Coworking with AI

I don't use AI. I work with it. That distinction took me a while to understand, and longer to get right.

For the first year or so, I treated it like a very fast intern. I'd give it tasks, it would do them, I'd review the output and either accept it or send it back with corrections. The relationship was transactional. Request, response, done.

That's not how I work anymore. These days it's closer to having a thinking partner who happens to be available whenever I need them, knows everything I've told them, and never gets tired of my half-formed ideas.

The shift

The change happened gradually. I noticed I was having conversations rather than issuing commands. I'd share context, think out loud, let the reasoning develop across multiple exchanges. The AI would push back on things, offer alternatives I hadn't considered, ask clarifying questions that made me realise I didn't actually know what I wanted yet.

It stopped feeling like a tool and started feeling like a collaborator. Not a human one, obviously. But not a search engine either. Something in between that didn't have a name yet.

The thing people misunderstand

My wife pointed out recently that I spend more time talking to my AI than to her. She wasn't wrong. But the thing is, I'm not really talking to the AI. I'm talking to myself.

The AI is the medium that forces me to articulate what I'm thinking. To take the half-formed thing rattling around my head and put it into words. And once it's in words, I can see whether it makes sense. The AI responds, and that response becomes a surface to push against, a way of challenging the thought as part of the process.

It's not companionship. It's externalised thinking. The conversation is with yourself, the AI just makes it possible to have it.

That's why "using AI" never felt like the right frame. You don't use a thinking process. You engage in one.

The rhythm

My day has a shape to it now that's built around this collaboration.

Morning is for thinking together. I'll bring whatever's stuck from the day before, whatever needs untangling, whatever I woke up with half an idea about. We'll talk it through. Sometimes that means drafting something. Sometimes it means exploring options. Sometimes it means being told that the thing I thought was the problem isn't actually the problem.

Afternoon is for building. Code, content, structure. The kind of work where I know what I want and need help executing it well. The collaboration is tighter here, more back and forth, more iteration.

End of day is for review. Looking at what got made, catching the things that slipped through, planning what tomorrow needs.

The rhythm matters because context matters. A conversation that's been building for an hour has momentum. Starting fresh every time means losing that momentum, explaining things again, rebuilding the shared understanding.

What I've learned

Context is everything. The more the AI knows about what you're trying to do and why, the better it gets at helping. I front-load context now. Background, constraints, what good looks like, what I've already tried. It feels slow at first but saves time overall.

Push back on the first answer. Not because it's wrong, but because it's often safe. The interesting stuff comes when you say "that's fine but what if we tried something bolder" or "I don't love that, what else?" The AI isn't precious about its suggestions. You shouldn't be either.

Know when to start fresh. Sometimes a conversation goes down a path that isn't working and the best thing is to open a new one with a clearer sense of what you actually need. Sunk cost fallacy applies to AI conversations too.

Treat it like a collaborator, not an oracle. It's not here to give you the answer. It's here to help you find it. That means staying engaged, staying critical, staying in the driving seat. The moment you start accepting output without thinking about it is the moment the quality drops.

Let it see your working. I used to clean up my thinking before sharing it. Now I share the mess. The half-finished thoughts, the contradictions, the "I'm not sure about this but..." The AI is better at helping when it can see where you actually are, not where you wish you were.

The resistance

People are sometimes uncomfortable with this framing. It feels like anthropomorphising, or like giving up agency, or like pretending a language model is something it isn't.

I understand the discomfort. But I think it misses the point. I'm not claiming the AI is conscious or that it cares about my work or that we're friends. I'm saying that treating it as a collaborator produces better outcomes than treating it as a tool. The frame changes the behaviour, and the behaviour changes the results.

When you treat something as a tool, you use it. When you treat something as a collaborator, you work with it. The work is different.

Still learning

I'm three years into this way of working and I'm still figuring it out. The technology changes, the capabilities expand, the patterns that worked last month need updating. That's part of what makes it interesting.

But the core thing stays the same: this works better when you show up as a collaborator rather than a commander. When you bring your thinking, not just your tasks. When you stay curious about what's possible rather than just trying to get things done.

It's not using AI. It's coworking with it. And once you make that shift, you don't go back.