Hire an AI Agent Developer: Freelancer, Agency or DIY?

Hire a freelancer or fixed-scope service for one well-defined agent; an agency when it spans several systems and needs ongoing ops; build it yourself if it’s one workflow and you can read code. Whoever you pick, ask the same ten questions before you pay, and walk away from anyone who wants your agent to run on their Claude account.

Disclosure: Claudekit sells this service. We build custom AI agents priced by feature count, so we have a stake in your answer. To keep this guide fair, the checklist below applies to every option, and we answer it for ourselves in its own section, including the parts that might put you off. Prices were checked on October 25, 2026 and link to their sources.

The four options compared

Option	Fits when	Published prices we found	Main risk
Build it yourself	One workflow, you can read code, you’ll maintain it	Your Claude plan or API tokens	Your time, and nobody else understands it
Freelancer	One well-defined agent, you can describe each job	$50/hour; fixed builds $1,200–2,000 (Contra profile); $2,500 for one production workflow (Contra service)	Availability, and quality varies widely
Fixed-price service	You have a feature list and want a set price and terms	A$5,000 and up for an agent build (Sandlabs); from $3,000 for a Claude Code project (eSEOspace); $490–4,500 by feature count (Claudekit)	Anything outside the written scope costs extra
Development agency	Several systems, unclear scope, you need discovery and monthly support	$12K–22K for a single-tool chatbot MVP up to $60K–120K for a multi-agent system (Groovy Web); $15K–40K for a simple agent (Space-O)	Price, and long timelines

The full cost picture, including running and maintenance costs, is in our guide to AI agent development cost.

How to choose in one minute

• Can you write each job as one line, with an example input and expected result? If not, you need discovery first. An agency or a paid scoping session fits; a fixed-price service will turn you away or guess.
• Does it write to systems that hold customer or financial data? Look for a builder whose process includes a human code review, or plan to review it yourself.
• Who will fix it in month three? If nobody on your side can read code, budget for a retainer from whoever builds it.

Where to find a developer

• Freelance marketplaces such as Contra and Upwork. Search for the systems you use (“Shopify”, “HubSpot”, “MCP server”) together with “AI agent”, and read past work, not just ratings.
• Fixed-price services, which publish packages and terms on a pricing page. You can compare them without a sales call.
• Agencies that publish case studies in your industry. Expect a discovery call before any number.
• Communities where people share Claude builds, including the open-source repos behind tools in our directory. A developer whose code you’ve read is a safer hire than a profile.

Ten questions to ask any developer

Ask all ten in writing, and keep the answers with the contract.

1. What exactly is in scope? Ask for the list of jobs the agent will do, one per line. Vague scope is how fixed prices become hourly.
2. Who owns the code, and where does it live at the end? You want a repo you control, not access to someone else’s platform.
3. Whose API key and Claude account does it run on? It should be yours. More on this in the red flags below.
4. How is it tested? Ask whether there are automated tests for each job, and whether you get them.
5. Does a person review the code before delivery? Some builders do, some don’t. Know which you’re buying.
6. What will it cost per month to run, at my volume? Ask for the model, tokens per run and runs per month, then check the math with our API cost calculator.
7. What do you need from me? Example inputs, sandbox or test accounts, and who on your side answers questions.
8. What does the warranty cover, and for how long? Bugs against the agreed list, for a set number of days, is the usual form.
9. What happens after the warranty? Retainer, hourly or nothing. Get the price.
10. What won’t you do? A good builder has a list: regulated data, production credentials during the build, deadlines they can’t meet.

Red flags

• “We’ll run it on our Claude subscription.” Anthropic’s Claude Code legal page says developers building products should use API key authentication, and that third-party developers may not route requests through Free, Pro or Max plan credentials on behalf of their users. It also says companies that run Claude Code in their products may not pay for, resell or intermediate Claude usage on their end users’ behalf: each end user authenticates with their own credentials. An agent that depends on someone else’s subscription can stop working without notice.
• They ask for your Claude login. The same page says developers may not collect, store or intermediate Claude.ai credentials or session tokens.
• They want production passwords to start. A build should run against sandbox or test accounts. Production access comes at handover, set by you.
• No repo, only a hosted dashboard. If the business closes, so does your agent.
• No example inputs requested. If they don’t ask what “right” looks like, they can’t test it.
• Hourly with no cap on a defined job. Ask for a fixed price or a not-to-exceed figure.
• Claims of an official Anthropic partnership you can’t verify. Check the claim on Anthropic’s own pages, not the vendor’s.
• Use cases that break someone’s terms, such as scraping a site that forbids it or bulk messaging. Anthropic’s usage policy applies to whatever you build with Claude.

How Claudekit answers the ten questions

We sell custom AI agents, so here are our own answers, taken from our Hire page on October 25, 2026. Hold us to them.

Question	Our answer
Scope	Your feature list, one job per line. The count sets the tier: Starter 1–3 features ($790; $490 for the first 10 clients), Business 4–7 ($1,900; $1,190), Pro 8–12 ($4,500; $2,900). 13 or more: a fixed quote before any payment
Code ownership	You own it. We transfer the private GitHub repo, or give you admin access, after full payment
API key	Yours. We don’t resell or pay for Claude usage on your behalf
Testing	Automated tests for each feature, included in the repo
Human review	None. AI agents write the code and run the tests; no person reviews the code before it reaches you. That is how the price stays fixed
Running cost	You pay API and hosting costs. We don’t host or run the agent
What we need	Your feature list, an example input and expected result per feature, a sandbox or test account for each outside system, your Anthropic API key if the agent calls Claude
Warranty	14 days on Starter, 30 days on Business and Pro, for bugs against your feature list. 2 revision requests after delivery on every tier
After the warranty	No maintenance included. You can hire anyone to maintain the repo
What we won’t do	Regulated data such as health records or card data, production credentials during the build, deadlines shorter than the tier’s target, hourly work, and anything that breaks Anthropic’s usage policy or another service’s terms

Who that fits: someone with a clear feature list who wants a fixed price and a repo, and who can check the delivery against their list. Who it doesn’t fit: anyone who needs a human code review, hosting, ongoing operations or discovery to work out what to build. For those, a freelancer you work with closely, or an agency, is the better choice.

When you don’t need to hire anyone

• It’s one workflow and you can read code. Claude Code can help you build it; follow how to build an AI agent with Claude and run it on a schedule with headless mode.
• It’s answering questions from your documents. A Claude Project does it with no code.
• The steps never change. That’s an automation, not an agent. Anthropic’s own advice on building effective agents is to find the simplest solution possible, which “might mean not building agentic systems at all.”
• You aren’t sure what the agent should do yet. Do the job by hand for a few weeks and write down each step. That list is the spec any builder will need, and it may show the job is simpler than it looked.

Not sure which jobs are worth automating? Our list of AI agents for small business has twelve examples, each with a feature count and a DIY level.