I help sustainability professionals build AI tools that save a week out of every month.

Hey {{first_name|there}},

Today I'm sharing the 5 things I'd get right first if I wanted to be good at using Claude as a sustainability manager.

These 5 give you 80% of the result for 20% of the effort. The 80/20 rule is almost unfair. Make the most of it.

The rule: I almost always start with Sonnet, and move to Opus only when the task is hard enough that I want a second opinion.

Example, regulatory analysis. Upload a new EU regulation and ask Claude to extract requirements, deadlines and responsible functions into a table. That's Sonnet.

Now give it the regulation, your existing procedures, your operations, and past legal interpretations, and ask: "Based on all of this, where are our biggest compliance risks, what's still unclear, and what should we prioritize over the next 12 months?" That's when I'd reach for Opus.

Same topic. Different level of difficulty.

Next: the setting that can quietly work against you.

We work on a hundred projects at once, all at different stages of maturity. Most of the time, Memory is great.

But when it gets the context wrong, it gets it wrong. No way around it.

You might be running an EcoVadis assessment at global level while also supporting one site with its own assessment. You might be discussing several decarbonization projects for your Scope 1 and 2 report, but not every one of them is approved yet.

You don't want an unapproved project showing up in your draft report. And you definitely don't want Claude assuming it while you're reviewing a Sustainalytics assessment.

That's where Memory turns from helpful to annoying. The good news: it's easy to turn off.

For work where context accuracy really matters, I'd rather hand Claude the right context deliberately than hope it remembers correctly.

Next: turning your own judgment into something Claude can reuse.

Example, audit findings. Every time you review an environmental audit, you might tell Claude: classify the finding, determine severity, identify the relevant requirement, draft it in this structure, suggest a corrective action, don't assume when evidence is missing. Fifteen instructions before Claude even starts.

That's a Skill. Put your classification logic, your expected structure and a few examples of good findings into one file.

Next audit: upload the evidence, use the Skill, get the output in the same structure every time.

You've stopped prompting Claude. You've started giving it an SOP.

Now: getting Claude to reach your evidence directly.

Example, ESG rating prep. You're preparing for an EcoVadis assessment. The evidence is everywhere: policies in Google Drive, previous submissions in another folder, energy data in spreadsheets, supplier documentation, meeting notes, maybe emails with decisions buried in them.

Without connectors, your job becomes: find, download, upload, ask Claude, find another document, upload again.

With the right connectors, Claude reaches the information where it already lives. That's the real benefit.

Last one, and it's the one that changes how you think about all of this.

Example, quarterly environmental reporting. Every quarter, 10 sites send you their data. You check submissions, validate energy, emissions, water and waste figures, spot unusual changes, chase missing information, consolidate the numbers, update Excel, build the dashboard, prepare the management presentation.

You can do pieces of this in Chat. Upload the Excel file, explain your validation rules, analyze it, build a presentation. Three months later, you start again.

Cowork changes the question from "Can Claude help me create this quarter's report?" to "Can I build a reporting process that Claude helps me run every quarter?"

Now your instructions, validation rules, site information, historical data, templates, Skills and expected outputs sit around the process itself. Next quarter, you bring in the new data and run it again.

Which brings me back to the question I ask before starting anything new: will I do this again? Will I need this context again? If both answers are yes, I stop chatting and start building around it.

That's it for this issue. Hit reply and tell me what you're building. I read every one.

— Zyad