Two people run the same search in the same week and want opposite things. One sells a product customers already pay for and wants a model, a chat window or a scoring pipeline built into it. The other has an app that a tool wrote from prompts, it is live, something inside it is wrong, and they need somebody who can open the generated code and say what is actually in there. Google hands both of them nearly the same nine results.
Two jobs share the name AI developer. One puts AI inside the product your customers use. The other takes over the product an AI builder wrote for you. None of the three seller pages on that search that allowed an automated read on 29 August 2026 mentioned the second job.
Which one you are is settled by a single question, and the answer changes who you write to, what you ask them to show you, and whether any price you are quoted means anything. If you typed hire an AI app developer because a tool built your app and you now need a person, the second half of this page is the part written for you.
For this page three Google result sets were pulled on 29 August 2026, for hire ai developer, hire ai app developer and ai app developer; the seller pages on them that allowed an automated read were read in full that day, the marketplaces refused and are credited by address instead; Lovable’s own GitHub sync documentation was read the same day; and the reading list below was drawn from the 26 applications AxonBuild audited in June and July 2026. Nothing was installed, signed up for or run for this page.
The question that decides which developer you need
Ask it in these words: is the AI something your customers use, or is it the thing that built what your customers use?
If your customers use it, you want the first job. The work is a model, an API account, a prompt that has to behave, a pipeline that has to keep running, and a bill that grows with traffic. This is the job the market has named. It is what people mean by hire an AI engineer, and it is the same job when the wording is hire a generative AI developer or ai engineer for hire. The candidates advertise Python, TensorFlow, PyTorch, natural language processing and computer vision, and the marketplace filters are built around exactly those words.
If the AI built it, you want the second job, and it is a different purchase. Nothing new is being invented. There is a working application, written by a tool from prompts, carrying real users or real money, and the job is to read what the tool produced, find what is wrong in it, and change that without breaking the parts that work. The candidate never has to train anything. They have to be able to read fast and be right about what they read.
| AI inside your product | The app an AI builder wrote | |
|---|---|---|
| What you are asking for | A feature your customers interact with | Someone to take over code a tool generated |
| What the work touches | A model account, prompts, a pipeline, a per-request bill | Your database rules, your endpoints, your keys, your deploy |
| What they must show you | Something similar they built and still run | That they can read your repository out loud, today |
| How you know it worked | The feature does the thing, at a cost you can carry | The named failure is gone and nothing else moved |
| Who sells it | AI development companies, staffing shops, marketplace listings | Individual developers and small teams that only take this work |
The market has already made this split, though not in a way that helps you filter a shortlist. Paid roles are being advertised for people who work with these tools every day, and the buyers say in the advert itself that the opening is something other than a conventional developer role. Some of those adverts use the word AI-native. Nobody publishes the test for it and nobody can fail it, so it removes no candidate from your list.
The person who prompts a builder and the person who reads what the builder wrote are two different hires, and the first one is a search of its own. Hiring for an app with no AI anywhere in it, and none behind it either, is the general version of this question. Adding one AI feature to an app that already runs is priced and checked differently from hiring somebody whose whole job is AI. And if what you want is somebody who does only the second job all day, the people who do only this already have a name for it.
What page one is selling when you hire an AI developer
The search returns nine organic results and eight of them are sellers. The ninth is a forum thread from a small business owner asking where to find a talented AI developer, which is the reader this page is written for, sitting on page one with no answer above it.
From their search snippets and the three that could be read in full, the seller pages share one shape: a hero promising vetted AI talent, a skills grid, a list of industries, a short list of ways to buy, and a contact form. Nothing on the three that were read is dishonest. They are listings and pitches rather than answers, and they are built to catch the phrase hire ai developers and every neighbour of it, including hire artificial intelligence developer, ai developer for hire and hire remote ai developers.
Three of them opened to an automated read on 29 August 2026 and were read in full. LeewayHertz, at leewayhertz.com/hire-ai-engineers/, names three ways to buy: a dedicated development team, team extension, and a project-based model. It publishes no price anywhere on the page and routes you to a form. Arc, at arc.dev/hire-developers/ai, publishes numbers about itself: its own page says Top 2%, runs a live counter of remote AI developers and experts available to hire that moved between two reads on 29 August 2026, and says freelance matches take 72 hours and full-time hires 14 days. Those are Arc’s figures about Arc, quoted here as what the page claims rather than as a fact about the market. Appinventiv, at appinventiv.com/blog/how-to-hire-an-ai-developer/, is the only real article on the result set: eleven headings, a skills matrix, a six-step hiring process that runs from working out what you are building to checking the person actually fits, and a section on cost.
Three marketplace category pages sit across these searches, two of them on this one, and all three refused an automated read, so nothing on them is quoted here at all, including the hourly bands their search snippets advertise. They are credited by address: upwork.com/hire/ai-developers/, toptal.com/developers/artificial-intelligence and fiverr.com/categories/programming-tech/ai-coding/ai-applications. Devico, Uptech and Agicent hold the rest of the page as staffing shops and agencies, and are named here for the same reason. One result sits at an address for hiring an AI developer while its listed title advertises agent developers, which is a good sign of how blurred this term has become. Wiring an agent into a business, where the real questions are which tools it can reach and whose bill it can spend, is a different hire from this one.
The gap is the same on all three. Not one of them mentions taking over, inheriting, repairing or continuing an application that an AI builder already generated. Appinventiv does not use the words Lovable, Replit, Base44, Bolt or Bubble anywhere. That is a documented absence on those three pages on that date rather than a claim about every seller in the market. Everybody on page one reads AI developer as the developer puts AI inside your product. The reader who means an AI wrote my product and I need somebody who can read it is not being served by any of them.
