You've met the loop, the rungs, and the job card. Now the recipe box. Each agent below tells you what it does, which rung it runs on, the full job-card text, a quick review checklist for when it reports back, and roughly how much time it buys you. Start with the one that matches the chore you wrote down in Lesson 1 — you don't need all ten, you need one that sticks.
Every job card keeps the golden rule: draft, don't send. Nothing here spends money or publishes without your click.
1 Morning briefRung 1 · scheduled daily · saves ~20 min/day
What it does: Every morning at 7am it summarizes yesterday — your email, your key numbers, any mentions — and hands you today's top three priorities, so you start the day oriented instead of scrolling.
GOAL: Give me a 2-minute morning brief. Done means: what happened yesterday + my 3 priorities today. INPUTS (read-only): my email since yesterday, [my metrics source], [mentions source]. STEPS: 1) Scan each source. 2) Pull what actually matters. 3) Propose today's top 3. RULES: Summarize only — never reply, send, or delete. Flag urgent, don't act. If a source is empty, say so. DELIVERABLE: A short brief in the chat, run daily at 7am. REPORT BACK: The brief, the 3 priorities, and anything urgent I should see first.
- Did it miss anything that mattered? Add that source next time.
- Are the 3 priorities actually yours, or generic?
- Did it stay read-only?
2 Content repurposerRung 2 · on demand · saves ~1 hr/post
What it does: Takes one blog post and turns it into pins, captions, and a newsletter blurb — in your voice — so a single piece of writing feeds a whole week of channels.
GOAL: Turn my latest post into a week of social content in my voice. Done means: 5 pin descriptions, 3 captions, 1 newsletter blurb. INPUTS: only the post at [link] and my voice guide at [file]. STEPS: 1) Read both. 2) Pull the 3 strongest ideas. 3) Write each piece from them. RULES: Drafts only — never post or schedule. Match my voice; ask if unsure. No invented stats. DELIVERABLE: One doc "Repurpose — [post]", each piece labeled, in my Content folder. REPORT BACK: What you wrote, anything unsure, which piece is strongest.
- Does it sound like you, or like AI?
- Any claim not backed by the post? Cut it.
- Would you actually publish these as-is?
3 Inbox triageRung 1–2 · daily · saves ~25 min/day
What it does: Labels and summarizes your inbox and drafts replies — but never sends them. You skim, tweak, and hit send yourself.
GOAL: Cut my inbox to a 5-minute job. Done means: everything grouped + draft replies waiting for the ones that need one. INPUTS: my email from the last 24h. STEPS: 1) Group: needs reply / FYI / junk. 2) One-line summary each. 3) Draft (don't send) replies for the "needs reply" ones. RULES: NEVER send, archive, or delete. Drafts stay as drafts. Flag anything urgent or sensitive for me. DELIVERABLE: The grouped list + draft replies saved to drafts. REPORT BACK: The groups, which drafts you wrote, anything you weren't sure how to answer.
- Read every draft before sending — always.
- Did it draft anything it shouldn't have touched?
- Nothing sent or deleted without you?
4 Research assistantRung 2 · on demand · saves ~2 hrs/decision
What it does: Compares your options — suppliers, tools, plans — into a table with a cited recommendation, so you decide from evidence instead of a dozen open tabs.
GOAL: Compare 5 options for [thing] and recommend one. Done means: a comparison table + a one-paragraph pick with reasons. INPUTS: web search, and my notes at [file] if any. STEPS: 1) Identify the 5. 2) Compare on price, [factor], [factor]. 3) Recommend for my situation. RULES: Cite a link for every fact. Say so if sources disagree or are thin. No purchases or sign-ups. DELIVERABLE: The table + recommendation in the chat. REPORT BACK: The table, your pick, how confident you are in the sources.
- Click a couple of the cited links — do they say what it claims?
- Are the comparison factors the ones you actually care about?
- Is the recommendation reasoned, or just confident?
5 Bookkeeping prepRung 2 · monthly · saves ~2 hrs/month
What it does: Reads your receipts folder once a month, categorizes each one, and flags anything odd into a spreadsheet — so your books are prepped, not dreaded. (It preps; you or your bookkeeper still review.)
GOAL: Prep this month's expenses for review. Done means: a categorized sheet + a flag list of anything unclear. INPUTS: only the receipts in [folder]. Read-only on the receipts. STEPS: 1) Read each receipt. 2) Categorize (vendor, amount, date, category). 3) Flag duplicates, missing info, or oddities. RULES: Never delete or rename a receipt. Don't guess a category — flag it instead. This is prep, not filing taxes. DELIVERABLE: A new sheet "Expenses — [month]" + a "Flags" tab, in my Bookkeeping folder. REPORT BACK: The totals by category, the flags, and anything I need to find a receipt for.
- Spot-check a few categories against the actual receipts.
- Do the flags make sense?
- A human still signs off before anything hits your taxes.
