Claude Chat, Code, or Cowork

You have thirty PDPs to audit against the 10-element rubric. Do you:

Open Chat, paste the rubric, paste a product page, paste the next product page, paste the next product page, until you lose your mind?
Build a Skill in Code that runs the audit automatically and produces a scored CSV?
Hand the whole job to Cowork with a one-line instruction and walk away to do something else?

The right answer is the third one. The second one works too if you're going to run this every week. And the first one is what most marketing teams actually do, because Chat is all they know.

This chapter is the decision framework. Five minutes of reading. Years of avoided friction.

The Three Surfaces Again, With the Job Each Does Best

We covered the three Claude surfaces in Chapter 1. Here's the same map, this time with the job each surface does better than the others.

Chat is best at conversation. Strategy work, exploration, drafting, analysis you want to riff on with Claude before committing. You're in the loop on every turn. The output lives in the conversation thread.
Cowork is best at execution. Multi-step work that produces a finished deliverable. You set the outcome, Cowork figures out the steps and runs them. The output lands in a file or in an app through MCP.
Code is best at integration. Direct file system work, custom scripts, technical pipelines, anything that needs to run on a schedule without a human present. Your engineering team or a developer works in Code. The Skills they build then run inside Chat or Cowork.

Three surfaces, three jobs. The trick is matching the job in front of you to the right one.

The Decision Question

You can route 90% of marketing work correctly with a single question.

"Will I be in the loop on every step, or do I want to come back to a finished thing?"

If you want to be in the loop, choose Chat. If you want to come back to a finished thing, choose Cowork.

That's it. That's the framework.

The other questions are edge cases.

"Will I do this exact thing repeatedly?" That wants to be a Skill, so it runs identically every time.
"Does this need to run while I'm asleep or touch our infrastructure?" That's an engineering ask. Code.
"Am I just thinking out loud?" Definitely Chat.

But the core decision lives in that first question. In the loop, or finished thing.

When to Choose Chat

Chat is the right tool more often than people realize. The mistake isn't using Chat too much. The mistake is using Chat for work that should be a Skill or a Cowork pipeline.

Use Chat for the work where the thinking matters more than the output.

Strategy conversations. You're working through positioning, brand voice, or campaign concepts. The value is in the back-and-forth, not the document at the end. You'll refine for an hour and the final version is in your head, not the chat window.
Exploration. "What are five ways we could think about this new SKU?" You want options, not a finished plan. Conversation, not execution.
Drafting from scratch when the brief is unclear. You don't know exactly what you want yet. You'll figure it out by writing and reacting.
Analysis on a one-off question. "Look at this customer survey data and tell me what stands out." One conversation. One answer. Done.
Learning and reference. Asking Claude how something works. Asking for an explanation of a marketing concept or a tool you're considering.

The thing all of these have in common is that you're using Claude to think, not to produce. Chat is where strategy lives. Cowork is where production lives.

When to Choose Cowork

Cowork is the right tool whenever the work has multiple steps and you don't want to drive each one.

Use Cowork for the work where the output matters more than the conversation.

Audits at scale. Run the 10-element PDP rubric across a hundred product pages. Output a scored report.
Content production with a defined process. Topic in, finished blog post out, with formatting and citations and Surfer optimization handled along the way.
Research synthesis. Drop fifteen interview transcripts in. Get back a synthesized persona document.
Recurring reports. Pull weekly Shopify, GA, and Klaviyo data. Format the digest. Drop it in the team's Notion page.
Multi-step workflows that touch several tools. Take a client meeting transcript, update Linear with new tickets, draft a client follow-up email, and post a summary to Slack. All in one Cowork run.

The thing all of these have in common is that the inputs are knowable in advance, the steps are repeatable, and you want a finished deliverable instead of a conversation.

Cowork shines when you can describe the job and walk away. That's the test. If you can write a one-paragraph brief and trust the output, Cowork is the right surface.

When Code Comes Into the Picture

Most marketers won't touch Code directly. But it's worth knowing what Code is for so you know when to call your engineering team or your AI Ops partner.

Code is the right tool for the kinds of work that sit underneath Chat and Cowork.

