Skip to content

The spring launch: one sample task, from brief to finished work

Sample data. Everything on this page is made up to show how the product works. The marketing team, its people (Priya, Sam, Leo, Ana and Kai), its tickets and its quotes are sample data, not a customer and not a result. The page follows one piece of work, the marketing team's spring launch, from Priya's brief to the VP to the finished work, so you can see every pillar at work on a single task: Steer by intent, At scale, Together.

The brief: Priya says what she wants

Priya opens Constraint on her own computer and types one line to the VP, the one AI she talks to: "Get the spring launch ready for Thursday." The VP runs the marketing department, so it aligns with her first. It asks what "ready" covers, and Priya answers on Home: the announcement, the pricing page, the customer email, the sales FAQ and the launch blog post.

Then the VP files the tickets in the ticketing service the team already uses and puts an agent team on each. The announcement gets a Planner, a Writer and a Reviewer, working in order; the pricing page goes to a single Writer. One ticket, the demo script, fits none of the department's roles, so the VP asks Priya before it creates a role for it. She says yes.

What Priya sees: the VP's reply, listing every ticket it filed and the agent team on each, and then the board filling as the agent teams start, all at once, on the team's computers. She has said what she wants once, and the work is under way. (Steer by intent)

A sentry finds work nobody asked for

The marketing department has a sentry, an automatic check that finds work, pointed at the pricing page. Sentries do discovery and analysis on their own schedule, or when something happens in a connected app. This one notices that the pricing page keeps being updated by hand, and that the launch will change the prices again.

It proposes a ticket: keep the pricing page current through the launch, with a Writer on it. The proposal waits for a yes on Priya's Home, next to her other decisions. She says yes, and the ticket is filed with an agent team on it, like the others. Nobody briefed this piece of work; the sentry found it. (Steer by intent)

Sam assigns a ticket to himself, and it starts

Sam owns the customer email, so he assigns that ticket to himself in the ticketing service. That is all he does. In auto mode, the right agent team starts on the ticket by itself, in the stages Sam allows, and it will ask on Home when it needs him. The Writer's session runs on Sam's computer, while his desktop app is running.

Everyone on the team can see that session live on the ticket, marked with Sam's computer, whoever opens the ticket and wherever they sit. Tickets start themselves; the team watches them run. (At scale)

A question reaches Priya, who asks the right person

The Writer on the announcement, running on Priya's computer, needs a decision: which customer quote should lead the announcement? It does not stop Priya in a thread. The question appears as a card on her Home, with the research behind it, next to the one question the VP asked earlier. Home shows Priya only the decisions that are hers, in one place.

Priya does not own the customer relationship; Sam does. So she assigns the question to him. Sam gets it in his app, on his own computer, even from home, and picks the quote. His answer reaches Priya's session on her computer, which carries on, and the Writer notes the answer on the ticket. Sam has answered a question in Priya's work without touching her computer, and a teammate's answer never approves a tool from another computer. (Assign a question)

Leo steps in, and the whole team weighs in

Leo opens the announcement ticket from his own computer and sees the Writer's session live, marked as running on Priya's computer. Both computers are on, so he steps in. He messages the Writer: say that the old reports stay available. The Writer reads it and adds the line to both drafts. On Priya's screen, Leo's message sits inline in her session, marked as his.

Ana never opens the session. She comments on the ticket itself, in the ticketing service: keep the one-pager to one page. Several people are now on one ticket, and the AI working it reads everyone's comments and makes both changes. Nobody waited for Priya, and nobody worked in the dark; every agent on the launch is visible to the whole team, with whose computer runs it. (Together)

Inside the rules, and checked before done

Every role in the marketing department reads the department's rules first, before it touches a ticket: claims come from the research, prices come from the pricing page, and the one-pager is one page. Each person has also chosen how far their agents go before they ask. Priya allows edits; Kai asks first when a permission is needed, and allows one for the rest of a session when he is sure. They chose these settings, and they can change them.

The Reviewer on the announcement has checks to run before it says it's done: every claim against the research, and the one-pager's length. It runs them before it says it's done. The work gets done well without pulling Priya back in. (At scale)

Thursday: the finished work, and its record

By Thursday the launch tickets are done: the announcement, the pricing page, the customer email, the sales FAQ, the blog post and the demo script. Each session's record, its trace, is saved with its ticket to Constraint's cloud service. Before it is saved, screenshots are removed and keys, tokens and passwords are blanked out. The Traces page shows every trace, by team and by person, and Priya can open any one and read every step the AI took. People see these records only for the teams they can read.

Sam's answer about the quote, Leo's line, Ana's comment and the Reviewer's checks are all there, with the reasons. (Where data goes)

Later: the next launch learns from this one

When the team starts the summer launch, its agents look up how the spring launch was done before they ask anyone. They find that leading the announcement with a customer quote was deliberate, Sam's choice, so they keep to it. The demo script's length looks accidental, so that is the one thing they ask about, on Home.

The Analysis Hub reads the whole team's traces, not one person's. It shows that the launch-messaging skill isn't working: the Writer stopped to ask the same question in every run. It proposes a better version that settles the question up front. Priya checks it, and because skills are saved in versions, every role that uses the skill can take the new version. Every run has taught the next. (Analysis Hub)

What the example shows

Priya said what she wanted once, and the VP and a sentry turned it into tickets with an agent team on each: steer by intent. Every ticket ran at once, on the team's own computers, inside the department's rules, without pulling her back in: at scale. Each question went to the person it belonged to, teammates stepped in, and every run was saved so the next one could build on it: together.