How to Hire an AI Agent Developer: 12 Questions, Red Flags & Contract Clauses (2026)
Picking someone to build your AI agent is a strange kind of purchase. You can't see the product yet, every proposal sounds confident, and the quotes can be wildly far apart. Meanwhile every vendor now says "agentic".
That last part is a real problem. Gartner calls it agent washing: relabelling chatbots and scripts as agents. Its estimate was that only about 130 of the thousands of self-described agentic vendors were the real thing (Gartner didn't publish its method, so take that as directional).
This guide gives you a way to cut through it: what to ask, what a good answer sounds like, and what to put in the contract.
What the failure numbers actually say
Read these carefully, because they're often misquoted:
- MIT's "95%" means pilots with little or no measurable profit-and-loss impact, from a mix of 150 leader interviews, a survey of 350 employees and 300 public deployments. It is a generative AI figure, not an agent-specific one, and it has been debated.
- Gartner's 40% is a forecast, not a measurement. Its stated causes are rising costs, unclear business value and inadequate risk controls. Model quality isn't on the list.
- S&P's 42% comes from a survey of 1,006 IT and business professionals in North America and Europe. The same survey found the average company scrapped 46% of its AI proofs of concept before production.
The useful finding is on the buying side. Per Fortune's coverage of the MIT report, projects bought from specialised vendors or built through partnerships succeeded about two-thirds of the time, while internal builds succeeded only about a third as often. Picking a good partner matters more than picking a good model.
First: do you need to hire anyone?
An honest guide has to say this. Many "AI agent" projects shouldn't be custom builds.
Related: our comparisons of AI agent platforms for business, Meta's Business Agent vs a custom WhatsApp agent, and when you need an MCP server.
// ai integration & agents
Ready to add AI to your product?
We build AI agents, RAG chatbots and automation pipelines for businesses — scoped and shipped fast.
12 questions to ask before you sign
Ask all twelve. Look at how they answer, not only what they say.
Proof: can they build and ship?
| # | Ask this | A good answer sounds like | A worrying answer sounds like |
|---|---|---|---|
| 1 | Can I see a live system with real users, not a demo? | A live URL or store link, plus a past client you can call | Slide decks, prototypes, or "everything is under NDA" |
| 2 | Who will actually build it, and can I meet them before signing? | Named engineers; the person on the sales call is on the project | "Our pool of developers", sales only |
| 3 | What went wrong on a past project, and what did you change? | A specific failure and a specific fix | "All our projects succeed" |
| 4 | Is this really an agent problem? | They'd talk you down to something simpler if it's cheaper | Everything is "multi-agent" |
Build quality: will it work when it counts?
| # | Ask this | A good answer sounds like | A worrying answer sounds like |
|---|---|---|---|
| 5 | How will you measure accuracy before launch? | A test set of real examples and a pass mark agreed with you in advance | "It's very accurate", demos only |
| 6 | What happens when the agent is wrong or unsure? | Human hand-off, confirmation before risky actions, full logs | "It won't be wrong" |
| 7 | Which models does it depend on, and how do we switch? | Named models, a fallback, and a plan for price changes | "The latest AI technology" |
| 8 | How do you handle our data and compliance? | Where data goes, how long it's kept, which rules apply (GDPR, EU AI Act, Australian privacy law) | "Bank-grade security" with no detail |
If compliance matters to you, see our notes on the EU AI Act timeline for agents and Australia's automated decision-making rules.
Money and ownership: what will it really cost?
| # | Ask this | A good answer sounds like | A worrying answer sounds like |
|---|---|---|---|
| 9 | What will it cost to run each month, not just build? | Line items: model usage, hosting, monitoring, maintenance | A quote that covers the build only |
| 10 | Who owns the code, prompts and data, and where does the code live? | IP assigned to you on payment; repo in your organisation from day one | A "licence", or code delivered in a zip at the end |
| 11 | What's in the contract after go-live? | A support window, bug-fix response times, handover documents | The contract ends at launch |
| 12 | How do you price, and what happens when scope changes? | Short discovery first, then fixed units, with a written change process | One big fixed price after a 20-minute call |
Fixed price, hourly or retainer?
Sources agree there's no single best model. The answer depends on how well defined the work is. The most common sensible structure is a mix.
Why not one fixed price for everything? AI work surfaces surprises such as messy data, and every surprise becomes a change order. Why not open-ended hourly? Without milestone gates and weekly reporting, costs drift.
Green flags and red flags
A scorecard you can copy
Score each vendor 1 to 5 on each row, multiply by the weight, and add up. Under 3.5 out of 5 is a pass.
| Criterion | Weight | What a 5 looks like |
|---|---|---|
| Production proof | 25% | Live system, real users, reference call completed |
| Evaluation method | 20% | Test set and pass mark agreed before build |
| Ownership and contract | 20% | IP to you on payment, repo in your org, post-launch terms |
| Team transparency | 15% | Named engineers you've met |
| Run-cost clarity | 10% | Monthly running costs itemised |
| Security and data | 10% | Clear data handling and compliance answers |
What AI agents cost (agency-published estimates)
These are marketing benchmarks from agencies that sell this work, not quotes. Use them to spot a quote that's wildly off, in either direction.
Bar length uses the midpoint of each range. Sources: agency cost guides (RaftLabs, Azilen, GroovyWeb and others). Not independently verified.
Two cost traps to ask about:
- Running costs. Guides suggest budgeting roughly $500 to $5,000 a month to run a first production agent (model usage, hosting, monitoring, upkeep). One source says first-year cost typically lands 40% to 80% above the build quote.
- Scope drift. Integrations, compliance and traffic move the price more than the choice of model.
Start with our free app cost calculator if you also need an app or dashboard around the agent, and see AI app development cost in 2026.
The contract checklist
Before you sign, make sure these are in writing:
- IP assignment to your company, effective on payment (not "on completion").
- Code in your repository (your GitHub or GitLab organisation) from day one.
- Prompts, evaluation test sets and data belong to you.
- Model and API accounts in your name, so you pay the provider directly and can switch.
- Named personnel, with notice before senior people are swapped out.
- Acceptance criteria for each milestone, tied to payment.
- Post-launch support window with response times, plus handover documentation.
- Data handling terms, including where data is processed and how it is deleted.
- An exit clause, so you can leave with working code and documentation.
If you've already been burned, read our guides on rescuing a failed outsourced project and what to do when a developer abandons your app.
How we work (so you can hold us to it)
We build custom AI agents and AI automations, starting from $2,000, with typical delivery in 4 to 6 weeks. We're also happy to be measured against the checklist above:
- You own 100% of the code and IP, and it lives in your repository from day one.
- Every engagement is NDA-backed, with weekly demos and milestone payments tied to working software.
- A low-risk trial week before you commit.
- Live proof you can check: we built Affortable AI, an AI marketplace, in 160 hours (now 18,000+ users), and an AI lead-generation machine using n8n with AI email and voice calls.
Want to test us? Ask us all twelve questions. Book a free 30-minute call and we'll answer them, including telling you if a simpler option fits better than a custom build.
Sources: Gartner press release, 25 June 2025 (figures as reported by Digital Applied and others). MIT NANDA, "The GenAI Divide: State of AI in Business 2025", via Fortune and trade coverage. S&P Global Market Intelligence survey via CIO Dive. Cost ranges are agency-published estimates. This is general information, not legal advice.
// ai integration & agents
Ready to add AI to your product?
We build AI agents, RAG chatbots and automation pipelines for businesses — scoped and shipped fast.
