Automate the Boring Stuff

Core concept 01

Chat, workflow, agent: pick the right rung.

There's a ladder here, and most people stand on the bottom rung without knowing there are more. Rung one is chat: you open a window, type the ask, and drive every single step yourself. Rung two is a workflow: you set the steps once, on a trigger, and they run themselves from then on. Rung three is an agent: it plans its own steps toward a goal instead of following ones you wrote. Anthropic's engineering team draws this exact line for developers, workflows run through predefined paths, agents direct their own process. This lesson lives almost entirely on rung two.

That third rung gets all the marketing budget, but the second one is where actual businesses save actual hours this month. A workflow doesn't need to think, it needs a clear trigger and a clear job: new inquiry comes in, AI drafts the reply, you approve, it sends. Anthropic's own advice to its developers is to start with the simplest system that works and add complexity only when it earns its keep. For almost everyone reading this, an agent that plans its own way through your business is a solution looking for a problem. A workflow, tuned to one task you already repeat, is the whole assignment.

Go deeper: this is the autonomy dial from Lesson 01, turned to work

Lesson 01's three settings, chat, workflow, agent, are the same three rungs here. The reason the middle rung wins for a business is not that agents are bad, it is that a task worth automating is by definition one that runs the same way every time, and sameness is the exact condition under which paying a model to re-plan from scratch adds cost and risk for nothing.

Put plainly: you climb to the agent rung when the steps genuinely cannot be known in advance. A weekly inquiry reply is not that. It is a workflow wearing an agent's price tag if you let it.

Source: Anthropic, "Building Effective Agents"

Core concept 02

Four properties decide it. Three say go, one says stop.

Start with evidence, not a hunch. Spend one honest week jotting down every task the moment you notice yourself doing it, especially the ones that make you sigh, then look for what shows up three, four, five times, not the one dramatic afternoon that will not repeat. Now run each repeat task through four properties.

Zapier's team frames the first three: is it frequent, is it repeatable (same steps every time), and is it low-judgment (execution, not a decision). Frequent, repeatable, low-judgment is the sweet spot. The fourth property is the one that keeps you out of trouble, and it is not about whether to automate but how far: blast radius, how much damage a wrong result does before a human sees it. A misfiled note is nothing. A wrong email to a client is a lot. The first three tell you what qualifies. The fourth tells you where the approval gate has to sit.

Source: Zapier, "When you should automate a task"

Scroll the diagram sideways →

Frequency × sameness · your first workflow is the top-right box, nowhere else

Core concept 03

Keep the approving. Automate the drafting.

Every workflow in this lesson has the same shape: trigger, draft, review, send. That third step, review, isn't the boring leftover part you'll eventually cut to go faster. It's the actual product. OpenAI's guide to building agents calls human-in-the-loop review the mechanism that lets a business trust an AI system enough to keep using it, catching the odd edge case before a client ever sees it.

Practically, this means you write the review moment into the workflow on purpose, not "whenever I remember." Ten minutes every Monday, or a single pass before anything leaves your drafts folder, works better than a vague promise to check on it later. Draft it, don't send it, is the whole rule. The tasks worth handing off are the ones where a wrong first draft costs you nothing but an edit. The tasks worth keeping in your own hands are the ones where a wrong first move costs you a client, a relationship, or a signature.

Go deeper: place the gate by blast radius, not by habit

Blast radius from concept two decides where the human sits, and it sorts cleanly. Actions that only ever produce a draft, an email waiting in your folder, a post saved but not published, a note filed, are the inner ring: a wrong result costs one edit, so the gate can be a relaxed weekly pass. Actions that send, spend, publish, or delete are the outer ring: a wrong result reaches a client or the public or is gone for good, so the gate must sit immediately before that step, every time, no exceptions.

The whole discipline reduces to one line: automate right up to the edge of the outer ring, and put your eyes on that edge. Draft, do not send, is that rule stated in four words.

Source: OpenAI, "A Practical Guide to Building Agents"

Your first automation · trigger, draft, review, send, with a loop back for edits

Scroll the diagram sideways →

Blast radius · draft freely on the inside, gate every crossing to the outside

The scorecard you leave with

Score your candidate task before you build the workflow

Take the one task your honest week surfaced most and score it. Three yeses and a known blast radius mean build it. A no on either of the first two means pick a different task.

01Frequent?did it show up 3+ times in one week — yes / no
02Repeatable?same steps every time, not a fresh decision each case — yes / no
03Low-judgment?execution, not the call that needs you — yes / no
04Blast radius?what a wrong result costs before you see it — draft-only, or does it send/spend/publish?
05The build = trigger · draft · gate · sendif 01–03 are all yes, put the gate where line 04 says, then give it two weeks before you judge it

Fresh from the lab

What changed this month

ChatGPT can now finish a whole project, not just answer one question. On July 9, OpenAI launched ChatGPT Work, an agent inside ChatGPT that gathers context from your apps and files and turns it into a finished document, spreadsheet, presentation, or small web app, working on its own for hours on one job. It runs on Scheduled Tasks, the same feature already driving the "Sunday at 4pm" workflows in this lesson, so a bigger engine just landed under a trigger you may already be using. It's rolling out first to Pro, Enterprise, and Edu plans.

