How to hire an AI developer: write a useful brief
Start with work your team can describe from beginning to end, such as taking an inquiry, checking availability and preparing a draft reply. Pick a repeated process with clear inputs and a person responsible for exceptions. Keep disputed business rules, irreversible decisions and poorly understood data out of the first automatic release.
Before you hire an AI automation expert
Explain where judgment is needed and where a fixed rule already exists. A calendar check or duplicate-payment check should not depend on a model guessing correctly; AI agent development makes sense when the task needs context and controlled tool use. Our AI automation hub covers the different types of work.
Include these details in the job post
- Describe the current process.Who starts it, what they do manually and where requests get stuck.
- Name the systems involved.List your CRM, calendar, payment tool and any existing automation.
- Prepare representative data samples.Use anonymized examples, including missing fields and unusual cases.
- Define what done means.Describe the expected output, approval point and failure notification.
- State a budget range and priorities.Separate the essential first release from features that can wait.
- State ownership and handover requirements.Your accounts, source repository, documentation and access recovery.
How to hire an AI developer: verify the proof
Ask for a similar project, the candidate's personal contribution and a demonstration of the relevant workflow. A polished homepage shows presentation; you also need to see what happens when someone submits a request, the system reads data and something fails. If confidentiality prevents a live demonstration, ask for a sanitized walkthrough with the same operational decisions.

Our role in WLTR.AI was the product build. In an interview about a business-data assistant, ask the builder to explain access boundaries, which records support an answer and how a user checks a disputed total. Those questions reveal more than a list of model names.

See every cluster, budget and screening theme from the same sample in our AI automation hiring report 2026.
↑ Back to topFreelance AI developer vs agency vs in-house
Choose who will own delivery and ongoing decisions before comparing hourly rates. Each arrangement can work; the practical difference is how much coordination, coverage and product knowledge your business must supply.
Freelancer
Useful for a bounded task when you can provide direction and review. Ask about availability, documentation and who can take over if the developer becomes unavailable.
Agency
Useful when discovery, integrations, interface work and launch need coordination. Confirm the named delivery lead, who actually builds the system and what support the quote includes.
In-house
Useful when the product changes continually and requires daily business context. Plan for recruitment, technical supervision, operational coverage and access to specialist help.
AI developer interview questions grounded in real jobs
In 100 recent Upwork job posts we reviewed in October 2026, 32 had screening questions, with 111 questions in total. Of those posts, 26 asked for proof of real past work and 16 posed technical scenarios. This is our review of a job-post sample, not a survey of completed projects or a success-rate benchmark.
- Posts asked about testing and rollback
- 9
- Asked how AI-generated code is reviewed
- 5
- Asked about availability and time zone
- 6
Our October 2026 review of 100 recent Upwork job posts.
About 8 posts asked about budget, rate or monthly running costs. Buyers also asked candidates to describe a similar project, identify what they personally built, link a project and demonstrate it. Copy that approach: request something inspectable, then ask a follow-up about a decision the person made.
A payment webhook arrives twice. What happens? Two Generate clicks, same second, credits for one. Ad conversions drop; analytics look flat. Why? Give an AI character memory without full history. How would you cap AI cost per user? Which token fits an unattended hourly job? One account fails. Can the other accounts finish? AI-written PR: 3,000 lines, no real DB test. Ship?
Questions to ask an AI developer: good answers
These are paraphrased interview prompts based on the reviewed posts. You do not need to judge the code yourself: ask what the customer would experience, what is recorded and how the candidate would demonstrate the safeguard. A strong answer names assumptions before prescribing a tool.
| Question | Good answer mentions |
|---|---|
| Duplicate webhook? | Verify, deduplicate, log |
| Credits for one, two clicks? | Atomic credit reservation |
| Conversions drop? | Trace click to CRM |
| Character memory? | Scoped facts and retrieval |
| AI cost grows? | Meter, cap, stop |
| Hourly job token? | Server identity, limited access |
| One account fails? | Isolate, retry, report |
| Untested AI-written PR? | Review, test DB, rollback |
Payments and credits need enforceable rules
For duplicate payments, expect signature verification, stored event identifiers and a record of completed work. Stripe's webhook guidance covers duplicate deliveries; Stripe's idempotency documentation covers safely retrying supported API requests. These are separate protections.
