Gå til indhold
Use casesLærOm mig
cleverest
Tilbage til The Broadcast

I built the perfect AI agent. I almost never opened it.

Obs. The Broadcast er skrevet på engelsk.

I had an AI agent pulling together a reading digest for me every morning. It worked perfectly. And yet... I almost never opened it.

The digest lived in a file in my Obsidian vault, which meant reading it required opening my laptop. Opening my laptop before my reading hour is how reading hours die. Something always grabs me first: a quick email, a small website tweak, one prompt I wanted to test.

So I moved the delivery to Slack. Now the digest pings my phone at 6:30 AM. I read it on the couch with coffee. Laptop stays shut until I'm actually starting the workday.

Watch the tutorial

Get the AI onboarding doc — paste it into your favorite AI assistant and it'll walk you through the build step by step.


What I'm reading

Stanford 2026 AI Index Report — Stanford HAI. Two interesting stats from this report: 88% of organisations now use AI somewhere. But AI agent deployment is still in single digits across nearly every business function. If you're already deploying agents for yourself or your team, you are not behind. You are years ahead of most organisations.

Anthropic Economic Index: Learning curves — Anthropic. Anthropic looked at 1M Claude conversations and found that experienced users get a 10% higher success rate. While there's no definitive answer as to why, signs are pointing to learned habits: knowing what kinds of tasks to bring and how to communicate what you want. AI competence compounds with experience, but if you want a shortcut to get there, the answer is training.

Want to bring AI training to your team? Let's talk. →

Sign of the Future: GPT-5.5 — Ethan Mollick / One Useful Thing. Anthropic's Opus 4.7 release was the first time I felt a negative change in my workflows. In some cases I changed the model to Sonnet and in others I tweaked the prompts. Most of the time, though, new releases don't deliver on the hype announcements. Mollick gives us a reality check on the GPT-5.5 launch. He puts it through its paces and finds that it is stronger on structured tasks (he generates an academic paper from raw data), but still weak on creative writing.


Slow note

This morning I did everything right. Coffee, Made to Stick, notebook open, all my tech in another room. I resisted the laptop. I'd built the routine for exactly this hour and it was working.

Then my partner called from the kitchen: your phone is ringing.

Just after 8 AM, which is unusually early. I couldn't find the phone. I ran around the house tracking the sound. Picked up out of breath. It was a client with a last-minute request. He laughed: did I have you running for the phone?

I'd built everything around protecting this hour, and the hour still got interrupted. The system worked fine. Sometimes the phone rings and you spring into action anyway.

Until next week,

Louise

Få den næste i din indbakke