An Iron Man suit for the mind

When people ask me how I work with AI, I reach for Tony Stark.

The suit makes Stark faster and stronger, able to do things no unaided human can. He fuses with it. But the suit doesn't replace him. He is still the one deciding where to fly, what to lift, when to stop. The moment the suit starts flying him instead of the other way around, the same power that lifted him can put him through a wall.

That is how I think about AI. It is an Iron Man suit for the mind: something you put on, that takes you to greater heights, and that can make you dizzy and get you hurt if you stop being the one in control.

I talked through all of this with Dan Blumberg on his podcast, Future Around & Find Out. Dan and I first worked together years ago at The New York Times, and it was good to pick the thread back up with him on the question I spend most of my days inside: how do you move fast with these tools without quietly handing them the parts of the work that should stay yours?

Two ways to get it wrong

The first is what people now call vibe coding. You describe what you want, the model produces something that runs, and you ship it without really understanding it. It feels fast. It is fast, right up until that code reaches production and you find that nobody, human or machine, actually owns it. The speed was borrowed against a debt that comes due later.

The second is handing the whole job to an autonomous agent and stepping away. Let the system choose the architecture, make the trade-offs, decide what "good" means. This also feels efficient. But the architecture, the judgment, the thinking, and the learning are exactly the parts I want to keep. They are where the real work of engineering lives. Give them away and you are no longer the engineer. You are a spectator with a deployment button.

The middle path

Synthesis engineering is what I call the path between those two.

Use AI to go as fast as it will let you, and keep for yourself the things a human should not give up. I keep the architecture: how the pieces fit, and what the whole thing is for. I keep the judgment: which trade-off is right here, what "done" means, when to stop. I keep the thinking and the learning, because the day I let the suit think for me is the day I stop getting better at my craft.

The AI drafts; I direct. I read what it produces the way an editor reads a writer, or a senior engineer reads a junior's pull request: closely, with taste, ready to throw work away. Some of the most satisfying days I have had lately were spent directing a team of AI agents the way I once directed teams of people. The gain is real. So is the responsibility.

There is a question Dan asked that I keep turning over. Would Tony Stark have trusted his suit if he didn't know how it was built?

I don't think he would have, and I don't think we should either. That is much of why I have open-sourced all of the synthesis skills: the methods for synthesis engineering, coding, writing, and project management. As we hand more of our thinking to systems we can't see into, the ability to look inside matters more, not less. A tool you can inspect gives you a basis for trust: you can question its methods, improve them, and judge the results. A black box you have fused with is something else.

The conversation

The full episode with Dan covers more than I can fit here: what AI slop actually looks like, whether all of this is making us sharper or duller, how a non-programmer can now build a better version of his own software, and what AI does to the economics of open source.

It is on YouTube, Spotify, and Apple Podcasts. If you have been trying to find your own footing with these tools, faster but still yours, I think you will get something from it.

The synthesis skills are free and open at synthesisengineering.org and on GitHub. Put on the suit. Just stay the one flying it.