I help sustainability professionals build AI tools that save a week out of every month.
Hey {{first_name|there}},
Last week, I said it plainly: if there's one thing to take from this newsletter, it's to pay attention to Cowork and how it's becoming the future of knowledge work, sustainability work included.
Today I want to answer the question that comes right after that.
Because here's what usually happens.
Someone gets excited about Cowork
They open it up and give it a prompt
They get the same output as chat, if not worse
Two things are going on here.
First, you're using it wrong. It's not chat.
Second, you might be giving it the wrong task.
Claude is the umbrella. Under it sit three ways to work: chat, Cowork, and code. And the only question that matters right now is which one to use when.
Chat is where you talk. You ask, it answers, you steer.
Cowork is where you delegate. You set the goal, it does the work and checks in.
Code is where you build. You describe a tool, it builds the software that runs it.
That's the whole thing. Each step up gives you more, but it asks more of you too: more features to learn, more setup to invest in.
Master Claude Cowork in 9 Hours. Get 9+ Hours Back, Every Week.
You've seen what Cowork can do. The training is where we build your system together, on your real work: materiality, reporting, audit prep, until it runs without you. Sooner or later this lands on your desk. The waitlist decides whether you're the one who already knows how to run it.
No payment, no commitment. The waitlist sees the price before anyone else.
Chat: the technology we all know by now
You already know this one. It's the conversation window. You ask a question, you get a response, you ask the next one.
What makes something a Chat job is simple: the whole thing lives and dies in that conversation. One or two exchanges, maybe an attachment, and you're done. You won't come back to it.
In our world, that's a lot of genuinely useful work:
Interpreting a single ESRS datapoint you're unsure about
Checking one green claim against the EU Empowering Consumers Directive before it goes public
Drafting a supplier email, or tightening a paragraph for a report
Thinking out loud: "help me pressure-test whether this belongs in Cat. 1 or Cat. 7 of Scope 3"
Notice what these have in common. You're in the loop for every step. You copy, you paste, you ask, it answers.
How you work with it: you steer every message.
What to expect back: an answer, a draft, a second opinion.
The skill you need the most: good prompting.
Cowork: when you need the job done
Here the logic changes. It's not chat back and forth anymore, not a Q&A. It's an agent you assign a goal to, and it works toward it.
With that comes a superpower and a curse.
Give it a vague goal with no system behind it, and you'll spend more time figuring out what went wrong than celebrating what worked.
Tasks I personally run with Cowork:
I stopped jumping across dozens of websites and newsletters for my weekly sustainability updates. I built a skill that does exactly this. The Monday Brief, downloaded by over 1,000 sustainability professionals.
I ran a full materiality assessment using Cowork, another skill with a plugin in the works.
I screened products for EUDR compliance across multiple data sources: product SKUs, purchasing data, the regulation annex. A multi-step sustainability problem, solved via Cowork.
The idea is simple: one goal, a clear SOP for Claude, task done.
Is this easy? No. It means understanding how the goal can be achieved, creating a clear roadmap for the agent, providing the right data, and building in the right checks. Only then do you have a system that actually delivers.
How you work with it: you set the goal, it executes and checks in.
What to expect back: a finished deliverable. A report, an assessment, a filled questionnaire, a dashboard built from your data.
The skill you need: learning to direct an agent and build a system around it. No code. Just control.
When to jump to Code
The line between Cowork and Code is worth getting right, because it's where people over- or under-invest.
If the process is still moving, still being figured out, stay in Cowork.
Jump to Code when a proven solution already exists but doesn't fit your process, your problem, or your way of working, and you want to build it your way.
Two examples from our field:
EUDR is a Cowork problem. Still evolving, many steps that benefit from AI judgment, not ready to freeze into software.
Carbon accounting is a Code problem. Proven methodology, most tools are generic and not designed for your setup. Build the one that fits.
In the coming weeks we are going to dive deeper into how to select the right tool for your problem, with real examples. But this is the framework you can apply right now, and I'd genuinely love to hear what you think of it.
FREE VALUE
Do you want to test cowork yourself?
Check this video out, you can build yourself a deliverable that save you at least an hour or two using cowork. Test it out yourself and let me know if you have found it helpful.
That's it for this issue. Hit reply and tell me what you're building. I read every one.
— Zyad