For simultaneous Generate clicks, the answer should reserve credits atomically, meaning competing requests cannot both spend the same balance. Ask how failed work releases or refunds the reservation and how a retry finds the original job. Disabling a button alone does not protect the balance.
Ask where the information went
For an advertising discrepancy, expect the developer to trace consent, click identifiers, events and CRM outcomes before changing tags. For character memory, ask what facts are stored, how relevant context is retrieved and how users correct or delete it. Memory must stay within the right user's data boundary.
Ask how unattended work stays controlled
Listen for per-user usage measurement, enforceable spending limits and a defined stop when the allowance runs out. For an hourly job, the candidate should first identify the service and its supported authentication, then explain a server-side identity, limited permissions and renewal. Supabase's API key guidance distinguishes public keys from backend secrets, which must remain private.
A failing account should produce an isolated failure, bounded retries and an actionable report while other accounts continue. For the untested AI-written pull request, expect review in manageable parts, a representative test database and critical-path tests before release. Ask how database changes would be recovered if reverting application code is insufficient.
↑ Back to topGreen flags, red flags and approval rules
Ask the candidate to draw the workflow and mark where it reads private information, changes records or contacts a customer. Start with drafts or read-only access when a mistake would be expensive. Give a named person approval responsibility and agree what happens while that person is unavailable.
- Request or scheduleA person asks, or the agent wakes up on a timer.
- Read contextApproved documents, CRM records, past conversations.
- Plan and use toolsSearch, read a record, draft a reply, calculate. Each tool has narrow permissions.
- Ask before risky actionsMoney, deletions and messages to customers wait for a human yes.
- Act and logThe result, the sources used and the cost of the run are saved.
Do: look for these green flags
- Asks about your process before quoting.
- Writes scope, assumptions and exclusions.
- Explains running costs and spending controls.
- Uses your accounts and source repository.
- Demonstrates logs, tests and approval points.
Don't: accept these red flags
- Guarantees business results without evidence.
- Quotes without asking how the process works.
- Requests passwords by email.
- Makes everything AI-driven without justification.
- Treats a successful demo as enough testing.
Budgets, running costs and realistic scope
In 100 recent Upwork job posts we reviewed in October 2026, 28 fixed-price posts ranged from $100 to $11,000, with a median of $1,850. The 51 hourly posts that supplied a range had a median range of $25 to $47 per hour. These are posted budgets, not final contract prices or Fer Marketing prices.
- 1. Build (one time)Discovery, development, testing and handover.
- 2. Platform subscriptionsAutomation tool plan, hosting, database, vector store, phone or WhatsApp provider.
- 3. AI usageBilled per token or per minute by the model provider. Grows with volume.
- 4. Maintenance and changesAPIs change, prompts drift, your process evolves.
- 5. Your team's timeReviewing approvals and exceptions. Plan for it.
| Cost | Ask for |
|---|---|
| Platforms | Accounts and billing owner |
| AI usage | Volume assumptions and cap |
| Maintenance | Included work and exclusions |
| Team time | Approvals and exceptions |
Request low-volume and busy-period estimates based on your expected requests, document sizes and tool calls. Include retries, failed jobs and the staff time spent reviewing exceptions; ask who sees alerts before spending escalates. Our AI automation cost guide explains the cost categories to put beside each quote.
↑ Back to topA hiring process with inspectable milestones
Use a staged commitment when the workflow or existing code is uncertain. Paid discovery should produce useful decisions and a clearer scope; a small first milestone should prove the hardest assumption. Agree on its deliverable before paying for it.
- Step 1
Write the brief
Describe inputs, systems, approvals and the desired result. Include anonymized examples and the cases your team currently handles manually.
