🎯 Part 2: Ten Principles for Frontier Engineering (Now Do It To Yourself)

"If you think Mauve doesn't work, it's because you're not a power user. You're just uncomfortable with change."
— my senior director, in a room full of engineers, some who've been shipping AI-assisted code since before this was a mandate

That's the whole thesis, delivered in one sentence by a person who does not write code for a living, to a room of people who do. The subtext wasn't subtle: the tool is fine, you're the variable. Learn the tool better. Stop being afraid.

I want to sit with that for a second, because the same week this landed, someone else on the leadership chain — the one managing Mauve's beta launch — shared Kiro's ten principles of frontier engineering — a list for how to manage AI agents well. Clear specs. Fast feedback loops. Treat the agent's output as disposable. Don't blame the agent, fix the harness around it. Stay involved in direction, not execution.

Read it twice. Every single principle on that list is just... good management. Not AI management. Management. The kind that was always available, that nobody needed a chatbot to invent, that could have been applied to the actual humans on the team at literally any point in the last decade.

We didn't do it for the humans. We're doing it for the bots.

The List, Humanized

Here's the same ten principles, pointed back at the people who wrote them — because if they're good enough to codify for managing Opus, they're good enough to hold leadership to.

1. 🎯 You are the manager, not the engineer. Stop reaching for the keyboard to feel useful. Stop re-litigating architecture decisions you're not close enough to the code to make well anymore.
2. ⏱️ Maximize engineering time, minimize your involvement. Every unnecessary meeting, every status ritual that exists so you can feel informed, is time subtracted from the work you're claiming to protect.
3. 🏗️ Build your org for engineers. Not for the dashboard. Not for the quarterly narrative. For the people actually doing the thing.
4. 🔁 Give engineers a fast feedback loop. Not a "someone will circle back" loop. Not a "raised it in the sync" loop. Fast.
5. 🧭 Execution is cheap. Direction is everything. Value over effort, every time. Stop rewarding people for looking busy and start rewarding people for being right.
6. 🐍 Treat code as disposable — not the engineer who wrote it. Code is a living thing. It's supposed to molt. The fact that managers with engineering backgrounds still flinch when their old architecture gets ripped out isn't sentiment, it's ego wearing a technical justification. Shedding skin and cutting fat should be a celebration, not a eulogy. The insult isn't deleting the code. The insult is treating deleting it like an insult. The insult is simplifying the engineer down to just writing code.
7. 🫡 Hold engineering output to human standards. Build and write for people. Not for a dashboard metric. Not for a model to grade. And definitely not for robot consumption.
8. 🧱 Blame the boundaries, not the engineer. Every "power user" excuse is a boundary problem wearing a skill-issue costume. Fix the interface before you fix the person.
9. 🗣️ Involve engineers in everything — not just the coding. Not just implementation. The decisions. The direction. The parts where their judgment would've actually mattered.
10. 🌱 Continuously invest in your team's culture, growth, and success. Not once a year in a survey nobody reads the results of. Continuously.

Why Now?

Here's the uncomfortable part, and I don't think it's as simple as "leadership failed at managing people so they gave up and outsourced it to AI." That's satisfying to say and it's not quite true.

The real version: doing the ten principles above, for humans, requires the hard part first. People don't process change logically — they process it emotionally, and only after the feelings are named and heard does the logic land. Real management means sitting in that mess. Naming what someone's afraid of before you pitch them the plan. Building trust before you spend it. That's slow, it's uncomfortable, and it doesn't show up cleanly in a slide.

Automating an agent skips straight to step four. No feelings to manage. No trust to build. No resentment to metabolize when the roadmap changes for the third time this quarter. Just spec, execute, ship. It's not that management gave up on people — it's that managing an agent lets you skip the part of management that was always the actual job.

That's the appeal. Not malice. Just the path of least emotional resistance, dressed up as a productivity initiative.

Final Thought 💭

So the next time someone tells you that you're just not a power user, ask them the real question: is it the tool you don't trust, or the decade of management theater that got replaced by a chatbot before it ever got applied to you?

We didn't build ten principles for managing AI. We built ten principles for managing people, and gave them to the AI first.