There are three ways to work with Claude now, but the names do a pretty bad job of telling you which one to use for what.
Chat. Cowork. Code. They sound like three flavors of the same thing (but they're not).
Pick the wrong one and you'll spend twenty minutes coaxing a chat window to do a job it structurally cannot do, or you'll fire up a full agent to answer a question you could have gotten in one line.
Today I'm walking you through the differences between the three, so you know exactly which interface to use for which tasks 👇
10x the context. Half the time.
Speak your prompts into ChatGPT or Claude and get detailed, paste-ready input that actually gives you useful output. Wispr Flow captures what you'd cut when typing. Free on Mac, Windows, and iPhone.
Are you asking, or are you doing?
This is the question you can use to filter.
You can think of asking as when you’re looking for an answer, a draft, an explanation, a second opinion, etc.
Doing is when you want Claude to get some work done for you:
Files created and saved.
Apps opened.
A spreadsheet filled in.
A report that runs every Monday whether you're at your desk or not.
Chat is for asking. Cowork and Code are for doing.
Claude Chat: the thinking surface
Chat is the one you already know. It's claude.ai in your browser or the chat tab in the desktop app. You type, it answers. It’s fast, reactive, and conversational.
Use Chat when the output is text. Stuff like:
Brainstorming a campaign.
Drafting an email.
Pressure-testing an argument before a meeting.
Summarizing a document you paste in.
Asking it to explain a concept three different ways until one lands.
This is your second brain for any single piece of content. It's also the only surface on the free plan (which is why it's most people's whole picture of Claude).
The limit is the thing that makes it fast. Chat doesn't reach into your computer, doesn't run jobs in the background, and forgets the moment you close the tab unless you've set up a Project to hold context.
It answers, and then it waits for you.
For a huge share of daily work, that's exactly right, and reaching for anything heavier is overkill.
Claude Cowork: the doing surface for everyone else
Cowork is where Claude stops talking and starts taking action.
(We wrote a beginner’s guide to Cowork a while back, you can check that out here as a good starting point)
It's an agent, which here means an AI that takes actions on its own instead of just replying.
Give it a goal and it makes a plan, breaks the job into steps, runs those steps on your actual computer, and hands you finished files.
It can see your screen, open apps, drive a browser, read and write to the folders you point it at, and connect to your work tools through connectors (plug-ins that link Claude to apps like your email, calendar, or Slack).
Use Cowork when you want the work done, not described. Some of what that looks like:
Turn a folder of messy invoices into one clean spreadsheet.
Pull this week's numbers from three places and build the report.
Go through my inbox, draft replies to the five that need them.
Take these notes and produce the slide deck.
The piece people underrate is scheduling. You can set Cowork on a recurring job (like a briefing every morning, a report every Friday, a file cleanup every night). It runs on its own and the results are waiting for you.
You can even kick a task off from your phone and come back to finished work on your desktop.
Cowork is built for people who don't write code. If your job is full of repetitive steps across files and apps, this is the surface that gives you the most back.
Claude Code: the doing surface for codebases
Code is the same engine as Cowork, but pointed at software/code.
Under the hood, Cowork and Code share the same agentic core. The difference is the terrain.
Code is built to live inside a codebase: read across many files at once, understand how a project fits together, and make changes that span the whole thing while keeping the architecture straight.
It runs as a command-line tool for developers and also has a tab in the desktop app.
Use Code when the task is a codebase. Building a feature, fixing a bug, refactoring, reviewing a pull request, wiring up a script.
For most readers of this newsletter, this is the one you'll open least, and that's fine.
The useful thing to know is simply the boundary: the moment a job is really about source code, Code is the surface with the right context for it. If you're not working in a codebase, you want Cowork.
The fastest way to internalize this is to run your actual tasks through it once.
Paste this into Claude, fill in the blank with a rough list of what you do in a normal week, and let it sort the list and explain the calls👇
I use Claude in three places: Chat (asking: fast answers, drafts, brainstorming, no access to my files), Cowork (doing: an agent that works on my files and apps and can run on a schedule), and Code (doing, but for working inside a software codebase).
Here are the recurring tasks in my week:
[list 8 to 12 things you regularly do, messy is fine]
Sort each task into Chat, Cowork, or Code. For each one, give me a one-line reason. Then flag any task I'm probably doing the slow way (something I handle manually that Cowork could run for me, ideally on a schedule) and tell me how I'd set it up.You'll usually find two or three jobs you've been grinding out by hand in a chat window that should have been running themselves in Cowork or Code.
You’re gonna want to see this live
On July 16th at 1PM ET, beehiiv is unveiling the next chapter for audience-led businesses.
For years, creators and brands have been forced to stitch together bloated stacks of tools just to publish content, grow an audience, and make money online.
Newsletters in one platform. Websites in another. Podcasts somewhere else. Analytics scattered everywhere.
beehiiv thinks there’s a better way. Now, they’re ready to show it off at their Summer Release Event.
Your roundup of the latest model releases and updates from the biggest AI labs.
Microsoft ships its own "Cowork" and a desktop agent. Copilot Cowork hit general availability on June 16, the agentic version of Copilot that runs longer multi-step tasks across your apps in the background, even with your device off. Alongside it Microsoft unveiled Scout, an always-on desktop agent that works on your files and drives a browser. Notable footnote for this edition: Anthropic's Claude is now a selectable model inside Copilot Chat. The conversational-tool-plus-agent-tool pattern is becoming the standard shape of these products. (A Guide to Cloud & AI)
OpenAI leans hard into agents doing the work. On June 25 OpenAI published "How agents are transforming work," and the headline number is the adoption curve: by June 2026, median agent usage was 56 times higher than in November 2025. ChatGPT's agent mode now plans and executes multi-step jobs (analyze three competitors and build a slide deck, that sort of thing) rather than just answering. The whole field is shifting weight from "ask" to "do." (OpenAI)
Google bets on running agents in parallel. Google's Antigravity 2.0 introduced a setup that takes a complex goal, breaks it apart, and spawns independent subagents that run at the same time, one stream coding while another generates assets. The frontier is several agents working at once while you're off doing something else. (TechCrunch)
Advertise with Build with AI
Get in front of an audience of professionals using AI day-to-day: founders, engineers, operators, and product builders.
Interested in advertising? Respond to this email for rates and details.
Until next time,
William Ryan
Editor-in-Chief @ Build with AI
PS: Follow me on X for daily updates and AI workflows.