Pair Programming: You Drive, I Tell You What I'd Do as an AI Engineer

This is for the person who has an AI agent half-working and can't tell whether the next hour should go into fixing it or abandoning it. You might be an operator automating your own week, a founder who promised a demo, or an engineer who's new to agents and doesn't want to learn by burning a month of tokens. If you've been pasting the same instructions into a chat window every day and wondering what this is actually costing you, this is the hour that answers it.

You share your screen and keep the keyboard the whole time. I sit beside you the way a senior engineer would, telling you what I'd try next, why the thing you just tried didn't work, and what it's costing you in tokens and time while we do it. I don't take the wheel and I don't lecture; the point is that you can do it yourself on Monday, which is the only version of help that lasts. We finish by writing down what to do next week and what you should stop paying for.

One task you're actually trying to get an agent to do, in whatever state it's in, and access to the tools and accounts it needs. If you can tell me beforehand what you've already tried and roughly what you've spent, we skip the first fifteen minutes of archaeology and go straight to the part that's stuck.

The good outcomes are concrete. The agent does the task end to end and you know what a run costs. Or you find out the task doesn't price well with today's models and you stop spending on it, which is a result too. Either way you leave with a written next step, a number for what the work costs per run, and a clear sense of which rung of the ladder this belongs on: a chat subscription, an API with real accounting, a managed platform, or hardware you own.