6 Content calendar fillerRung 2 · monthly · saves ~90 min/month
What it does: Harvests your scattered notes and ideas and proposes next month's content calendar in your planner — turning "I have no idea what to post" into a draft you edit.
GOAL: Propose next month's content calendar. Done means: a dated list of post ideas across my channels, drawn from my own notes. INPUTS: my ideas/notes at [file], last month's calendar at [file]. STEPS: 1) Read my notes. 2) Cluster into themes. 3) Lay out a month, balanced across channels, with a title + one-line angle each. RULES: Proposals only — don't schedule or publish anything. Use my ideas first; only suggest new ones if I'm short. DELIVERABLE: A draft calendar in [planner/doc], marked "DRAFT — review". REPORT BACK: The calendar, which ideas came from my notes vs. new, any gaps.
- Do the ideas fit your brand, or feel generic?
- Is the mix across channels sensible?
- You approve before anything gets scheduled.
7 Listing refresherRung 2 · quarterly · saves ~2 hrs/quarter
What it does: Rereads your product or service pages each quarter, flags stale or weak copy, and proposes A/B title options — so your listings don't quietly go dusty.
GOAL: Audit my listings and propose refreshes. Done means: a list of stale/weak spots + 2 title options per listing. INPUTS: my product pages at [links/file], my voice guide at [file]. STEPS: 1) Read each listing. 2) Flag stale info, weak hooks, unclear benefits. 3) Draft 2 alt titles each. RULES: Proposals only — never edit or publish the live listings. Keep prices/facts unless I confirm a change. DELIVERABLE: A doc "Listing refresh — [quarter]" with flags + title options. REPORT BACK: The weakest listing, the strongest new title idea, anything factually out of date.
- Are the flags fair, or nitpicks?
- Do the new titles keep your voice?
- You make every live edit yourself.
8 Follow-up nudgerRung 2 · weekly · saves ~40 min/week
What it does: Scans your sent mail for threads that went quiet more than a week ago and drafts polite bumps — so warm leads and loose ends don't slip through.
GOAL: Find threads I never heard back on and draft gentle nudges. Done means: a list of unanswered threads > 7 days + a draft follow-up for each. INPUTS: my sent + received mail from the last 30 days. STEPS: 1) Find sent threads with no reply in 7+ days. 2) Skip anything clearly closed. 3) Draft a short, warm bump each. RULES: NEVER send — drafts only. Don't nudge personal or clearly-finished threads. Keep it brief and kind. DELIVERABLE: A list + draft nudges saved to drafts. REPORT BACK: Who's gone quiet, the drafts, any I should skip.
- Would you actually want to nudge each of these?
- Do the drafts sound warm, not pushy?
- Nothing sent without your read + click.
9 Weekly review builderRung 1 · scheduled Sunday · saves ~45 min/week
What it does: Every Sunday at 5pm it compiles your numbers, your wins, and next week's plan into one page — so you close the week with clarity instead of a vague guilt.
GOAL: Build my weekly review. Done means: one page — the week's key numbers, 3 wins, and next week's top priorities. INPUTS (read-only): [my metrics source], this week's calendar, my notes at [file]. STEPS: 1) Pull the numbers. 2) Note what went well. 3) Propose next week's top 3–5. RULES: Summarize and propose only — change nothing. If a number's missing, leave a blank, don't invent it. DELIVERABLE: A one-page review doc, run Sundays at 5pm. REPORT BACK: The page, the trend vs last week, what you think next week's #1 should be.
- Do the numbers match reality?
- Are the proposed priorities right, or do you re-order them?
- It proposes; you decide.
10 The intern sessionRung 4 · one-off · saves hours per project
What it does: The general-purpose pattern — "here's a folder and a goal, work through it" — for the one-off jobs that don't fit a recipe. This is the top rung, so it earns the most careful job card and the closest review.
GOAL: [the one-off outcome]. Done means: [what finished looks like]. INPUTS: only [this folder / these files]. Start read-only; ask before writing anything. STEPS: 1) [rough first step]. 2) [second]. 3) Check your work against "done" before reporting. RULES: Never delete. Never send, publish, pay, or promise. Show me before any change I can't undo. Ask when unsure. Stop at [time/scope cap]. DELIVERABLE: [what to produce], named [name], in [where]. REPORT BACK: What you did, what you couldn't, what needs my eyes, and what you'd do next.
- Because it's the most autonomous, review the most: check its work, not just its summary.
- Did it respect "ask before writing"?
- Save this only for jobs you can fully review afterward.
Ten recipes is a menu, not a checklist. Pick the one that matches the chore you're most tired of, run it well for a week, and let it earn its place before you add a second. An agent you actually trust beats nine you set up and abandoned.
- Open the recipe that matches your Lesson-1 chore.
- Copy its job card and fill in every blank with your real files and rules.
- Save it next to the card you drafted in Lesson 3 — whichever is sharper is the one you'll run in Lesson 6.