The 2026 hype says you need to “build an AI agent.” Most solopreneurs don’t. What you need is one
smart step inside a workflow you already run — an AI action that summarizes an email, classifies a
lead, or drafts a reply, then hands off to the rest of your automation. This is the “AI in the loop”
section of the solopreneur automation stack: AI as plumbing, not
a product.
TL;DR — the four AI steps worth adding
- Summarize: long email or transcript → three bullet points.
- Classify: form submission → the right tag/route based on what they wrote.
- Draft (with a human check): generate a first-draft reply you approve before it sends.
- Extract: pull structured fields (invoice number, amount) out of messy text.
- The rule: AI is a step, not the destination. Keep a human approval on anything customer-facing,
and watch the cost — AI steps often bill differently than normal ones.
Why “add a step” beats “build an agent”
Here’s the trigger, here’s the action — and now, in the middle, an AI step. That’s the whole idea.
You’re not building a standalone AI product; you’re dropping one AI-powered action into a scenario
you already trust. The trigger fires, the AI transforms some text, and the next step does something
with the result.
The reason this matters for a team of one: an AI step is testable, cheap to add, and easy to rip
out if it misbehaves. An AI agent that runs unsupervised is a maintenance liability you don’t have
the time to babysit. Start with the step. Graduate to agentic behavior only when a specific,
repeatable job earns it.
In scope here: AI as a step inside Make or Zapier. Choosing which standalone AI assistant to
pay for is a different question and a different category — this piece stays in the automation lane.
How to add an AI step in Zapier
Zapier’s built-in tool is AI by Zapier, and it plugs into any Zap as an action.
- In your Zap, add a new action step.
- Search for AI by Zapier (or a specific provider like OpenAI/Anthropic/Google if you have
keys). - Pick the action: Analyze Text or Transform Text are the workhorses for summarize,
classify, and extract jobs. - Write a clear prompt and map in the field you want it to work on (e.g., the email body from the
trigger). - Use the AI step’s output in your next action — a tag, a Slack message, a draft.
Zapier includes several models at no extra charge (for example a smaller GPT model and a fast Gemini
model), with the option to connect frontier models from OpenAI, Anthropic, and Google. Start with an
included model; upgrade only if quality demands it.
How to add an AI step in Make
Make exposes AI through modules — an OpenAI module (and others) you drop onto the canvas like any
other module.
- Add a module to your scenario and choose your AI app (e.g., OpenAI).
- Pick the operation (create a completion / analyze text).
- Map the incoming data into the prompt.
- Route the AI output to downstream modules with the usual mapping.
Cost warning, and it’s a real one: Make’s AI modules consume variable credits based on token
usage, not the flat one-credit-per-run of a normal module. A chatty prompt over a long transcript
can cost many credits per run. Before you wire an AI step into a high-volume scenario, test it and
check the credit cost on a few real runs. I’ve seen an AI summarize step quietly become the most
expensive part of a workflow. (For how credits work generally, see
Zapier vs Make vs n8n.)
The four steps, as real builds
1. Summarize — long input to short output
- Where it shines: the weekly digest from the
course automation guide. Feed the AI step your raw
numbers or a long client email; get back three tight bullets. - Prompt shape: “Summarize the following in three bullet points, plain language, no preamble:
{{text}}.” - Gotcha: cap the input length. Summarizing a 50-page transcript on every run gets expensive
fast. Filter or truncate first.
2. Classify — route based on meaning
- Where it shines: new-subscriber tagging. The AI reads a free-text “what do you need help
with?” field and returns a category you route on. - Prompt shape: “Classify this into exactly one of: [Coaching, Course, Freelance, Other].
Respond with only the category: {{text}}.” - Gotcha: constrain the output to a fixed list, or you’ll get chatty answers your filter can’t
match. Then branch with a router/paths step on the returned value.
3. Draft with a human check — the safe way to use AI on customer-facing text
- Where it shines: support replies, outreach follow-ups.
- The build: trigger → AI drafts a reply → the draft goes to you (Slack/email) for
approval → only on your approval does it send. Never let the AI send customer-facing text
unattended. - Gotcha: this is the single most important pattern in this article. AI drafts; a human ships.
The approval step is not optional on anything a customer will read.
4. Extract — messy text to clean fields
- Where it shines: turning an unstructured email (“hey, invoice #4471 for $250 is paid”) into
structured fields you can log to Airtable. - Prompt shape: “Extract invoice_number and amount as JSON from: {{text}}.”
- Gotcha: validate the output before you write it to a record. Add a filter that checks the
fields look right (amount is a number, invoice matches your format) so a bad extraction doesn’t
poison your data. Pipe clean data in — even when the “pipe” is an AI.
When NOT to add an AI step
- The task has a deterministic rule. If you can classify with a simple “if the email contains
X,” use a filter, not AI. It’s cheaper, faster, and can’t hallucinate. - The output ships to a customer unattended. Always insert a human approval step first. AI
drafts; you approve. - Volume is high and the input is long. Token-based billing (especially on Make) can make an AI
step the most expensive thing in your account. Test the cost before you scale it. - You’d be automating a judgment call. Summarizing is fine; deciding to refund a customer is
not an AI step.
Frequently asked questions
Do I need to know how to code to add AI to Make or Zapier?
No. Both offer no-code AI actions — AI by Zapier in Zapier, and AI modules (like OpenAI) in Make.
You write a prompt and map fields; there’s no coding required. The skill is in prompt clarity and
adding a human check.
How much does an AI step cost inside an automation?
It varies. Zapier includes some models at no extra charge on eligible plans; frontier models may
cost more. Make’s AI modules bill variable credits based on token usage, so long inputs cost more
per run. Always test on a few real runs before scaling.
Is it safe to let AI send emails automatically?
Not for customer-facing text. Use the draft-with-human-check pattern: AI writes the draft, it goes to
you for approval, and only then does it send. Reserve fully automatic AI output for internal,
low-stakes text.
What are the best AI steps to start with?
Summarize and classify. They’re low-risk, immediately useful, and don’t touch customer-facing
output. Add draft-with-approval and extract once you’re comfortable.
The bottom line
You don’t need to build an AI app to get AI’s value — you need one well-placed step in a workflow you
already run. Summarize and classify are the safe, high-return places to start. Draft only with a
human approval step, and extract only with output validation. Watch the token-based cost, especially
on Make, and keep AI where it belongs: as plumbing inside the automation, never the hand on the
customer-facing send button.
Zoom back out to the full stack:
The solopreneur’s automation stack — or put an AI step into
real recipes with
8 no-code workflows for your course and newsletter business.
Sources
- Zapier — “AI by Zapier: Easily add AI steps to your workflows” (Analyze Text / Transform Text actions; included models; connect OpenAI/Anthropic/Google). Verified 2026-09-26. https://zapier.com/blog/ai-by-zapier-guide/
- Make Help Center — “Introducing credits: new billing unit live in Make” (AI modules consume variable credits based on token usage). Verified 2026-09-26. https://help.make.com/introducing-credits-new-billing-unit-live-in-make
- Make — official pricing page (credit model context). Verified 2026-09-26. https://www.make.com/en/pricing
- Zapier — official pricing page (plan context for AI features). Verified 2026-09-26. https://zapier.com/pricing
