After the first Beyond the Prompt cohort wrapped, I already knew the next problem I wanted to solve. I wanted agentic workflows to feel approachable. Less intimidating. A door, not a wall.

So I pointed people at Cowork. It reads as the gentler on-ramp. The less technical runway. The version of this you can walk into without feeling like you need a computer science degree.

Here's the part I sat with for all three weeks: I don't run my own life in Cowork. I run it in Code.

I was using one tool and promoting another. Not because the on-ramp was wrong. Because "approachable" had quietly become a branding decision instead of a capability one.

Then one of the participants migrated to Code on their own. I didn't tell them to. The job they were trying to do pulled them there.

That was the moment the framing cracked open for me.

There's a line people draw when they talk about AI tools, and about Claude specifically.

Claude Code is on one side. Developers, engineers, people who live in a terminal. Claude chat and Claude Cowork for everyone else: the chat interface, the desktop app, the thing you open when you want to ask a question or get a document drafted.

I want to challenge that line. Not because it never had any truth to it, but because it's collapsing. And when we don't notice it's collapsing, we choose tools based on who we think we are, instead of what the work actually needs.

When Code launched as a command-line tool, the developer framing made sense. Terminal fluency was a real prerequisite.

That changed in April. Anthropic rebuilt the Claude Code desktop app around parallel sessions, a visual sidebar, an integrated terminal, and a drag-and-drop workspace. You can run multiple agents on separate tasks, watch them work, and intervene. No command line required.

I built MVP Momentum in Code. It's a Chrome extension with a FastAPI backend that powers my new tab page: live inbox, daily priorities, links. I am not a developer. I don't write production code for a living. Code was the right tool because the job was software development: a real codebase, a local server that had to run, shell commands, browser testing, iterating on something that had to function as an actual piece of software. That's what Code is built for. Not file management. Cowork handles that just as well. The distinction is building software versus working with it.

That's the shift. Code is no longer gated by whether you know your way around a terminal. It's gated by whether the job requires what Code does.

Start with the part most people miss: Code and Cowork run on the same engine. Same models, same API, same underlying intelligence. What changes is how much of it you can reach, and the interface you reach it through.

Code hands you the whole machine. Your full file system. Agents that run on a schedule at 3am whether your laptop is open or shut. Scripting, pipelines, several agents working different problems at once. And now Dreaming, a research-preview feature where Claude reviews its own past sessions and curates its memory between runs, so it sharpens without anyone retraining it.

Cowork reaches that same intelligence through a chat window, and trades range for one standout strength: it can drive your screen. Computer Use lets Claude see what you see, move the mouse, click through apps, and do the things that only happen by operating software the way a person does. Code can't touch that. So when the work lives in a browser, Cowork isn't the easy option. It's the only one that works.

The split was never about who's technical enough. It's about what the work in front of you actually needs.

Here's where I break from my own tidy framing.

I don't use both. I use Code. When I want a fast one-off, I drop into the chat prompt for a minute and come back. But there is no clean handoff between Code and Cowork in how I actually work. The "OS layer here, work surface there" version is neater than the truth.

That isn't a knock on Cowork. It does things Code can't, browser control chief among them. It's that my work, building and running the systems behind my day, lives in Code. So that's where I am.

And here's the part nobody says out loud: the guidance itself is murky. Claude won't reliably tell you which one to reach for, or when. Ask it directly and the answer wobbles. That confusion isn't a you problem. It's a real gap, and it's most of why this divide feels harder than it should.

So the question was never "should I use both." It's "which one fits the work in front of me." That one has an answer. You just have to ask it on purpose.

Pick one recurring task you do every week.

Ask three questions:

Does this need to run when my computer is off?
Does it need to navigate a browser or control another app?
Does it need to process many files, connect to a codebase, or run on a schedule?

If the first or third: Code is probably the right tool, regardless of your job title.
If the second: Cowork's browser control is the capability you need.
If none of the above: either works. Choose based on output format. Interactive and shareable points to Cowork. File-based and automated points to Code.

The goal is to choose by the job, not by the title you happen to hold. That's the Job-First Test.

Cohort 2 opens in July. We spend three weeks building exactly this kind of thinking. Not just how to use AI, but how to architect a workflow around what the work actually requires.

We work in Cowork. We talk about Code. Most leave knowing which one is actually theirs.

If you've assumed Code wasn't for you because you're not a developer, that assumption is the thing to examine. The tool has moved. The mental model is the part that lags.

[Join Cohort 2: puhala.com/beyond-the-prompt]

Cheers to finding flow this week.

– Michael
Founder, The Drop In
& Author of 'Human Traits, a novel exploring humanity's relationship with AI'