Automate

Use AI agents to automate repetitive work

Map your workflows, identify bottlenecks, and deploy AI agents that handle research, writing, and operations.

Who this is for

You are a solo founder spending too much time on tasks that do not require your judgment: drafting routine emails, summarizing research, populating content, responding to common customer questions, updating spreadsheets, or generating reports.

What problem this solves

The difference between AI tools and AI agents:

AI tool: You do something and ask AI to help. You are in the loop.
AI agent: You define a workflow and let AI execute it. You review the output.

This playbook teaches you to identify which tasks are worth turning into agents and how to build those agents without a technical background.

Step 1 — Audit your week with a time log

Before you can automate anything, you need to know where your time goes. For one week, log every task over 15 minutes with:

What you did
How long it took
Whether it required your unique judgment or could have been done by a competent assistant

After one week, you will have a clear picture of your automation opportunities.

Step 2 — Identify agent-ready workflows

Not all tasks are worth automating. The best tasks to convert to agents have these characteristics:

High-frequency: You do this task more than once per week.

Low-judgment: The task follows a clear pattern or set of rules. “Respond to customer questions about pricing” is a good agent task. “Decide whether to invest in a new product direction” is not.

Time-consuming: The task takes more than 30 minutes when done manually.

Examples of agent-ready tasks:

Draft responses to common customer questions
Summarize articles or research for your content
Generate social media posts from a blog article
Update a CRM or spreadsheet with new contact information
Create weekly status reports
Draft outreach emails to potential partners

Step 3 — Map the workflow before you build the agent

For each task you want to automate, write out the exact steps:

What triggers this task? (new email, new row in spreadsheet, scheduled time)
What information does the agent need? (where does it come from)
What does the agent produce? (draft email, updated record, summary)
What happens to the output? (you review it, it goes directly to the recipient)
What are the error conditions? (what should the agent do if it cannot complete the task)

This map becomes the agent’s instruction set.

Step 4 — Start with prompt-based agents before code-based agents

You do not need to write code to have an AI agent. Start with prompt-based workflows using ChatGPT or Claude:

Example — Research summarization agent:

Every Friday, take the 10 articles I saved in Readwise and:
1. For each article, summarize the key insight in 3 sentences
2. Identify one specific action I could take based on this insight
3. Output a formatted list I can review in 10 minutes
4. Flag any articles that seem outdated or unreliable

Give this to Claude with access to your reading list and it produces the output without you doing any work.

Example — Content repurposing agent:

Take the following blog post and generate:
- 3 Twitter/X posts (under 280 characters each)
- 1 LinkedIn post (150–200 words)
- 1 email summary (100 words) suitable for my newsletter
- 3 potential headline variations

Keep the tone consistent with my existing content style.

Step 5 — Layer in tools for real automation

When prompt-based agents are working, layer in automation tools to remove the manual trigger:

Zapier or Make: Connect apps so the agent runs automatically when something happens (new email → agent processes it → result goes to your inbox)
Scheduled prompts: Use a weekly recurring prompt in ChatGPT/Claude for recurring workflows
Notion/Airtable databases: Use them as input/output tables for agents that process structured data

Step 6 — Review and iterate

AI agents are not set-and-forget. Review every agent output for the first 2 weeks:

Is the quality acceptable?
Are there patterns of errors?
What edge cases is the agent missing?

Adjust the prompt based on what you observe. The first version of any agent prompt will be wrong — the iteration is where it becomes reliable.

Common agent types for solopreneurs

Agent type	What it does	Tools
Research summarizer	Summarizes articles, videos, podcasts	ChatGPT, Claude
Content repurposer	Converts one piece into many formats	ChatGPT, Claude
Email drafter	Drafts responses to common questions	ChatGPT, Claude
Outreach drafter	Generates personalized cold outreach	ChatGPT, Claude
Report generator	Creates weekly status reports	ChatGPT + Zapier
Social scheduler	Generates and schedules posts	ChatGPT + Buffer

Risks and tradeoffs

Automating tasks that need judgment: if a task requires nuanced decision-making, an agent will make errors that cost more than the time saved
Not reviewing outputs: always review agent outputs in the beginning — assuming they are correct leads to embarrassing or costly mistakes
Over-engineering: building a complex multi-step agent when a simple prompt would have done the job wastes more time than it saves
Losing the learning: if you automate research gathering without reading the summaries, you lose the depth that comes from engaging with the material

Weekend checklist

Log your time for one full week (every task over 15 minutes)
Identify your top 3 highest-frequency, lowest-judgment tasks
For the top task, write out the exact workflow in steps
Write a prompt-based agent instruction for that workflow
Run it manually this week and note the output quality
Based on what you learned, decide whether this needs a tool-based automation

FAQ

Do I need coding skills to build AI agents? No. Most agent workflows can be set up using prompt engineering + automation tools like Zapier or Make. Code is needed only for complex multi-step agents or custom integrations.

What is the biggest mistake solopreneurs make with AI agents? Automating tasks that need judgment and then not reviewing the outputs. Agents are reliable for pattern-following, not for nuanced decisions.

How many agents should I start with? Start with one. Get it working reliably, then add the next. Trying to automate five workflows at once leads to none of them working well.

What tasks should I never automate? Anything requiring legal judgment, financial decisions, personal relationships, or creative direction that defines your brand. These require your judgment and usually a human touch.

Dan Koe — uses AI systems to run his entire business without a team
Justin Welsh — lean systems approach to solo operations

Next step

Pair this with the Content Engine playbook — the automation systems you build here free up time for the content work that drives distribution.

← Back to Playbooks