Writing

How to choose tasks for your first AI agent when you run solo

By Scott Wilder · October 31, 2026 · 6 min read · All writing

The best first task is one you have already done by hand, that repeats often, needs little judgment and costs little when it goes wrong. For many solo operators that means follow-up with quiet leads, which Operater, a company that sells an agent platform, calls the highest-revenue, lowest-risk workflow in almost every solo business. Run one agent on that one task for two weeks before you add anything.

Quick version

Why picking wrong feels risky

If you work alone, every hour spent setting up the wrong thing is an hour you do not get back. The fear is reasonable.

An agent is also not a fixed script. Dust says agents adapt based on what is happening, choose which tools to use and handle tasks that require some judgment. That flexibility is useful. It is also why a first agent needs a narrow job and a short leash. Dust sells an agent platform, so read its enthusiasm with that in mind.

Our view: the real risk is not picking a slightly wrong task. The risk is picking a big one. A narrow task that fails costs you a couple of weeks. A broad one that fails can reach customers and money. Spending is its own topic, covered in AI agent spending limits.

A four question filter

Run every candidate task through these questions. Two honest "no" answers mean the task waits.

  1. Does it repeat often? One-off jobs rarely repay the setup time.
  2. How much judgment does it need? The less, the better.
  3. What does a mistake cost? If a customer reads the output as your voice, or money moves, treat the stakes as high.
  4. Have you done it by hand? If not, do not automate it yet.

Questions one to three follow a sorting that Dust describes. Dust sells an agent platform, but the logic holds up on its own:

Frequency Judgment or stakes What to do
High Low judgment Strong automation candidate
High Moderate judgment Worth automating with oversight
Low High stakes Keep a human in the loop

The same page adds that strategic decisions and high-stakes work should stay with humans.

Question four comes from Operater, which also sells an agent platform. It says a solo operator automating a process they have never done manually almost always automates the wrong thing. Its fix is to do the process by hand five times, write down what you actually did at each decision point, and only then hand it over. That written list is also your instruction sheet for the agent.

Tasks that tend to pass the filter

These are candidates, not orders. Your own work decides.

The lead response workflow gives a sense of the setup time. FixerAI says it can be built in 25 to 35 minutes if you have never used Make.com before. That fits inside one spare evening hour, though it is a vendor estimate for one workflow, not a promise for yours.

Content is where the sources disagree

Operater says AI for solopreneurs pays off least in content generation and most in follow-up, qualification and admin. FixerAI describes a SaaS founder whose content work went from 6 hours to 45 minutes per week. Both companies sell something, and neither is neutral.

Our view: content is a poor first pick even if the second source is right. Your writing is your voice, and a customer will read it as yours. Start where a bad output stays in your drafts folder.

What to keep for yourself

Operater's advice is blunt: do not delegate positioning, pricing, or anything a customer will read as your voice until you have checked it works. Dust puts it the other way around: low frequency, high stakes means keep a human in the loop.

Our view: add anything that touches your money, your reputation or someone's trust to that list. The full argument is in what to never automate with an AI agent. The Agent Leash, a one-page checklist of 15 checks before your AI agent touches your money, covers the money side.

How to tell you are over-automating

These are warning signs, and this list is judgment, not sourced fact.

If two or more are true, cut back to one workflow and fewer permissions.

Set a pass mark first

A test needs a number you picked before you saw the results. Dust gives one example: for a call summary agent, success might mean accurately capturing 95% of action items and decisions. Operater suggests a floor for time: the minimum useful result is three hours returned per week, and if it is under three, it says something is wrong with the setup, not the concept.

Be careful with big claims from vendors. Dust reports that Profound's post-sales team reclaimed 1,800+ hours per month. That is a team result. A solo operator's gain is smaller, and no source here measures it for you.

What it costs

Two vendors give price ranges, and both are estimates for their own kind of stack:

Operater also lists $39 a month on Solo for its platform and Apollo at around $49 a month for contact data. Added together, $39 plus $49 is $88 a month. Prices change, so check the vendor's own pricing page before you commit. Our view: start on free tiers where you can, and spend only after the two week test passes.

The two week test

  1. Pick one task that passed the four questions.
  2. Do it by hand five times and write down each decision.
  3. Write the pass mark before the agent runs.
  4. Start with drafts. Our view: have the agent prepare work for your approval before it sends anything.
  5. Run one agent on one workflow for two weeks. Do not add anything else during that time.
  6. Check the misses each week against your written steps.
  7. Decide: keep it, fix it, or drop it. Only then pick the next task.

If the test fails, you lost two weeks and learned which part of the process was unclear. That is a small price. It is also easier to pay when you have a day job and one spare hour at night. For the wider picture on fitting this around work, see shipping side projects with a full-time job.

Where to go next

Pick the task, do it by hand, set the pass mark and keep the leash short. If you want one plain note a week on work, money and keeping AI agents in check, Get Sunday Night Truths. It is free.

Questions

How many tasks should my first AI agent handle?

One. Operater, which sells an agent platform, advises to deploy one agent on one workflow and run it for two weeks before adding anything. A second task can wait until the first has a result.

Should I automate content creation first?

Our view: no. The sources disagree: Operater says content pays off least, while FixerAI describes a large time saving, and both sell automation. Content reads as your voice, so start with a task that stays in drafts.

What if I have never done the task by hand?

Do it by hand first. Operater says automating a process you have never done manually almost always automates the wrong thing, and suggests doing it five times and writing down each decision.

How do I know the agent is working?

Set a pass mark before you start. Dust gives the example of capturing 95% of action items and decisions for a call summary agent. Operater suggests three hours returned per week as a minimum useful result.

Should an agent handle invoices or other money tasks?

Treat money as high stakes. Our view: let it draft reminders for your review and keep amounts and terms yours. Check with an accountant or other professional on anything financial or legal.