LLMs: Not the Intelligence; It's the Momentum.

There's a specific kind of focus you reach when everything you need is in one place. No tab-switching, no context-switching, no Slack pings pulling you away from what you were actually thinking about. Just you and the problem in front of you. I didn't think I'd find that in an AI — but here we are.

Over the past month, working with LLMs has quietly collapsed my entire workflow into a single window. And once you experience that, it's very hard to go back.

Why flow matters more than you think

Flow state isn't some productivity buzzword. It's that rare condition where your thoughts and your hands are moving at the same speed, and the work just happens. The enemy of flow isn't laziness — it's friction. Every time you switch to a different tool to look something up, debug an error, or run a server that refuses to start, you lose the thread. You spend the next ten minutes just trying to remember where you were.

Before I started leaning on AI for real work, that friction was constant. Setting up environments, chasing integration errors, figuring out why the server was throwing a cryptic error — these weren't hard problems exactly, but they were slow problems. And slow problems kill momentum.

What changes when the heavy lifting disappears

When AI handles the mechanical stuff, I stop thinking about the syntax and start thinking about the idea. I'm not debugging import paths — I'm deciding what the product actually needs to do. I'm not reading documentation — I'm building.

This sounds obvious when you say it out loud. But the actual feeling of it is different. You realise how much mental energy you were spending on things that had nothing to do with your real job. The ideas were always there. The bottleneck was everything surrounding them.

I built my portfolio until 3am

A couple of weeks ago, I decided to rebuild my personal website from scratch — on a whim, late at night. I just started thinking out loud. What features would I actually want? I typed it all out roughly, got my thoughts in order, and started building. I didn't stop until 3am.

I went to sleep with the project still running. The AI had been filling in code, catching issues, suggesting better approaches the whole time. When I woke up and ran the server the next morning — half expecting to spend the day fixing it — it just worked.

I sat there for a moment, genuinely surprised.

There was a time when just getting a local server running could eat an entire afternoon. You'd install something, something else would break, you'd find a Stack Overflow thread from 2018, try the fix, introduce a new error. Hours gone. I wouldn't have even attempted that project before. Not because I couldn't — but because the cost of all that friction made it feel not worth it.

It becomes addictive — in the best way

Here's what nobody really talks about: this way of working is genuinely hard to stop.

When everything is in one place, when the friction is gone, when you're actually making progress on the thing you care about — you just don't want to stop. The 3am thing wasn't a one-off. It keeps happening. Not because I'm grinding, but because I'm in it. The problem is interesting, the momentum is real, and stopping feels like a loss.

That's what flow does to you. It's not just productive — it's enjoyable in a way that regular work often isn't. You're not fighting your environment. You're thinking, deciding, building. And the next thing you know, it's been four hours and you've shipped something you're actually proud of.

MCP made the single window actually mean something

LLMs alone are powerful — but they were still somewhat isolated. You'd get an answer, then go somewhere else to act on it. MCP — lets LLMs connect directly to your tools. Your CMS, your calendar, your design files, your codebase. Not just read them, but act on them. From the same window.

I publish articles through an MCP server I set up on my own website. I can tell Claude to list my drafts, update a piece, push it live — without touching the CMS once. I describe what I need, and it happens where I am.

Developers have real concerns about MCP. When you're a power user who cares about token efficiency, bloat matters. But for a designer, a writer, a founder — someone who just wants to get things done without managing infrastructure? It's brilliant.

The best co-worker I've always wanted

A great co-worker doesn't just do their part. They make your part easier. They catch things you'd miss, suggest approaches you hadn't considered, and handle the annoying stuff so you can focus on the thing that actually needs your brain.

That's what a good LLM with the right connections does. It doesn't replace your judgment — yours is still the one that matters. But it removes the thousand small interruptions that were quietly making you less effective.

One window. One thread of thought. The work, and you. That's it. That's the whole shift.