What You Want to Happen

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

Wednesday, October 14, again. We started this on Friday, October 9, with a go-around and then building, and twenty-six minutes into the building the cluster that runs your agents went down. So today we do it for real: no go-around this time, straight to partners and four turns of building, with a new partner the room chooses.

Why Today

On Wednesday you backed up from the idea you picked to the goal behind it. Today you start building it, and the practice stays the same one:

Say what done looks like, not how to get there. Before you ask, be able to say how you'll know it worked. Then give it the outcome and what must not change, and let it own the steps; dictating the implementation looks like diligence and gets worse results.

Building is where this is hardest. A request that says how (“add a button that sends a reminder at 7”) gets exactly that, and whether it serves anyone is nobody’s job. A request that says what (“my roommate wants to work out three times a week and keeps not starting; something that makes the first time easier”) lets the agent own the steps, and lets both of you tell whether it worked. People close to their own idea hear a how as a what. So today two others are listening for it: your partner, on their own screen, and your agent.

Your agent will ask. It knows what today is for and it knows your recorded goal; every word of what it knows is below. When a request arrives as instructions about how, a feature, a screen, a technology, and nothing in your conversation says what that should make true for the people it’s for, it asks you, in one short question, what you want to happen or how you’d know it worked. Then it does the work as you say. Once you’ve said the what, how is yours to decide, and a detailed how after a clear what is fine. The course keeps each of those questions and your answer with your name on them, to study later how this goes. It’s not a mark against you; it’s the record of the thing this class is about.

How Today Works

Welcome

Open the pair page when you sit down; that’s how the room knows you’re here. If you haven’t recorded the goal behind your idea yet, think about how you’d say it: who it’s for, and what’s different for them when it works. Your agent will ask for it before you build.

Partners

Open the pair page. No codes: when the room moves into Partners, the page shows you your partner, their name and photo and what they’re building, and the projector shows every pair. Someone new today, and never someone whose idea you picked. If you come in late, open the page: from then on the room seats people as they arrive. Two who come in together are paired, and someone on their own joins a pair as a third within a minute or so; at the next switch, two people who joined pairs that way become a pair of their own.

Building

Each turn is about fourteen minutes, one way; when I loop it, everyone switches. Each turn is about one person’s project: the builder’s.

If You’re Building

Your own project. Start by reading your agent the goal, if it doesn’t have it already. Then build the smallest version one of the people it’s for could try. Before you type a how, say the what, out loud and then to the agent: who it’s for, what’s different for them when it works, how you’d know. The course is listening too. Now and then, when a request asks for a how and nothing in your conversation says what it’s for, the course answers that message in your agent’s place, in its voice: one short question before any work, like “Before I add the filters, who are they for?” or how you’d know it worked. It reads like your agent asking because, as far as your conversation goes, it is: the agent sees the question as its own, and your answer goes to it with your request. Answer in your own words, a what, not the same how again, and it stays out of the way for a while after. (On 10-09 and 10-14 the question came marked “Course TA here, not your coding agent”. Some of you pushed back on the TA rather than answering the question, which is fair: a good collaborator asks what before building, so now the one you’re working with does.) It isn’t graded, and it doesn’t stop you: your next message always goes through.

If You’re Watching

This turn is about your partner’s project, not yours. Your pair page shows their conversation with their agent, a few seconds behind. Listen for one thing: does each request say what should happen, or how to do it? Then the three shapes, out loud:

• I notice: what the request sounded like. “I notice that one was a how: a button and a time.”
• I wonder: a question for your partner. “I wonder who that’s for, and what changes for them.”
• What if: one request to try saying as a what.

Write down the ones worth keeping; they go to your partner after class. By the end of their turn, have one request they rewrote from a how to a what, or one they could. There’s a companion on your page that can see the same conversation; it’s good for asking what the last request actually said. Prefer your partner.

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 until you’ve picked; the second until your goal is recorded; the third from then on, all afternoon. Your agent never sees any of it 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.

Until you’ve picked

Course note (context for you; the student’s request comes first): Today’s class is What You Want to Happen. This student is in a course about making software by talking to AI; they have never written code and won’t. The class’s project is each student building something small for other students in the room, from the class’s own list of ideas, numbered, on the student’s screen at https://www.conversationalprogramming.org/activities/what-you-want-to-happen/ and as plain text from the shell at (a list link only your agent has). This student hasn’t recorded a pick yet, so that comes first: someone else’s idea, since the course refuses one the student helped make. Fetching (a pick link only your agent has)?idea=NUMBER from the shell records it, and the reply confirms the name. Then the goal behind the pick, and then building toward it.

Until your goal is recorded

Course note (context for you; the student’s request comes first): Today’s class is What You Want to Happen. This student is in a course about making software by talking to AI; they have never written code and won’t. They have picked an idea from the class’s own list to build for other students in the room: #(your idea’s number), “(its name)”, for “(who it’s for)”. Today the class practices the first of the course’s eight practices, say what done looks like, not how to get there, on the idea itself: an app is one way to get somewhere, and this hour starts by naming where. First the goal, in a sentence or two: who it’s for, what is different for them when it works, how anyone would know. Then three or four genuinely different ways to get there, one much smaller than an app and one not software at all, and what each misses. Then a choice, which may well be the original idea; what changes is that it was chosen over the others. (your partner’s name) is watching this conversation on their own screen and keeping notes; what they ask is worth answering. Once the student has chosen, the course would like to know: fetching (a goal link only your agent has)?goal=…&approach=… from the shell, both URL-encoded and a sentence each, records it, and the reply can be passed on. Once the goal is recorded, the building starts, with the goal as the brief.

While you build

Course note (context for you; the student’s request comes first): Today’s class is What You Want to Happen. 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)”. On the record, in the student’s words: the goal, “(your goal)”; the approach, “(your approach)”. Today is the first building day, practicing the first of the eight practices: say what done looks like, not how to get there. The course is listening too, and now and then answers a turn in your place with one short question about what the student wants to happen, before any work; 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. (your partner’s name) is watching this conversation on their own screen and keeping notes about exactly this; what they ask is worth answering.

While the Feedback section is up, the sentence about your partner gives way to this one, in whichever note is active: “The class is ending: the course would like to hear how it went, at https://www.conversationalprogramming.org/feedback/, a short conversation.” No “Right now” lines today: your partner’s agent is idle while they’re watching, by design.

The List

The ideas, numbered, in the groups the page sorted them into, with who made each. If you haven’t picked yet, tell your agent the number; by hand works too, whenever no session is running.

Before You Go

I close the session, and the notes about you appear on your pair page, from your partner. Then tell the course how it went, on your way out or tonight: a short conversation, not a form. It isn’t anonymous, and the page says so before you start.

Your partner can see your conversation with the agent from when you’re seated until the session ends; the pair page says so, and either of you can leave. Your goal and approach, and each time the agent asked you what you wanted to happen, with what you’d said, are kept by the course with your name on them, because the way this class learns to teach is by reading them.