There's a ceiling to what you can do with AI in a browser tab. Most people hit it around the three-month mark. The novelty wears off, the answers start feeling generic, and you're back to doing most of the work yourself — just with a slightly smarter autocomplete in the loop.
I hit that ceiling. And then someone showed me how developers actually work.
I Hit a Ceiling
If you're using ChatGPT or Claude in a browser, here's what your workflow probably looks like: you type a question, read the answer, copy something useful, close the tab, open your document, paste it in, tweak it, and start again. The AI has no memory of yesterday. It doesn't know your clients. It can't see your files. Every conversation starts from scratch.
That's not a limitation of the AI — it's a limitation of how you're accessing it.
Developers don't work this way. When they use Cursor (an AI-native code editor), they're working inside an environment where the AI can see their entire project, remember context across sessions, run tasks in the background, and operate multiple workstreams concurrently. The AI isn't answering questions. It's doing work.
How I Got Here
Earlier this year I came across the work of Dave Killeen — Field CPO at Pendo, former BBC and Bumble product leader, and the creator of Dex (heydex.ai), an open-source personal AI operating system. Dave built Dex to run inside Cursor, not a browser. The premise: your AI should live where your work lives, with persistent context, structured skills, and the ability to act — not just advise.
I set it up over a weekend. Within two weeks I was doing almost all of my work through an IDE (the developer interface) instead of a browser.
"Two weeks of working almost exclusively with an IDE for my everyday non-technical tasks and I can tell you, this has completely reshaped how I think about using my computer. My productivity is up 5x again — through concurrency, less app-hopping and going much deeper into all trades I would usually be waiting to interact with."
— Claire Byrne
The productivity gain is real. But the bigger shift was conceptual.
What the Developer Approach Actually Unlocks
Persistent context
In Cursor with Dex, the AI knows who you are, what you're working on, and who your clients are — because that information lives in files it can always see. You don't re-introduce yourself every session. Your preferences, your style guides, your project status: it's all there, always loaded.
Concurrency
This is the one that surprises people most. I ran 12 complex tasks in parallel in a single session — the kind of work that would have taken a full week of back-and-forth. The IDE model lets you kick off multiple work streams and review the outputs, rather than waiting for each one to finish before starting the next.
Skills, not prompts
Dex comes with 65+ built-in skills accessible via slash commands or voice. /daily-plan pulls your project files, scans your calendar, and surfaces missed check-ins. /email-triage applies your custom rules and surfaces what actually needs attention. /slide-deck uses your saved style guide to generate a presentation. You stop crafting prompts from scratch and start running repeatable workflows.
The role shift
Don't tell the AI how to do something. Tell it where you want to end up. Then get out of its way.
What I Saw in the PM Sessions
I ran two sessions for product managers earlier this year, walking them through this live. The questions were revealing.
The people getting the least from AI were treating it as a smarter search engine — one question, one answer, close the tab. The people getting the most had already started treating AI as a collaborator with context. One attendee had connected his AI to his project management system and his company org chart. He was doing in an afternoon what used to take a week.
The insight
What separated them wasn't technical skill. It was whether they'd made the shift from chat to context.
The Setup — It Took Me a Weekend
I did this over a Saturday and Sunday. No coding involved. Here's exactly what I did.
Can you use this at work?
Dex runs entirely on your machine. Your files don't leave your computer — nothing is shared with anyone beyond your normal Claude usage. That makes it easier to clear with IT than most SaaS tools. If your organisation already has a Claude or Anthropic agreement, you're likely covered. If you're setting this up personally, all you need is a Claude Pro subscription (around $20/month).
Step 1 — Get Claude Cowork
Download Claude and set up Claude Cowork — Anthropic's desktop agent that lets Claude work directly with files and folders on your machine. This is what gives it persistent memory and the ability to act, not just chat. Anthropic's free course walks you through the basics in about 20 minutes. Worth doing before Step 2 — it makes everything below click into place.
Step 2 — Download Dex
Go to heydex.ai or directly to Dave's GitHub repository. You don't need to know what GitHub is — think of it as a library where Dave stores his files. You're just downloading a folder. Click the green Code button → Download ZIP → unzip it somewhere on your machine (Documents is fine).
Step 3 — Add Dex to Cowork
In Claude Cowork, point it at the Dex folder you just downloaded. That's your last step. Claude reads through the files, understands the system, and from that point on it has access to all 65+ skills, your preferences, and your project context — in every session. When you open your first conversation, Dex walks you through a short setup: your role, your clients, how you work. Five minutes, and then it's ready.
This Isn't Just for You
One thing worth saying clearly: this isn't a solo productivity trick. The same shift that happens for an individual happens at team level. When your team stops bottlenecking on what they know how to build — and starts directing AI that can build it for them — the output ceiling lifts for the whole organisation.
I run half-day sessions where everyone leaves with a working setup, a personalised skill library, and a clear plan for what they're going to build first. Book a call if that's useful.
The Point
The gap between using AI and building with AI is a workflow gap, not a capability gap. The models are the same. The difference is whether you're accessing them through a chat window that forgets you every time — or through an environment built for you to build.
The ceiling isn't the AI. It's the interface.
"Remove the interface. See what you build."
— Claire Byrne
_1775536111226-BMn_eQIK.png)