Even the best AI agent still breaks a business rule on more than half its runs. On July 6, Zapier and Artificial Analysis launched AutomationBench-AA, an independent test of 657 real workflow tasks across finance, HR, marketing, sales, and support, run inside 40 simulated apps including Gmail, Salesforce, and Slack. The top score, from Anthropic's Claude Fable 5, was 48.6% of tasks completed without violating a guardrail. That's the best agent tested, on a scoreboard built for exactly this. It's the whole case for concept three below, made with data instead of a hunch.

OpenAI's standalone browser app is folding back into ChatGPT. Also announced July 9, ChatGPT Atlas, the separate browser-agent app, stops working on August 9, 2026. Its browsing and computer-use abilities move into ChatGPT itself and a new ChatGPT desktop app. If you'd started automating anything through Atlas, the same ability just moved one app over, no extra download required.

Sources: OpenAI, "ChatGPT is now a partner for your most ambitious work", July 9, 2026 · Artificial Analysis, "Announcing AutomationBench-AA", July 6, 2026 · OpenAI Help Center, "Evolving Atlas into ChatGPT", July 9, 2026

Vocabulary

Five words for your first automation

Learn these five and the next automation conversation, whether it's with a teammate or a rep from Zapier, gets a lot easier to follow.

Trigger

The event that starts a workflow without you clicking anything: a new form response, an email landing, a date on the calendar. No trigger, no automation, just a chatbot you have to remember to open.

Connector

The plug that lets one app hand data to another, sometimes called an integration. Zapier and Make call them apps; Claude and ChatGPT call the same idea connectors too.

Human-in-the-loop

The step in the workflow reserved for a person, on purpose. This is the safety valve for everything in this lesson, and the one part that's never optional.

Scheduled task

A prompt or workflow that runs itself on a timer, hourly, daily, or weekly, without you opening the app. Both Claude and ChatGPT now offer this on paid plans.

No-code

Building a workflow by connecting pre-made blocks in a visual tool like Zapier or Make, instead of writing code. Most of what's in this lesson is no-code.

Deeper how-tos, straight from the builders: ChatGPT automations docs · Claude Help Center · Zapier

Make it yours

Automate your #1 time sink

Same method, your one real time sink. Pick your lane and build the whole thing: a trigger, a drafting prompt, and a review moment you actually keep. Budget an afternoon to set it up, then two weeks to trust it. A free or a paid Claude, ChatGPT, Zapier, or Make account covers every workflow below.

For female founders

I'm CEO, marketer, bookkeeper, and intern in one body, and the actual building only happens after 9pm.

The 9pm rescue

Tool: Claude Cowork or ChatGPT Scheduled Tasks · weekly, date-triggered

Name the sink: for most founders it's the weekly content batch and follow-up round that keeps sliding to 9pm.
Pick a date trigger, not a "when I get to it" trigger. Sunday at 4pm, before the week starts, beats any amount of willpower.
Set up the scheduled task: "Every Sunday at 4pm, draft this week's three social posts and reply drafts for anyone I haven't heard back from in five days, using last month's posts as the voice guide."
Set the review moment for Monday morning coffee, ten minutes, approve or edit, never mid-week.
Let it run two weeks before deciding if it's working. One skipped Sunday isn't a failure, it's a data point.

The payoff: the 9pm slot goes back to being optional.

For artists

My statements and applications eat the exact studio time the work itself needed.

The deadline autopilot

Tool: Google Calendar + Claude or ChatGPT · date-triggered

Name the sink: the application answers and statement rewrite you redo from scratch for every open call.
Put every deadline on a shared calendar the moment you see it. That entry is the trigger, not your memory.
Two weeks out, draft it: "Using my artist statement and CV in this project, draft the application answers for [this open call], matching their word limits."
Set the review moment for that same evening, read it once out loud, that's the whole check.
Let it run two weeks, or two applications, before you decide the prompt needs work.

The payoff: one deadline stops costing you a full studio week.

For actors

Between survival jobs, self-tapes, and submissions, the admin of the career crowds out the acting.

The same-day submission line

Tool: Gmail templates + Claude or ChatGPT · triggered by a new breakdown

Name the sink: the cover note and slate script you rewrite from zero for every single submission.
The trigger is a new breakdown landing in your inbox, not "when I have a free hour" that never comes.
Draft it the moment it lands: "Here's the breakdown and my resume. Draft a tailored cover note and pick which two credits to lead with."
Set the review moment for right then, read it once, fix anything that doesn't sound like you, send.
Let it run two weeks of submissions before you trust the voice completely.

