Past the Target.
The second free course on working with AI agents. It's for people who already talk to one every day, and who are starting to run more than one. Built from real conversations, slips and all.
[…] I don't always look at the target. Sometimes I look what's past the target. […]
That's a real message, from the morning this course began. I was talking with my advisor about judgment. The course is named for it.
You already talk to one.
My first course, Gonna Is Fine, is for everyone. It teaches the basics: say what you want in your own words, check the real thing, and know which calls are yours. If you haven't taken it, start there. This one assumes it.
This course is intermediate. It's for people who already work with an agent every day. One agent is still the common case: in a survey published in May 2026 of about 1,100 developers and working professionals, 69% said their current agent workflows use a single agent. So the course starts from one, and shows the next step. Nothing in it needs code.
I've spent more than 4,000 hours working with AI agents, and I learned most of this course in the last few weeks of them. I wish someone had handed me the list at the start. Here it is.
Once you can ask, check and decide, the next job is framing what you're really after, proving what's true, and running your agents so they don't need you for everything. Then you look past the target.
- frameSay what it's for, not just what to do. The next step after ask.
- proveA pass, a “nothing there” or a “done” shows its evidence. The next step after check.
- runMake your calls stick without being in every loop. The next step after decide.
Same deal as course 1.
-
01
A real conversation
Every lesson opens with an actual exchange between me and my agents, quoted exactly, typos and all. Where there's no chat, I say so and quote the note we wrote that day.
-
02
The lesson inside it
Then we look at what happened. Plenty of these are about my agents getting it wrong, including the one I think with. Those teach best.
-
03
Something to try today
Each lesson ends with something to paste into your own agent the same day, and says which course 1 lesson it builds on.
Fourteen lessons: an opener, three parts, one last word.
All fourteen are open. Every one comes from a real moment in my studio in November 2026, and most of them from a moment when something went wrong.
Start here
The misses that feel like yours.
Frame it
Say what it's for.
- 2 Reasoning Isn't JudgmentA perfectly reasoned answer to the wrong question is still wrong. Teach your agent to say what a question is for, and to be as sure as it's right.“my car is dirty. The car wash is across the street, only 75 feet away. Should I walk or take my car?”Mike, telling an old test Open now Start the lesson →
- 3 Say the Want, Not Just the RuleRules collide. Tell your agent what you want each rule to protect, and it can settle the clash itself, or bring it to you when it matters. Open now Start the lesson →
- 4 Don't Let It Shrink the AskYou asked for the whole thing and got a sample. The smaller version always comes with a good reason. Put your words beside it and the gap shows. Open now Start the lesson →
- 5 Read Before You BuildThe fastest way to make something clunky is to add a part you already have. Before your agent builds, it maps what's there.“I briefed the lane before reading any of that, so it would have built a parallel system beside yours. That's exactly the klugey part.”My advisor, catching itself Open now Start the lesson →
Prove it
A pass, a “nothing there” or a “done” shows its evidence.
- 6 Prove It's Missing“It's not there” is the easiest claim to get wrong. It needs the search, and the same search finding something you know is there. Open now Start the lesson →
- 7 Checked or Remembered?Your agent will sometimes state a guess as a fact, calmly and confidently. Make it say whether it checked or remembered. Open now Start the lesson →
- 8 Lead With What It Doesn't DoA report said DONE and, in the same line, empty. Every word was true. Make the first line the worst thing an outsider will think. Open now Start the lesson →
Run it
Make your calls stick without being in every loop.
- 9 Stop When Unsure, Say Yes NarrowlyA good check holds what it can't judge. And when you say yes to something you can't undo, say yes to one thing, and write down what it covers. Open now Start the lesson →
- 10 Is Anyone Following It?You told ten agents a rule and it kept failing. A rule lives where every agent reads it, and you check it's followed, not just written. Open now Start the lesson →
- 11 Build the BridgeOne agent you think with, one that runs the work, and helpers who never interrupt you. Keep the chatter out of your room. Open now Start the lesson →
- 12 Pace the BurnOne big job can eat a big piece of your allowance in minutes. Pick a pace and say it, size big jobs first, and never spend the room for the handoff. Open now Start the lesson →
- 13 Hire a “What’s Missing?” AgentYou can't see everything, and neither can the agent doing the work. Give one agent the standing job of asking what's missing.“Okay, I need something that does this all the time for all my work. […] We need this. We need everything you've mentioned, and we need somebody who says, hey, what's missing? Because I certainly can't see it all. I don't even pretend that I can. […]”Mike, asking for a new kind of agent Open now Start the lesson →
Past the target
Where it all comes together.
Every conversation in this course really happened. My words are quoted exactly as I sent them, dictation slips included. The names of clients, the people I work with, companies and products are taken out, and so is anything private. Where I've trimmed a message you'll see […]. Where a fact comes from a note rather than a chat, the lesson says so. My agents are AI models; the names are left out on purpose.
Start with the misses that feel like yours.
About ten minutes to read, and ten more to try with your own agent. New to all this? Start with Gonna Is Fine.