AI agent and automation developers

How to hire someone to build an AI agent

To hire someone to build an AI agent, describe the job, the systems and data it will touch, what it must never do and how you will test it. Then ask for a working demo, hire in stages (prototype, pilot, launch) and budget for the model’s monthly usage costs. An AI agent is a program built on a language model that carries out tasks with some autonomy, such as sorting inquiries or updating your CRM.

Updated September 26, 2026

What people usually hire for

• An assistant for your team that answers questions from your knowledge base, policies and documents.
• A customer chatbot on your website, WhatsApp or Messenger that answers questions, books appointments and hands tricky cases to a person.
• A voice agent that answers calls, books appointments and passes the rest to your staff.
• Inbox and document processing: details from emails, forms, invoices and contracts extracted into your CRM or a spreadsheet.
• n8n, Make or Zapier workflows where a model does one of the steps, such as sorting support tickets by topic.
• An MCP server so Claude, ChatGPT and other agents can work with your service, for example search your catalog or place orders. MCP is the standard way AI assistants connect to outside tools.

What drives the price

• How many systems it connects to, and whether they have an API.
• How much data there is and how clean it is: scattered files and scans cost more than a tidy spreadsheet.
• What a mistake would cost. Where the agent talks to customers or touches money, it needs extra checks and a human in the loop.
• Where it runs: a hosted cloud service, your own server, or an on-premises environment with no outside access.
• Support after launch.
• Running costs: the model is billed by usage (tokens, roughly the amount of text processed), plus hosting and paid tools, so it costs money every month it runs. Ask the developer to estimate it at your volume.

What to put in your request

1. The job and the result: what the agent does, for whom and how often.
2. Where it gets its data and where it writes the result.
3. What it must never do, for example promise discounts or send emails no one has reviewed.
4. Volume: how many messages or documents a day.
5. Privacy rules: where data may be stored and which model providers are allowed.
6. Ownership: the code, prompts and workflows are yours, running on accounts in your name, with documentation and a handover.
7. Who maintains it once it is live.
8. How you will accept the work: 10–20 typical cases, anonymized, that it must handle correctly.

How to vet a developer

• Ask to see something working, not only slides.
• Ask how they test quality: which examples they run and how they catch made-up answers.
• Ask what the agent does when it is unsure of an answer or gets a question outside its job.
• Ask how its access is limited: read-only where possible, and your approval before it sends, pays or deletes anything. Ask how it resists prompt injection, such as an email telling it to ignore its instructions.
• Ask how you will see what it did: a log of every action and a way to correct mistakes.
• Split the work into stages: prototype, pilot on real tasks, launch. Pay per stage too.

Don’t hand over real customer data before you have a contract (and, in the EU or UK, a data processing agreement). Anonymized examples are enough for a prototype.

The board doesn’t vet developers or hold money: you agree and pay directly. Our guide on how to agree and pay safely covers deposits and milestones.