How It All Started
Most people who find Claude start in Chat. You open a browser tab, type a question, get a surprisingly good answer, and wonder what else it can do. That was me. I spent a good chunk of time using Claude Chat as my primary AI tool — including, for a while, using it to write code.
I'd paste in a chunk of HTML, describe what I wanted, iterate on the output, try things, break things, go back and forth. It worked — sort of. The code came out reasonable. But there was always friction: copying from the chat window into my editor, losing context mid-session, explaining the same project setup again after the conversation got too long. Chat is a conversation, not a workspace. I was using it like a workspace, and that mismatch showed.
The shift happened gradually. I started using Cowork for intensive builds, then eventually discovered Claude Code for in-IDE development. Chat didn't go away — it just found its actual home, which turns out to be most of the rest of my life outside of building the site.
Three tools, three relationships. Here's what each one actually is.
Claude Chat: The Life Layer
Once I stopped using Chat for serious coding work, something interesting happened: I started using it more, not less. It just shifted to a completely different category of questions.
Current events I want to understand better. History I vaguely remember but want to actually understand. Articles I don't have time to read fully — I paste them in and ask for the key takeaways and what I should think about. Vacation planning. Recipes when I'm standing in a grocery store aisle. Day-to-day questions that used to send me down a Google rabbit hole for 20 minutes.
Chat is the generalist layer. It runs in the background of daily life. It's the difference between having to figure something out and having a knowledgeable person you can just ask. And because it has no agenda — no ads, no SEO-optimized results trying to sell me something — the answers tend to be cleaner and more direct than most of what I'd find searching.
The mistake most people make is treating Chat as a utility for one specific thing: writing, or coding, or research. The better mental model is that it's a general-purpose intelligence layer you can bring to almost any question. I've had substantive conversations about the history of monetary policy, what a specific clause in a contract actually means, why a particular piece of music is structured the way it is, and whether the itinerary I built for a family trip to Colorado makes geographic sense. All of that is Chat.
Claude Cowork: The Builder's Room
Cowork is a different category of tool entirely. It's not for quick questions. It's for sitting down with a goal and not getting up until something ships.
The model is: you come in with a project, you give Claude access to your files and working context, and you build together in real time. The sessions are intensive. They require investment — especially upfront, building the reference document that gives Claude the full context of your project across sessions. But that investment compounds. Once the context system is in place, every session picks up exactly where the last one left off.
I've written at length about what Cowork actually produced — the full story is in I Rebuilt My Entire Site with Claude Cowork. The short version: in about three weeks of evening sessions, I went from a stock WordPress site to a fully custom static build with ten live apps, a 16-module Make.com automation, and 39 published blog posts. None of that was possible in the old model of copy-pasting from Chat.
A few of those apps are worth calling out specifically because they're things I actually use and that other people can try:
All of those were built with Cowork. Each one is a Make.com automation on the backend — a webhook triggers a multi-step scenario that calls Claude, formats the output, saves to Airtable, and delivers the result via email and a unique link. None of that existed before I started using Cowork as a serious building environment.
Claude Code: The New Frontier
Claude Code is what happened when Claude moved from the browser into the terminal and the IDE.
The difference feels structural. In Chat, you describe code and paste it around. In Cowork, Claude can see your files during a session. In Code, Claude lives inside VS Code. It can read any file in the repo, edit specific lines, run commands, check what changed, and move to the next thing — all without you switching windows or copying anything.
The feedback loop is fundamentally different when the tool lives where the work actually happens.
Here's a concrete example of what that looks like in practice. A few weeks ago I needed to add a security token to every Make.com webhook call across the site — a token guard to protect against unauthorized requests. Doing that manually would have meant opening 40+ HTML files, finding each fetch call, and adding the token parameter one by one. In Claude Code, I described what I needed, and it read every file that contained a webhook URL, identified each fetch call, added the token parameter, and verified the change. Across 40+ files. In one session.
That's not a task I would have tackled in Chat. It would have been too fragmented — too much context to maintain across too many files. Cowork could have done it but would have required careful session management. Code handled it the way you'd expect a competent developer to handle it: systematically, completely, with verification at each step.
More recently we've been using Code to build out the Reading Buddy app — the AI reading comprehension tool for elementary kids. The iteration cycle is fast. Describe a behavior, see the code change, reload the page, test it. When something breaks, Claude Code reads the error, traces it back to the source, and fixes it. When something works, we move on. There's no copy-paste friction between the AI and the actual file.
The other thing Code does well: it's honest about what it's doing. Every file edit is shown. Every command is visible. You can approve or reject each step. It doesn't make assumptions or go rogue. For someone who's not a professional developer, that transparency matters a lot — it means I'm always in the loop, not just trusting that something happened correctly.
When I Reach for Which One
After using all three tools regularly, the decision framework has gotten simple. It's almost never ambiguous anymore.
| Tool | When I open it | What it's good for |
|---|---|---|
| Claude Chat | Daily, throughout the day | Life questions, learning, article summaries, planning, recipes, anything not about the site |
| Claude Cowork | When starting a new project or feature | Intensive multi-session builds, app architecture, new Make.com scenarios, complex new features |
| Claude Code | When I'm already in VS Code building | Active development, file edits, bug fixes, cross-file changes, real-time iteration |
The clearest way to put it: Chat is the life layer. Cowork is the project sprint. Code is the development session. If I'm thinking about something, I open Chat. If I'm planning to build something, I might start in Cowork. If I'm actively building, I'm in Code.
There's some overlap, especially between Cowork and Code — both can handle serious development work. But Code has become my default for anything that touches actual files, because having Claude operate directly in the file system removes a layer of friction that adds up over time.
The thing I didn't expect: having three tools doesn't mean more complexity. It means each tool is used for exactly what it's good at, which makes all three feel more effective than any one of them did when I was trying to do everything in Chat.
What This Actually Means
I've been building with Make.com for over 18 months. I understand automation. I know how to think in modules and data flows. But before Cowork and Code, my ability to execute on technical ideas was constrained by the time it took to write and debug code — which, as a non-developer building in the evenings, is a real constraint.
That ceiling has shifted. Not disappeared — Claude Code still makes mistakes, still needs clear direction, still requires me to understand what I'm asking for. But the gap between having an idea and having a working implementation is dramatically smaller than it was a year ago.
The practical result: I've shipped more in the last few months than in the year before it. Not because I suddenly have more time — the day job and the family haven't gotten any less demanding. It's because the tools meet me at a higher level of abstraction than they used to.
If you've been living exclusively in Claude Chat and haven't tried Cowork or Code, I'd encourage you to. They're not incremental upgrades to the same experience. They're different tools for different modes of work — and once you have all three, you'll wonder how you were managing with just one.
Explore what I've built in the portfolio, try one of the apps, or read more on the blog. The whole site is the output of this system. Make of that what you will.