- Step 2
Shortlist on proof
Inspect relevant work and ask what the candidate personally built. Discuss scenario questions, availability, time zone and who will answer during delivery.
- Step 3
Use discovery or a small first milestone
Resolve uncertain permissions, data quality or an integration before committing to the full build. Require a written finding, working slice or explicit go/no-go recommendation.
- Step 4
Agree on the written scope
Record inclusions, exclusions, accounts, operating assumptions and ownership. Define change requests, support responsibilities and the evidence required for acceptance.
- Step 5
Release against acceptance tests
Check normal use, duplicates, missing data, permission failures and recovery. Approve each milestone from demonstrated behavior, then receive documentation and a handover.
Ownership, handover and white-label delivery
A handover should let your business operate the system and hire another maintainer. Request the repository, account inventory, setup instructions, test evidence and a short walkthrough of alerts, pausing and recovery. Clarify what support covers and how future changes are quoted.
Client-owned delivery
At Fer Marketing, source code, accounts and documentation belong to the client. Every project gets a fixed written quote before work starts; payment is by card through Stripe after scope is agreed.
For agencies: white-label
We work under your agency's brand, with NDA and non-solicitation available. Use our white-label development service to agree on client communication, review and handover.
Fer Marketing is operated by Fer LLC in Wyoming. Founder Fernando Marcano is an IT engineer with 20+ years of hands-on IT experience; our homepage shows 100+ Upwork projects and 100% Job Success. We serve the US market in English and Spanish.
Send your brief through contact, WhatsApp or phone. We reply the same business day, 9 AM to 9 PM ET, and use your process and constraints to define what should be built first.
↑ Back to topWe can build this for you
Want us to build your automation or AI product?
Send the workflow, systems and result you need. We will define a practical scope and a fixed written quote before work starts.
- Written scope with assumptions and exclusions
- Agreed automation, agent or integration build
- Acceptance tests, logs and approval controls
- Client-owned accounts, source code and documentation
- Handover covering operation, alerts and recovery
Frequently asked questions
How do I hire an AI developer without technical knowledge?
Describe the business process and ask for a similar working project. Have candidates explain failures, running costs and approval rules in plain language, then use written acceptance tests to judge delivery. For a high-risk system, include an independent technical review in your scope.
What are the best AI developer interview questions?
Ask what the candidate personally built, how duplicate requests are handled and how spending is limited. Ask for a demonstration of testing, permissions and recovery. A useful answer explains the business outcome and a way to verify it, rather than listing tools.
Should I hire a freelance AI developer or an agency?
A freelancer can suit a bounded task when you can manage direction and review. An agency can suit work needing coordinated product, integration and launch responsibilities. Compare the named people, scope, availability, handover and support before choosing.
How much does it cost to hire an AI automation expert?
The useful comparison is a quote for the same defined result, plus operating costs. Data quality, permissions, integrations and acceptance tests affect the work. Fer Marketing gives a fixed written quote before work starts; posted marketplace budgets are not final contract prices.
Should I pay for discovery before the build?
Discovery is useful when important assumptions remain unresolved, such as access to an API or the condition of an existing app. Agree on a concrete output, such as a validated integration, risk list and proposed scope. Avoid discovery with no defined deliverable.
Is using AI-generated code a red flag?
Ask how the developer reviews and tests it. The important evidence is reviewed changes, tests against a representative database, permission checks and a release recovery plan. Generated code still needs a person responsible for its behavior.
Who should own the accounts and source code?
Put ownership in the written scope and keep account recovery under your control. At Fer Marketing, source code, accounts and documentation belong to the client. Collaborator access should be removable without losing the system or its history.
Can an agency hire Fer Marketing under its own brand?
Yes. We offer white-label delivery under the agency's brand, with NDA and non-solicitation available. Agree on the client communication route, approval responsibilities, deliverables and handover before the project begins.
Contact us for a fixed written quote and a same-business-day reply, 9 AM to 9 PM ET, with white-label delivery available for agencies.