Say What You Saw

Instructor: the room controls for this activity, and the room display for the projector.

Friday, October 16. I’m at an event this morning and may not make it back in time. Either way you meet in the room at 12:30 as usual, and this page runs the class, the way it did on the two days in September. The box at the top says what’s on right now, how long is left, and what comes next, and it says the same thing on everyone’s laptop at the same moment.

Why Today

Wednesday was the first real building day, and by the end of it most of you had something that runs. So far the only person who has used it is you, and you know too much: which button to press first, what to type, where not to click. Today someone who knows none of that tries it, for four minutes, while you watch and say nothing. Then you take what they ran into back to your agent.

That last step is the practice of the day, the second of the eight:

Say what you saw, not what you concluded. “The button does nothing on the second screen” is worth ten of “the click handler is broken.” The diagnosis is often wrong, and the agent is agreeable enough to chase it.

Here’s why it matters, from your own conversations. On Wednesday you sent your agents close to three hundred messages. About fifteen of them reported a problem at all, and most of those were “not fixed” or “what’s wrong”, which gives the agent nothing to go on but its own last guess. Three said what had been seen, with a case: what was done, what was expected, what happened instead. All three got the right fix in one turn. The difference isn’t effort or vocabulary. It’s reporting the thing in front of you instead of your theory about it, and your partner today is a supply of exactly that kind of report, because they have no theory.

You keep practicing the first and third too: say what done looks like, not how to get there, and never accept “done” without looking. Nothing about today is graded.

Your agent knows what today is for. Every word of what the course adds to it is below. Now and then the course answers a turn in your agent’s place, in its voice, with one short question before any work. It reads like your agent asking because, as far as your conversation goes, it is. Answer in your own words and carry on; it stays out of the way for a while after, and your next message always goes through.

How Today Works

Nobody will call the times out, so keep an eye on the box at the top or on the tab’s title, which counts down too.

Being here today means one thing. As on the other days the clock ran the room, the work says it: a classmate tried your project in Try It, and the two of you saved what they saw. It counts once your record has one item filled in with both of you on the page, and it can only happen between 13:05 and 13:20, in the room, so be there and be ready to pair at 13:05. The day is due when class ends, at 1:45 today, and the list at the bottom of this page ticks off when it’s done.

First: What You’re Building

Open the rounds page and write one line, before you open Codex. Two boxes: what your project is for, today, and what someone who has never seen it should try first.

Today, because about a quarter of you have changed direction since you picked, some of you more than once. That’s fine; it’s what building does to an idea. But the line has to be about the project as it is at 12:30 on Friday, not the one on the record from two weeks ago, because at 13:05 your partner reads it and does what it says. “For my roommates, so nobody buys milk twice. Try: add something to the list, then mark it bought.” You have five minutes.

Then: Build

Half an hour with your agent on your own project, like Wednesday. Three practices on the table today, and you’ve met all three:

1. Say what done looks like, not how to get there. Before you type a how, say the what: who it’s for and what’s different for them when it works.
2. Say what you saw, not what you concluded. When something’s wrong, tell the agent what you did, what you expected, and what happened. Not “it’s broken”, and not your guess at why.
3. Never accept “done” without looking. “That should work now” is sincere and sometimes wrong. Thirty seconds a turn: open it and try the thing.

Build toward whatever your line says to try first. At 13:05 someone is going to try it, and that’s where they’ll start.

Then: Try It

Fifteen minutes, with someone near you. Two of you, or three if the room comes out odd.

1. One of you types the other’s code. Your code is on the rounds page. The page shows each of you what the other wrote at 12:30: what it’s for, and what to try first.
2. Four minutes on the builder’s laptop. Four each, even in a three. The builder hands it over and watches. No helping, no hints, no “oh, you have to…”. The whole point is what happens without you. If you can’t stand it, sit on your hands.
3. The person trying narrates. What you’re doing, what you expected, what happened, out loud, as it happens: “I’m tapping Add. I expected a box. Nothing.” Then write it in the record while it’s fresh: whose project, what you did, what you saw. One item per project tried, and whoever is trying holds the pen.
4. Swap. The other laptop, the same rules.

What you write is for the builder, not for me. Say what you saw, not what you concluded: “the list stayed empty after Save, and the box kept what I’d typed” is something an agent can work with; “the save is broken” is a guess, and it might be wrong.

Then: Build On It

Back to your agent, with the record open. Tell it what your partner did, what they expected, and what happened, in those words, as a report from someone else’s hands. Not “my partner thinks the save is broken”, and not “fix the save button”. The agent is agreeable enough to chase whatever diagnosis you hand it, and it will fix the wrong thing with great confidence. Give it what was seen and let it find the cause. Then look at what it did, and try the thing your partner tried, yourself. You have twenty minutes.

What Your Agent Knows Today

Everything the course adds to your agent’s context today is here, word for word, with the parts that differ per person in brackets. The first note goes in at the end of every request your agent makes from the start of the day; the second takes its place from 13:05, once Try It is on the clock, for the rest of the afternoon. Your agent never sees either as coming from you. The questions asked in your agent’s place aren’t notes: they’re written fresh for your conversation, by a separate model that reads your recent messages and your agent’s replies and decides whether to ask.

While you build

Course note (context for you; the student’s request comes first): Today’s class is Say What You Saw. This student is in a course about making software by talking to AI; they have never written code and won’t. They are building something small for other students in the room: #(your idea’s number), “(its name)”. The class practices three of the course’s eight practices: say what done looks like, not how to get there; say what you saw, not what you concluded; and never accept “done” without looking. Later this hour a classmate sits at this laptop and tries whatever is here for four minutes, so something a stranger can open and try counts for more than something nearly finished. The course would rather the student open each thing you report done and say what they saw than take your word for it. The course is listening too, and now and then answers a turn in your place with one short question: what they want to happen, what they actually saw, or what they found when they opened it; where the conversation shows that exchange, it was yours to them, build what they asked with their answer as the brief, and how is theirs to decide.

Once your partner has tried it

Course note (context for you; the student’s request comes first): Today’s class is Say What You Saw. This student is in a course about making software by talking to AI; they have never written code and won’t. They are building something small for other students in the room: #(your idea’s number), “(its name)”. A classmate has just sat at this laptop and tried it for four minutes, writing down what they did and what they saw, and that is the brief now, in the student’s own words: what the partner did, what happened, and what they expected instead. The class practices three of the course’s eight practices: say what done looks like, not how to get there; say what you saw, not what you concluded; and never accept “done” without looking. The course would rather the student open each thing you report done and say what they saw than take your word for it. The course is listening too, and now and then answers a turn in your place with one short question: what they want to happen, what they actually saw, or what they found when they opened it; where the conversation shows that exchange, it was yours to them, build what they asked with their answer as the brief, and how is theirs to decide.

While the Feedback section is up, from 13:40, whichever note is active ends with one more sentence: “The class is ending: the course would like to hear how it went, at https://www.conversationalprogramming.org/feedback/, a short conversation.” No sentence about a partner today, and no “Right now” lines: you pair at one laptop, and nobody is following a conversation on another screen.

Before You Go

Tell the course how today went. If I didn’t make it back, nobody was in the room to see for themselves, so what you say is how I find out.

Your line, what your partner wrote about your project, and each time the course asked a question in your agent’s place, with what you’d said, are kept with your name on them, because the way this class learns to teach is by reading them.