The Drip
Are we offloading too much of our thinking to AI? The question hit the top of Hacker News this week, and then the news gave it teeth. Meta is being sued by former employees who say an AI helped rank who got laid off. We read it as an accountability story. Companies have used stack-and-rank tools forever. The new part is that a human judgment got handed off, and now nobody will own the call. The tools have never been more able. The question is what we're handing them.
Inside The Bottle
We run AI all day, every day, so we're a decent test case for where AI-first tips into AI-only.
The boundary we keep coming back to is one Justin put on the table this week. Where is the line between what the AI thinks about and decides, and what you think about and decide?
What we hand off without hesitation: research, triage, scaffolding tasks, first-pass cuts on this very show. The toil. What stays human: decisions about people, the strategic calls, anything where the answer is a judgment rather than a synthesis.
Two practices from our own systems that make the boundary real:
Confirm current state first. AI only sees what it has access to. Justin's AI coworker once hit five-alarm-fire language about a sales opportunity that was actually fine. It was just missing the last conversation. His grounding rule now is to get current first, then decide what happens next. If a system can conclude the wrong thing that confidently, you don't want it deciding autonomously.
Count the human decisions. Kellan's daily assistant runs on a rule: "I'm an assistant, not a decision maker." Every report tallies how many decisions were made by a human versus autonomously, and if it's fewer than eight human decisions a day, something gets tweaked. The number is just a check. Getting to a decision in the most informed way possible is the most valuable thing AI does, more than building the thing or automating the process.
And apply one filter to anything you're about to hand off. How bad is it if this thing makes a mistake? Small thing you can survive, or big thing? Constrain the blast radius to match.
Lab Notes
| ■ | Justin's note: No more AI first drafts. Sit down and write for 15 minutes until your fingers can't type anymore, then hand it over: "clean this up — is my point clear?" It tightens the writing without changing your words. |
| ■ | Kellan's note: AI only knows what you feed it. Give it half the story and that's the narrative that plays out. When a recommendation feels off your moral compass, it's usually missing context, not judgment. |
What Stopped Our Scroll
|