The payoff: submissions go out the same day, not the same week.

For musicians

I'm a musician, not a manager, but I'm the one booking gigs, chasing deposits, and answering "do you do weddings?" at midnight.

The midnight inquiry catch

Tool: Gmail filter + template · Claude or ChatGPT for the draft

Name the sink: the venue and wedding inquiry replies that pile up because they always land at odd hours.
Build a filter that labels anything with "booking," "available," or "rate" in the subject line. That label is the trigger.
Draft from the label: "Reply to this booking inquiry with our rate, availability process, and next step, in my usual friendly-but-firm tone."
Set the review moment for once a day, same time, approve or tweak before sending.
Let it run two weeks before you stop double-checking every reply by hand.

The payoff: inquiries get an answer before they go cold.

For fitness pros

I'm booked teaching all day, and then it's 9pm and I still haven't posted, answered DMs, or chased the clients who ghosted their re-sign.

The Sunday programming autopilot

Tool: Claude Cowork or ChatGPT Scheduled Tasks · weekly

Name the sink: Sunday's programming rewrite, redone from scratch for every client's week.
Set a standing Sunday trigger, same time as a class you teach, non-negotiable.
Draft the batch: "Here are this week's notes for each client. Draft their adjusted programming and flag anyone who hasn't rebooked."
Set the review moment for fifteen minutes before dinner, approve each program, done.
Let it run two weeks before you decide Sunday nights are actually different.

The payoff: Sunday admin drops to one sitting, not a whole evening.

For coaches

I sell transformation, but I spend my week on scheduling, session notes, and the follow-up emails I keep meaning to send.

The post-session autopilot

Tool: Calendar-triggered checklist + Claude or ChatGPT

Name the sink: the recap and follow-up email that should go out right after every session and usually doesn't.
The trigger is the calendar event ending, with a checklist reminder five minutes after.
Draft from three bullets: "From these three bullet points about today's session, draft the recap email and next step."
Set the review moment for right then, before you open the next tab, no exceptions.
Let it run two weeks of sessions before judging whether it saves real time.

The payoff: clients hear from you the same day, every time.

For therapists

Between sessions I'm doing intake calls, insurance, and admin, and I went into this to help people, not run an office.

The intake reply, minus the names

Tool: Google Form + Zapier or Make · draft only, no client data

Name the sink: the intake and insurance FAQ replies you retype for every new client.
The trigger is a new submission on your intake form, connected through Zapier or Make.
Draft with blanks, never names: "Draft a warm intake reply using this template, with [NAME] and [INSURANCE PROVIDER] left as blanks for me to fill in." No client information goes into a general AI tool, ever; clinical notes stay in a HIPAA-compliant system with a signed BAA, a different tool entirely.
Set the review moment for filling the blanks and reading once before sending, always.
Let it run two weeks before deciding the templates are actually done.

The payoff: the admin hour between clients goes back to notes, lunch, or nothing at all.

For authors

I want to be writing the next book, and instead I'm writing about the last one, forever.

The first-of-the-month newsletter

Tool: Claude or ChatGPT Scheduled Tasks · monthly

Name the sink: the monthly newsletter you draft from a blank page every single time.
The trigger is the first of the month, scheduled, not "whenever inspiration hits."
Draft from your notes: "Using this month's notes on what I've been reading, writing, and thinking about, draft the newsletter in my usual voice."
Set the review moment for coffee, add the one line only you would write, send.
Give it two send cycles before judging, a newsletter's real test is the second month, not the first.

The payoff: the newsletter stops eating the mornings that belong to the book.

For chefs

I'm cooking or shopping all day, so menus, quotes, and event logistics live in my head and my texts, until something slips.

The instant quote draft

Tool: Google Form + Zapier or Make · triggered by a new inquiry

Name the sink: the catering quote email you assemble from scratch for every single inquiry.
The trigger is a new submission on your catering inquiry form.
Draft from your numbers: "Using our per-head costs and this client's headcount and date, draft the quote email with our standard terms."
Set the review moment for once a day, adjust anything unusual, send.
Let it run two weeks of inquiries before you trust the pricing logic fully.

The payoff: quotes go out same day, and the kitchen never waits on the paperwork.

For consultants

I'm delivering client work all day, so my own pipeline goes cold every time I'm busy, which is exactly when it shouldn't.

The same-day proposal draft

Tool: Zapier or Make · form-to-draft, plus calendar follow-ups

Name the sink: the proposal you rewrite from a blank doc after every good call.
The trigger is a calendar event ending or a new lead landing in your form, either can fire it.
Draft from your notes: "From these call notes, draft a proposal using our standard structure and the two case studies most relevant to this problem."
Set the review moment for before it leaves your outbox, adjust price and scope, send same day.
Let it run two weeks before deciding it's saving you real turnaround time.

