Directing @yagni · Lesson 3 of 4
How do you ask @yagni for help?
Published June 9, 2026 · Updated August 14, 2026
You can ask from three doorways: the pinned composer on the Front, the contextual rail scoped to whatever is in the center, and the full-page Ask. One voice, one memory. Answers are grounded in the live rows of your connected tools and cited, and the chat can act, not just answer.
Where do you talk to @yagni?
Three doorways, one voice. The Front carries a pinned composer. Most working surfaces carry a contextual rail scoped to what is in the center: open a case file and it is ready to discuss that item, open a Brief and it is ready to discuss that snapshot. And the full-page Ask is the destination for longer investigations. Command-K or Control-K takes you to the nearest one.
What can it do, not just say?
Act. The chat is not a commentary track on the product; it is a control surface. Ask for a Team and the Team gets created. Ask for an investigation and the work launches. Ask for a Library page to be fixed and it is edited. The line it never crosses is the same line everything in YAGNI observes: reversible acts execute in the thread, and anything irreversible stages a Decision inline, in the conversation, and waits.
What grounds the answers?
Three layers, all cited. The live rows of your connected tools, so “is the webhook fix deployed” is answered from GitHub, not from memory. The Library, the shared store of docs and context your Teams read and write. And the decision ledger, so “why queues instead of cron” comes back with the decision, who made it, and what it rejected. Any citation opens the real thing.
Will it flatter you?
No, and this is deliberate. You get its read and its confidence. When you propose something the evidence undercuts, it shows you the evidence. When the call is genuinely close, it says so and hands it to you. An assistant that always agrees is a mirror, not a colleague, and mirrors do not catch the incident before it happens.
Can your engineers ask from the terminal?
Yes. The same grounding rides inside YAGNI Code: which flag gates this, why is the table written here, what did we decide about retries. It is also applied automatically as code completes, when the agent checks the change against the ticket, the doc, and the thread where scope moved, so the code fits the decision and not just the diff.
Common questions
Is the Front composer a different agent from the rail?
No. There is one voice with one memory. The composer, the contextual rails, and the full-page Ask are different doorways into the same conversation; Command-K or Control-K takes you to the nearest one.
If there is one voice, who does the work?
The workers engaged on your Teams. @yagni speaks for them, briefs you, and takes requests, but the work is done and credited under the Team that carried it. One voice in the chat, many names on the record.
Will it just agree with me?
No. It gives you its read and how confident it is. When the evidence cuts against you it says so, and when it is not sure it says that too and hands you the call.
Can @yagni do things, or only answer?
Both. The chat acts with full power: it can create Teams, launch work, and edit Library pages. Reversible acts execute in the thread; anything irreversible stages a Decision inline and waits for your approval, as everywhere in YAGNI.