Building custom Skills your team will use repeatedly in Chat or Cowork.
Connecting Claude to systems that don't have an MCP server yet.
Running automated pipelines on a schedule (overnight reports, daily monitoring).
Anything that requires direct file system work across many files.

For a marketing team, Code shows up in one of two places. Either your developer builds custom Skills that you use in Chat and Cowork, or you bring in a partner to build the infrastructure for you. Either way, you're the consumer of what Code produces, not the author.

The signal that you've hit a Code problem is usually one of these.

You've described the same workflow three times and you keep wanting to run it the same way. That wants to be a Skill.
You want something to happen automatically without anyone opening Claude. That needs Code.
You want Claude to do something to fifty files at once. That needs Code.

When you hit those signals, route to engineering. Don't try to muscle it through Chat.

The Skills Wrinkle

Skills sit across all three surfaces, which confuses people. Here's the simple version.

A Skill is a folder of instructions Claude follows when triggered. It doesn't belong to Chat or Cowork or Code specifically. It runs wherever you invoke it.

That means a Skill built in Code can be used in Chat ("run the brief-generator skill on this topic") or run autonomously in Cowork ("build a brief from this topic using the brief-generator skill, then post it to Notion").

The practical implication is that Skills are how you encode your team's procedures once and use them everywhere. The mode decision (Chat or Cowork) is about whether you want a conversation or a finished deliverable.

The Skill decision is about whether this procedure is repeatable enough to bottle up.

Most marketing teams have ten to twenty procedures that get run weekly. Each one should be a Skill. Then the mode decision becomes simpler. Use the PDP audit Skill in Chat if you're working through one page interactively. Use the same Skill in Cowork if you have a hundred pages to score.

How the Decision Plays Out at Tymoo

Here's how this maps to our actual work, on real days.

Brand strategy session with a client. Chat. We're exploring positioning, talking through customer language, riffing on competitive responses. The conversation is the value. Cowork would be wrong here because there's no defined output to chase.
Content brief from a topic. Cowork. We have a brief-generator Skill that runs SEMRush keyword research, drafts the 5-section outline, generates the hook angles, and saves to Airtable. One topic in, finished brief out. Four minutes.
Drafting a tricky client email about a delayed deliverable. Chat. We want to feel the tone, see two or three options, adjust the framing. The judgment is the work. Cowork would automate the part we actually need to be careful about.
PDP audit across a client's catalog. Cowork. We have a PDP audit Skill that scores each page against the 10-element rubric. Cowork runs it across the catalog and produces a scored CSV.
One-off question about whether a campaign concept might run into FDA guidance issues. Chat. We want to talk it through, ask follow-ups, push on the reasoning. Cowork would give us a confident answer with no thinking-out-loud, which is the wrong shape for a judgment call.
Building a new Skill from scratch. Code, usually with help from a developer. Or we use the skill-creator Skill inside Claude itself, which is also a Code workflow under the hood.

The pattern is consistent. Work that's exploratory or strategic stays in Chat. Work that's defined and repeatable goes to Cowork, often through a Skill. Work that builds the infrastructure happens in Code, usually invisibly.

The Wrong-Mode Tax

Every marketing team pays a wrong-mode tax. It's the time you waste when you pick Chat for something that should be Cowork, or when you reach for Cowork for something that should be a conversation.

The Chat-when-you-should-Cowork tax shows up in three ways.

Re-prompting the same workflow ten times instead of building it once.
Copying outputs into other tools by hand.
Doing the boring orchestration work Cowork would handle for free.

The Cowork-when-you-should-Chat tax shows up differently.

Getting a finished deliverable when you wanted a thinking partner.
Skipping the strategic judgment by automating a decision you should have made yourself.
Trusting Cowork's output on a question that actually needed real exploration.

Both taxes are invisible until you start paying attention to which surface you defaulted to. The cure is the decision question. In the loop, or finished thing.

What's Next

Now that you know which mode to pick, the next chapter is the highest-leverage one in the whole guidebook. Five Cowork workflows every CPG marketing team should build. Concrete, ready to copy, with the inputs and outputs spelled out for each. This is where the guidebook earns its bookmark.