Thanks for being here!

Every week, I'm sharing links, articles and resources I found interesting.

This week's readings are dominated by one theme: AI agents are going from concept to reality—fast.

Claude Opus 4.6 dropped with massive upgrades, Anthropic launched a plugin system for Cowork, and personal AI assistants are getting eerily capable. Meanwhile, Notion's custom agents are getting closer to general release, and proving they're not just a feature—they're a fundamentally different way to work.

Here's what caught my attention. 👇🏼

While you were debating if AI would take your job, other people started using it to print money. Seriously.

That's not hyperbole. People are literally using ChatGPT to write Etsy descriptions that convert 3x better. Claude to build entire SaaS products without coding. Midjourney to create designs clients pay thousands for.

The Hustle found 200+ ways regular humans are turning AI into income. Subscribe to The Hustle for the full guide and unlock daily business intel that's actually interesting.

Start making AI money

Anthropic released Claude Opus 4.6 this week, and it's a significant leap. The headline: a 1-million-token context window—the first for any Opus-class model—plus state-of-the-art performance across coding, planning, and agentic tasks.

What I find most interesting isn't just the benchmarks. It's the adaptive effort system.

The AI Corner published a hands-on breakdown of what actually changed in Opus 4.6 and how to get the most out of it. The key insight: stop treating all prompts the same. The model responds differently to ambiguous versus well-scoped work now. Front-load your context instead of drip-feeding it—the 1M token window can handle it.

🔗 Read the full guide

Speaking of, I wrote about this earlier this week because I think it matters for anyone living in Notion.

The difference between a tool and a collaborator comes down to how well it actually follows through on complicated requests. With the 1M token context window, you can now feed it entire research papers, financial data, or complex project documentation—and it remembers all of it while working.

My recommendation: start with your most annoying recurring task. The one that takes 45 minutes every week because it requires pulling data from multiple places. Give it the real messy data. That's your test case.

🔗 Read my full take

OpenClaw (formerly ClawdBot then MoltBot) has been a hot topic this week, and with good reason: it actually feels like an AI Agent that can work on its own.

Peter Yang went deep on OpenClaw this week, and it's the best walkthrough I've seen. OpenClaw is an open-source AI assistant that runs entirely locally on your Mac, stores everything in Markdown files, and—here's the wild part—can modify its own capabilities when you ask it to.

What makes Peter's breakdown stand out is the practical use cases. He set up a personal assistant called "Zoe" that manages Google Calendar, edits Docs and Sheets, replies via voice, sends personalized daily briefings, and generates weekly insight reports—all through Telegram.

This marks a huge evolution for practical AI workflows. Peter replaced dashboards and manual workflows with scheduled cron jobs set up through a five-minute conversation. It makes you wonder: what happens to standalone apps when users can just ask bots exactly what they need?

🔗 Watch the full video

I told you AI was taking over this week!

Anthropic launched plugin support for Cowork, their desktop AI workspace.

They're open-sourcing 11 plugins covering sales, legal, finance, marketing, product management, and more. The idea: define your process once—connect your CRM, encode your sales workflow, teach it your documentation standards—and Claude pulls from that context whenever it's relevant.

I think it’s notable because it mirrors what we're seeing with Notion's custom agents. The pattern is the same: encode your processes into AI, then let it execute consistently. The file-based architecture makes plugins easy to edit, share, and customize. If you're building workflows in any AI tool right now, this is worth exploring!

🔗 Read the full article

I wrote about this one because I keep getting the question. Short answer: no.

Custom agents will be a fundamental shift from on-demand AI (you ask, it answers) to autonomous execution (it watches for triggers and acts without you). The moment it clicked for me was automating client onboarding—when a proposal status changes to "Signed," the agent creates the project page, builds the dashboard, and populates tasks. I don't touch it.

Here's the thing most people miss: your SOPs are already half the work. If your team has documented playbooks for key workflows, those documents can be repurposed directly as custom agent instructions. The heavy lifting is already done.

🔗 Read the full article

On February 26th, I'm hosting a live session where we'll build a custom agent from scratch together. We'll cover the real difference between Notion AI and Custom Agents, the tips that separate agents that work from ones that sit unused, and a sneak peek at advanced features like Slack integrations and MCP connections.

If you've been curious about custom agents but haven't taken the plunge yet, this is a great place to start.

🔗 RSVP here

That's all for this week.

The through-line is pretty clear: AI agents are moving from experimental to actually useful.

More to come next week, but curious: any topics you’re interested in deep diving on? Could be Notion, productivity, consulting, or AI. Send me a reply!

See you in the next one,

Dave de Céspedes
Founder, Workcraft Labs
YouTube • LinkedIn • X • Threads • Instagram

ps. if you’ve noticed a different look, it’s because we’ve moved to Beehiiv!