In partnership with
It's a slower news day today, so instead of highlighting something marginal just to fill space, we're going deeper on yesterday's Skills announcement—specifically, what it actually means to become an AI tutor. How do you teach an AI to do something well?
As we often say, AI skills are management skills.
You manage AI almost exactly the way you would manage a junior employee. Lots of context, clear definition of deliverables, and plenty of help understanding the process. When a new hire joins your team, you don't just hand them a desk and say, "Figure it out." You onboard them. You explain the workflow, show them where to find information, clarify what good output looks like, and point them toward the right people when they get stuck.
This is exactly the same as how you teach AI a new skill.
But let’s dig a bit deeper and see how Anthropic itself is doing it. If you take a look at their company newsletter writer, FAQ answering agent, or “Progress-Plans-Problems” reporting skill, a consistent architecture emerges.
Here are five practices worth stealing:
1. Start by explaining the assignment, not the steps. Before jumping into "how," make sure the AI knows what it's making and why it matters. The 3P update skill doesn't assume you know what "Progress, Plans, Problems" means—it defines it upfront, explains who reads it (executives, teammates), and sets the constraint (readable in 30-60 seconds). Think of it like briefing a freelancer: context first, task list second.
2. Don't upload information. Point to where it lives. The newsletter skill tells the AI exactly where to hunt for content, assuming it has access to those tools: "check Slack for team updates, look in Drive for the latest deck, scan the calendar for big meetings." This is completely different from Custom GPTs, where you'd upload a bunch of PDFs and hope for the best. Skills assume Claude can go find things—they just need to know where to look.
3. Show the format, don't just describe it. All three skills include actual examples of what a good deliverable looks like. The 3P skill shows the exact template: [emoji] [Team Name] (Dates) followed by three crisp sections. The FAQ skill demonstrates the question-answer structure. When the format is ambiguous, execution gets creative in ways you don't want. The same goes for briefing your team: "keep it short" means nothing, but "three bullet points, one sentence each" is crystal clear.
4. Tell it what to ignore, not just what to focus on. Each skill has explicit boundaries. The newsletter skill says "company-wide impact only, skip team-specific details." The FAQ skill says "questions that affect most employees, not niche issues." Without these guardrails, Claude (like a well-meaning intern) will try to include everything and optimize for volume instead of relevance. Negative space matters.
5. Break it into steps, even obvious ones. The 3P skill walks through: clarify scope, gather info, draft, review. The newsletter skill suggests organizing into thematic sections before writing. This isn't micromanaging; it's offloading the mental work of figuring out "what comes next." When the process is mapped out, people (and AIs) can execute without constantly asking, "Am I doing this right?"
The real insight here is that Skills don't just make AI better at tasks—they force you to articulate how work actually gets done. Most teams can't clearly explain their own workflows because they've never had to. Skills make implicit knowledge explicit.
Even if you don't plan to implement AI skills today, understanding them is extremely beneficial because it helps you become a better manager even without AI. The discipline of writing, a skill (defining the task, specifying the sources, showing the format, setting the boundaries), is identical to the discipline of onboarding someone well.
If you can't explain the process to Claude, you probably can't explain it to your new hire either.
— Torsten & Peter
We’ve just published a new workshop for our Pro members about using OpenAI’s Agent Builder.
In the video walkthrough, we build a sales enablement agent that taps into your company’s knowledge base, collects all the product information and case studies relevant to a lead, and even proposes follow-up questions your sales rep can ask on the discovery call.
Companies routinely pay thousands of dollars for such sales enablement tools. We show you how to build it in 20 minutes.
Click here to upgrade to Pro and get access to all existing and future video workshops, dozens of prompt templates, resources, and more.
Field Note: The 2×2 That Saves Millions
We once worked with a team that picked AI projects the old-fashioned way — whoever had the loudest idea in the meeting. Six pilots later, they had zero wins and a CFO ready to pull the plug.
So we brought in one grid that changes everything: the Impact–Effort Matrix.
Plot every idea on two axes:
Impact — time saved, revenue relevance, executive visibility.
Effort — input quality, integration pain, learning curve.
Then act with discipline:
High Impact / Low Effort → pilot now.
High Impact / High Effort → design for scale.
Low Impact / Low Effort → automate quietly.
Low Impact / High Effort → delete.
That single chart kills hype, saves millions, and earns you the right to scale. AI ROI isn’t about doing more — it’s about choosing better.
→ Inside Pro: Module 2 — Impact–Effort Matrix templates and pilot-scoring guide. Click here to upgrade.
Find your next winning ad creative in seconds with AI
Most AI tools promise you thousands of ads at the click of a button. But do you really need more ads—or just better ones?
Kojo helps you cut through the noise. We analyze your paid social data to uncover the ideas with the highest chance of success. Then, our AI predicts which concepts will perform best, so you don’t waste budget testing what won’t work.
Instead of drowning in endless variations, Kojo sends your best idea straight to a real human creator who makes it engaging, authentic, and ready to win on social. The entire process takes less than 20 seconds, giving you certainty before you spend and better performance without the waste.
Why gamble on guesswork or settle for AI spam when you can launch ads proven to work, made by people, and backed by data?
