Free guide · 7-page PDF

How to Build AI Agents: The Free Guide for Business Owners

How to Build AI Agents is a free 7-page PDF for business owners. It shows which manual tasks are worth handing to an AI agent, how to build your first one in Claude in four steps (a project, instructions, connectors and a trigger), where the DIY route hits its limits, and how the done-for-you route compares.

6workflows
4build steps
2paths compared

By Talk to Me Data · Updated October 2, 2026

It's free. I'll send it straight to your inbox.

Free download

Get the guide in your inbox

Enter your details and we'll email you the 7-page PDF right away.

Free PDF by email. No spam, unsubscribe any time.

Inside the PDF

What's inside the guide

01
6 workflows worth automating
From invoice processing to lead research, each with a manual versus agent breakdown so you can see the time it takes today.
02
Your first agent in Claude
Create a project, add instructions, connect your tools with connectors and run the workflow. Includes an example instruction you can copy.
03
The honest DIY limits
Session and weekly usage limits, agents stalling mid-task, API keys and hosting. Know where building it yourself breaks before you rely on it.
04
The done-for-you route
A side-by-side comparison of DIY in Claude and a hosted, managed agent, so you can pick the path that fits.

Which manual tasks can an AI agent take over?

Any task that is clearly defined, predictable and repetitive. If you could write the steps down once and hand them to someone forever, an agent can do it. The guide breaks down the most common ones:

Manual tasks compared with an AI agent
Workflow	Done by hand	With an agent
Invoice processing	20–40 min per invoice, typed into accounting software	Reads, categorizes and posts each invoice
Social media content	2–3 hours a week turning one idea into posts for every platform	Drafts, formats and schedules per platform
Lead research	10–15 min per lead to find, check and write an opener	Researches and enriches hundreds of leads overnight
Follow-ups and outreach	Inconsistent, or forgotten	Timed, personalized sequences that always go out
CRM and data entry	5–10 min per contact after every call	Logs and updates records in real time
Review requests	Hit-or-miss timing	Sent at the right moment, every time

How do you build an AI agent in Claude?

You can build a first agent in Claude in four steps:

1Create a new projectOpen Projects in Claude and create one. Name it after the workflow, like "Invoice Processor" or "Lead Research Agent". The project holds the instructions and context for this one agent.
2Add your instructions and knowledgeAdd project instructions and upload any reference files. Describe the task, the output format, the rules the agent must follow and examples of what good looks like.
3Connect your tools with connectorsGo to Customize, then Connectors, click + and turn on the tools your workflow needs, such as Google Drive, Gmail or Notion. Connectors are built on MCP (Model Context Protocol), so Claude can read from, write to and take actions in those apps.
4Start a chat in the project and run itOpen the project and start a new chat. Claude follows your project instructions automatically. Give it a trigger, such as "Process the invoices in the /Invoices/Pending folder", and it works through the task with the connected tools.

The PDF includes a full example instruction for a social content agent. For the long version, with three ways to build (no-code in Claude, n8n or Make, or code), read our step-by-step guide to building an AI agent.

What are the limits of building an AI agent yourself?

Building it yourself works well for small jobs you can watch. On long or always-on workflows, it hits three walls:

Usage limits. Claude's paid plans have a five-hour session limit and a weekly limit. A long job, such as 100 invoices or 200 leads, can use up a session, and the agent stops mid-task.
It only runs while you do. The agent works inside your chat session, so it isn't running overnight, and it uses the same allowance you need for your own work.
Running it properly is a project. Moving it out of the chat app means managing API keys, rate limits, hosting, error handling and monitoring.

Should you build it yourself or have it built?

Build it yourself if you want to learn and the task is small enough to run while you watch. Have it built if the agent needs to run every day, at volume, without you.

Building an AI agent yourself in Claude compared with a done-for-you agent
What you need	DIY with Claude	Done for you
No usage limits to manage	Session and weekly limits apply	Runs on our infrastructure
Runs around the clock	Only while you're running it	Always on, fully hosted
No setup	Projects, connectors and instructions	Describe it and we build it
Your Claude stays free for you	The agent uses your plan's usage	Separate, so use Claude freely
No API keys to manage	Needed to run outside the chat	Fully managed
Monitoring and updates	Your responsibility	Included

See the agents we build and run for businesses, or check where an agent would help most first with the AI Agent Readiness Audit.

Frequently Asked Questions

Can I build an AI agent without coding?

Yes. In Claude you can build one with a project, written instructions and tools connected through connectors, without writing any code. The guide walks through each of the four steps with an example instruction you can copy.

What is MCP?

MCP (Model Context Protocol) is an open standard that lets an AI model connect to real apps and data, such as Gmail, Google Drive, Notion or QuickBooks. Once a tool is connected, the agent can read from it, write to it and trigger actions in it.

Which task should I pick for my first AI agent?

Pick one task that happens often, follows the same steps every time and has a clear output: processing invoices, turning a blog post into social posts, or updating your CRM after calls. Avoid tasks that need a judgment call on every run.

Why does my agent stop halfway through a task?

Usually because it hit a usage limit. Claude's paid plans have a five-hour session limit and a weekly limit, and a long job, such as working through 100 invoices, can use up a session in one go. The agent stops wherever it is. Smaller batches help; running the agent on dedicated infrastructure removes the limit.

How is this guide different from the blog post on building an AI agent?

The PDF is a short, printable guide for business owners deciding whether to build an agent themselves or have it built. The blog post is the longer walkthrough, with three ways to build an agent (no-code in Claude, n8n or Make, or code) and example instructions to copy.

Keep going