First week with your agent

Your agent learns by doing. Every message you send, every correction you make, every preference you share — it remembers. The clients who get the most value in week 1 are the ones who treat it like a new hire on day 1: clear, patient, and constantly correcting.

• Talk to it the way you'd talk to a smart, eager assistant who's never met you. Tell it your name, your title, what you do, what you don't.
• Correct it out loud. "Too long, give me 2 sentences." "That's not how I'd phrase it." "Don't use exclamation marks." Each correction sticks.
• Ask it to remember. "Remember I prefer Slack over email when possible." "Remember Sarah is my COO, treat her requests as priority."
• Don't apologize for asking too much. The whole point is to push it.
• If it does something brilliant, tell it. The agent saves successful patterns and replays them.

A loose week-1 schedule. None of it is mandatory. But clients who follow this shape get to a smart agent in 7 days instead of 30.

The agent picks up your tone fast if you're explicit. Vague feedback ("I don't like that") takes 5x longer to land than specific feedback ("too formal — use lowercase, no exclamation marks, max 2 sentences").

Things to tell it explicitly

• Email tone (formal / casual / punchy / warm / direct).
• Length default ("keep emails under 4 sentences unless I say otherwise").
• Sign-off ("always sign emails 'Cheers, Alex' unless it's a first-time client").
• Capitalization ("never use sentence case in subject lines").
• Forbidden words (every founder has 5-10 — "never use 'thrilled', 'circle back', or 'just wanted to'").
• Default urgency posture ("always check in before sending anything to a client").

The fastest way to teach voice

Forward 5 emails you've sent in the last week to your agent and say "match this voice for everything you draft going forward". The agent reads them, extracts your patterns, saves them, and applies them. Takes 60 seconds and saves a month of back-and-forth.

Memory

Every important fact your agent learns gets saved to a memory file: who's on your team, your work hours, your preferences, recurring tasks, sensitive topics to avoid. You can ask "what do you remember about me?" any time and it will read its memory back to you.

Memories are editable. If something is wrong, say "forget that" or "update that — actually I'm the CTO not the COO". Done.

Skills

A skill is a saved how-to. "How I write a weekly client email." "How I prep for a sales call." "How I write a job description." You build them by walking the agent through the task once, then saying "save this as a skill called X". Next time you say "run the X skill" it does the whole flow.

Most clients have 5-10 skills by month 2. The library compounds — every skill is a personal SOP your agent now executes on demand.

Handle yourself (just talk to the agent)

• Wrong tone, wrong format, wrong info → tell it, correct it, save the correction.
• Want it to remember a preference → tell it, ask it to remember.
• Want a new skill → walk it through once, save it.
• Slow response → ask it to check its own status. It usually self-diagnoses.
• Refuses something it should be able to do → push back. Say "try anyway" or "explain why you can't, then try a different approach".

Ping us (drop a message in our shared Telegram)

• It's not responding at all (more than 2-3 minutes).
• An integration broke (Gmail OAuth expired, calendar disconnected, etc).
• You want a new integration (CRM, Slack, anything new).
• You want to change something deep — agent's name, personality, autonomy level, scheduled tasks, working hours.
• Something feels off and you can't put your finger on it.

• Agent forgets things between conversations. Usually means you didn't say "remember this". Memory is opt-in for context-window reasons — be explicit and it sticks.
• Agent over-explains. Tell it "shorter, less preamble". It'll dial back fast.
• Agent asks permission for everything. By design on day 1. After a week of trust, we tune that down per category — "send emails to anyone in this list without asking", etc.
• Agent sends a weird response. Sometimes a model has an off moment. Tell it "that was wrong, try again" — usually fixes it. If it keeps happening, ping us.
• Token budget warnings. If you're a heavy user, you might hit your Claude/OpenAI plan limits in week 1. We'll see it on our end and bump you to the higher tier if you're not already.
• Mac mini sleeps. Shouldn't, but if it does, the agent goes silent. Wake the Mac and you're back. Tell us — we'll lock down sleep settings.