The payoff: the pipeline stays warm while you're billing, which is the whole trick of consulting.

For health advocates

Every client is a crisis, and the paperwork is endless. I'm doing intake, appeals, and case summaries with no time left to grow the practice.

The fill-in-the-blanks autopilot

Tool: Template library + Zapier or Make · draft only, no client data

Name the sink: the appeal letter and intake packet you rebuild for every new case.
The trigger is a new case type, which starts one new template, not a rewrite from zero.
Draft with blanks, never real health data: "Draft an insurance appeal letter template for [DIAGNOSIS TYPE] with [PATIENT] and [DETAILS] left as blanks." No client or patient information goes into a general AI tool, ever.
Set the review moment for filling the blanks by hand in your own case file, then sending.
Let it run two weeks before the template library starts to compound.

The payoff: the paperwork mountain becomes a fill-in-the-blanks library, and client hours go to clients.

For makers

I make the thing, then I photograph it, list it, ship it, and market it, and there's no time left to actually make.

The listing night rescue

Tool: Claude or ChatGPT Scheduled Tasks · weekly batch

Name the sink: listing night, when every new piece needs a title, description, and caption.
Set a standing weekly trigger, same night every week, before you've had time to dread it.
Draft the batch: "Here are two plain sentences about each new piece. Draft the full listing: title, description, materials, care, and a caption."
Set the review moment for reading each listing once before it goes live, five pieces takes ten minutes.
Let it run two weeks of listings before deciding it sounds like you.

The payoff: listing night shrinks to listing hour, and the studio gets its maker back.

For nonprofit leaders

I'm chasing grants, thanking donors, running the program, and writing the report all at once. I'm the whole org, and the mission is waiting on my inbox.

The same-week thank you

Tool: Zapier or Make · triggered by a new donation row

Name the sink: the donor thank-you letter that should go out within a week and sometimes takes a month.
The trigger is a new row in your donation spreadsheet or a new payment notification.
Draft it personalized: "Draft a thank-you letter for this gift, personalized by amount and by whether they're a first-time or repeat donor."
Set the review moment for adding the one line that needs a human touch, then sending.
Let it run two weeks of gifts before deciding donors notice the difference.

The payoff: every gift gets thanked the same week, and the mission stops waiting on your inbox.

For photographers

I'm booked shooting on weekends and editing all week, so inquiries wait and the blog died years ago.

The same-day inquiry reply

Tool: Website form + Zapier or Make · drafted in Gmail

Name the sink: the inquiry reply that used to go out same-day and now waits until editing week ends.
The trigger is a new submission on your contact form.
Draft it automatically: "Reply to this inquiry with our packages, pricing, and availability, matching our usual warm tone."
Set the review moment for once a day, personalize the first line, send.
Let it run two weeks of inquiries before you retire the manual reply for good.

The payoff: inquiries get a same-day reply without hiring a VA.

For realtors

My database is a graveyard. I forget to follow up with last year's buyers, and every new listing is a scramble of descriptions and emails.

The graveyard revival

Tool: Claude or ChatGPT Scheduled Tasks · monthly, plus a new-listing trigger

Name the sink: the past-client follow-up that never happens because there's always a new listing to write up first.
Set two triggers: the first of the month for the database pass, a new listing going live for the description.
Draft the segments: "Segment my past clients into buyers, sellers, and dormant leads. Draft one re-connection note per group that doesn't sound like a campaign."
Set the review moment for reading each batch once, add a personal line for anyone you actually remember, send.
Give it two cycles before deciding the database is worth reviving.

The payoff: the graveyard becomes a referral engine, on one planning hour a month.

For stylists

I'm behind the chair all day, so rebooking, DMs, and content pile up, and the clients I forget to rebook just don't come back.

The weekly rebooking sweep

Tool: Claude or ChatGPT Scheduled Tasks · weekly

Name the sink: the rebooking text you mean to send and don't, until the client's already gone somewhere else.
Set a standing weekly trigger, same day every week, checking who's due.
Draft the texts: "Here's who's due for rebooking this week and their usual service. Draft a friendly, specific text for each."
Set the review moment for a slow ten minutes between clients, send from there.
Let it run two weeks before you trust it with your regulars.

The payoff: an empty chair becomes a choice, not an accident.

The house rule

Lauren's line, applied to every workflow in this lesson: what belongs to you and keeps your humanity, you keep. What frees you to be more human away from your laptop, you hand off. The approving is yours. The drafting is not. Build the trigger, write the prompt, and hold onto the review step like it's the whole business, because it is.

Next lesson: Ship something that pays →