Somewhere along the way, I found myself leading a team again. But this time the team wasn't just people. AI had quietly taken a seat too. What should I settle first? That was the first thing on my mind as I took the team back on, and it felt quite different from before.

Working alongside AI, I noticed one thing: the clearer the goal, the better the results. The more sharply I conveyed what I wanted, the better the quality of what came back. But when I sat with it for a while, I realized this isn't only true for AI. It's the same with people. A team does its best work when it's clear where we're headed. In fact, long before AI showed up, getting a team to look in one direction and run together always seemed to require some kind of 'baseline'. So I decided to draw up 'team principles' with both people and AI in mind.

As I set them down, I held to one standard: even as AI keeps advancing and society shifts, pick only the things that won't change. Principles built on what's changing would go stale fast. Narrowing it down that way, the team principles came together as two promises. One is 'trust', the promise between people. The other is 'AI Native', the promise between people and AI.

First, the promise between people: trust. As a team grows, processes and rules inevitably pile up. But when a team is small, keeping mutual trust seemed like the cheapest way to operate. When there's trust in each other's work, things come up that you can hand off and rely on without checking every detail, and a lot of invisible costs fall away. The reverse is true too: once something goes wrong, or trust starts to crack even a little, rules and procedures begin to sprout one by one, even in a small team. Maybe processes are what grow to fill the space that trust leaves behind.

So what builds trust? Looking back on my past experiences and reopening a few books, I came to settle on three things.

The first is predictability. We feel trust toward people whose words and actions line up. From small things, like the expectation that someone will keep a meeting time, to big ones, like having a sense of how a task you handed off will turn out. Because we can predict what comes next, we can work together with our minds at ease.

The second is transparency. I think the smaller the team, the more lively the discussion has to be. It matters for building a better service, and it matters for getting past the limits you can't reach on your own. So people need to be able to actively put their thoughts out there. The opposite, staying quiet while everyone is talking and then suddenly saying something different, or moving in another direction, once it's time to act on what was agreed, there's nothing that eats away at trust quite like that. Someone who lays their thoughts out transparently when discussing, and then cleanly turns what was agreed into action. That kind of person gives you trust.

The third is follow-through. There are people who are good at starting things but let the final wrap-up trail off every time, and there are people who, whatever the circumstances, see it through to a real result. I think I come to trust the latter. Because you can see the attitude of owning the promised scope and quality all the way to the end.

Next, the promise between people and AI: AI Native. Our team aims to be AI Native. Once we'd decided to work side by side with AI, that relationship needed promises too. What would be good to promise, and what can each side expect from the other? I held onto that question for a long time, and again I landed on three things.

The first is context sharing. AI will keep getting smarter. To raise the synergy between that smartness and our collaboration, I came to think the key is, in the end, handing over 'context' well. AI can expect to receive more context from people, and people can expect that the more context they pass along, the better the results they get back. This two-way expectation each holds for the other is, perhaps, the most basic promise between people and AI.

The second is reversibility, the idea that things have to be undoable. I think the default for what AI does should be a form you can always reverse. When something can't be undone, it gets hard to trust AI with bolder work. Because if an unwanted result comes out, undoing it costs both people and AI no small amount of resources. Only when there's the safety net of being able to undo can we reach out to AI more readily.

The third is entropy management. When I think about what AI does best, it comes down to 'generation'. Code, writing, documents, whatever it is, the sheer speed of producing something is overwhelming, far beyond what people can keep up with. But without a person setting a baseline or going through the work of verifying, unnecessary information keeps piling up in what AI pours out, and meaningless output drags the overall quality down instead. So I came to the thought that people have a duty to manage this entropy, to tend to it so that AI produces high-quality results.

After all this deliberation I did set the principles down, but I know full well this is only the start. In the end they'll only mean something once I keep sharing them with the team, strive to keep them myself, and we all work to uphold them together. Shortcomings will surface and the principles themselves will get refined bit by bit, but in the end what I want is to keep these promises, with people and with AI, one step at a time. And so, slowly but without stopping, I find myself resolving to build a good team.