He also revealed that AI adoption is expanding beyond engineering to include design, planning, and finance departments, and that AI is even contributing as a participant in internal decision-making processes. Armstrong stated, “We’re testing the limits of it — when can it actually start to be the decision-maker on some things and do better than humans,” implying that AI is moving beyond being an assistant to eventually transform the very way businesses operate.

John Collison: What are other ways in which Coinbase is crypto-pilled, AI-pilled, works differently from a company founded 10 or 20 years prior?
Brian Armstrong. Like a lot of companies, we’re leaning as hard as we can into AI.

What does that mean, concretely?
We’re doing a lot of the best practices. We made a big push to get every engineer on Cursor and Copilot. Then another the question was, ‘Well, are they actually going to use it?’ Because a lot of them onboarded when I—

Was it you or Tobi [Lütke, member of Coinbase board] who mandated it?
I mandated it. Yeah.

You required people to have a call with you who—
That’s true. I did do that.

You required people to justify to you, the CEO, if they weren’t using AI code.
That’s true. Originally—

Sorry, maybe I’m not meant to tell that story?
No, I don’t mind. It’s actually a good story. Originally they were coming back and saying, ‘All right, over the next quarter, two quarters, we’re going to get to 50% adoption.’ I said, ‘You’re telling me— why can’t every engineer just onboard by the end of the week?’ So I kind of went rogue. I posted in the all-in Slack channel.

Just a light dusting of founder mode.
I said, ‘AI’s important. We need you all to learn it and at least onboard it. You don’t have to use it every day yet until we do some training, but at least onboard by the end of the week. And if not, I’m hosting a meeting on Saturday with everybody who hasn’t done it, and I’d like to meet with you to understand why.’

Now, a few people were on vacation. There was a list of— Anyway, I jumped on this call on Saturday and there were a couple people that had not done it. Some of them had a good reason because they were just getting back from some trip or something, and some of them didn’t, and they got fired.

Wow.
Some people really didn’t like it, by the way, that heavy-handed approach, but I think it did set some clarity at least that we need to lean into this and learn about it.

What’s your experience of AI coding been so far? It’s clear that it is very helpful to have AI helping you write code. It’s not clear how you run an AI-coded code base and what the best way to do it is.
I agree. I think we’re still figuring that out too.

One thing we started doing is every month we host, we call it an AI Speed Run where one of the engineers volunteers that month to run a training for how they’re using it. We try to cherry-pick the people, the teams that are doing it the best.

We’re doing about 33% of code written by AI now. We have a goal to get to 50% by the end of the quarter. Let’s see if we get there. You probably can go too far with it. You don’t want people vibe coding these systems moving money. We’ve really encouraged people to really — you have to code review it and have the appropriate checks in place on that with humans in the loop. But some of the front-end piece, etc., you can iterate faster.

We want to make sure it’s used not just in the engineering teams. It really should be any team. Design is using it heavily. Product managers. I think FP&A could even be using this as, ‘Ingest all the data and tell me what you forecast the revenue to be.’

We’re getting to a world where, even as CEO, by the way, I use it a lot. We use a decision-making process called RAPIDS and everyone writes their input. We have a row now for AI that writes its input in as one of the people that help make decisions. We’re testing the limits of it — when can it actually start to be the decision-maker on some things and do better than humans.