Two limits on the evidence above. The pulls report that an AI Overview, a People Also Ask block and related searches were all on the live page; the endpoint used returns organic results only, so what those blocks said is not known here. And the question set recorded for hire ai developer carries no People Also Ask block at all. For the older phrasing, hire ai app developer, it records four questions and every one of them is about money. Buying this from a company instead of from a person changes what you are quoted and who answers the phone when it breaks, and that comparison, along with every price question in this family, belongs on pages that own it rather than here.
What they have to be able to read in your repository
The test below works because you do not have to understand the answers. You only have to notice whether the person can give one about your code rather than about code in general. Ask all five with your repository open in front of the candidate.
The counts below come from the 21 third-party applications inside the 26-application study AxonBuild audited in June and July 2026. Confirmed means each finding was checked against the code it was reported in rather than matched by a scanner.
- Who is allowed to call this endpoint? Point at one screen and ask who can make the request behind it. In those 21 third-party applications audited in June and July 2026, 11 had confirmed unauthenticated endpoints doing privileged work: the screen asks you to log in and the request underneath does not care. The cohort, the method and the rest of the counts are in what the 26 audits found.
- Does the server work out the price and the role again, or does it believe the browser? In the same 21 third-party applications from that June and July 2026 cohort, 10 had a confirmed case of the server accepting whatever the browser asserted. A generated checkout that takes the total from the page it just rendered looks identical to one that recalculates, until somebody changes the number in the page.
- Is there a limit on the call that costs money? Of those 21 third-party applications audited in June and July 2026, 13 had no rate limiting confirmed on their most expensive endpoint. The expensive one is usually the model call, the email send or the signup code, and it is the line item that turns a quiet week into a bill.
- Where does the key live? There is a right answer, on the server, and several wrong ones, and a candidate who can tell you which one your app does within a few minutes of opening it is reading, not guessing. Somebody who needs a week to find out is going to charge you for that week.
- What does the database rule actually permit, as opposed to what the screen suggests? The screen shows one customer their own orders. The rule underneath decides whether a request that skips the screen can ask for everybody’s. A candidate for the second job should be able to say which of those two things your app does, and say it out loud.
Two things about those numbers keep them honest. They are drawn from a distilled ledger of 420 findings that keeps the notable ones rather than all 958, so every count above is a lower bound. And across all 26 applications in that study the scores ran 29 to 81, mean 52.1, median 51, with 22 of the 26 carrying at least one confirmed critical finding.
The volume is not the job either. Those 21 third-party applications produced 958 confirmed findings between them, roughly 46 per application, and 58 of them were critical. Anybody can hand you a list of 46 things. Knowing which of them will cost you a customer this month, and which can sit for a year, is the part you are paying for, and it is the part that does not appear on a skills grid.
Somebody posting in a public builder community, one of the AI-app owners whose words this page draws on, asked for exactly this and had no name for it:
… how well can I secure the vibecoding output(generated code), the process and the tools/plugins to use while I build prod grade apps by vibecoding
What they want is a person who can look at what the builder produced and say whether it holds up. That judgement is the whole of the second job, and the five questions above are how you find out in twenty minutes whether a candidate has it.
What actually comes out of the builder
A candidate who lists Python, TensorFlow and PyTorch has told you about the first job. For the second one, the qualifying knowledge is what the builder actually hands over, and that is written down by the vendors themselves.
Take the most common one. Lovable’s GitHub sync documentation, read on 29 August 2026, sets out four things about the arrangement. Lovable edits and syncs a single branch at a time, and unless you change it that is the repository’s default branch. Importing a GitHub repository that already exists is listed as something the tool will not do, because the sync runs outward from Lovable. Disconnect and reconnect, and the documentation says Lovable creates a new repository with the current version of your project, leaving the original unchanged on GitHub and no longer linked. And the repositories the tool makes for you start private on every plan.
Read those four facts as a job description and the shape of the work changes. The person you hire is working in a repository whose branch model the tool decides, on a project the tool cannot be pointed back at, with a reconnect that leaves the old repository stranded on GitHub. Just clone it and fix it is not what the day looks like. A developer who has never met this arrangement will spend their first paid hours discovering it, and you will pay for those hours.
Every builder has its own version of that. What a Base44 project contains when you take it out is set out in what Base44 exports, and the general shape of what survives a move and what does not is in what does not come across. A candidate for the second job should be able to name your builder’s version of this before you tell them, because they have hit it before.
This is also the honest reply to the word on the adverts. AI-native is an advertising phrase, and it describes an attitude to tools rather than a capability you can check. The checkable version is duller and much shorter: this person can open your repository, name the tool that generated it, and tell you what that tool leaves behind. Telling from the code itself that a tool wrote it is a separate read, and it is the skill this whole hire rests on.
Three questions that tell the two apart
Three questions, all about your code rather than about theirs. Ask them in a call, with your repository shared. A candidate for the first job will answer generally and well. A candidate for the second job will point at a file.
| Ask this | What a real answer names | What a thin answer sounds like |
|---|---|---|
| Show me one place in my code where the server checks who is asking | A specific file or route in your repository, and what happens to a request that skips the check | A description of how authentication works in general, with no file named |
| What does my builder’s export actually contain, and what is left behind in the tool | The builder by name, what comes out, and the parts that stay in the vendor’s account | That it exports to GitHub and should be fine |
| What would you look at first here, and why that | One place, with a reason drawn from what they just saw on screen | Best practices, a security review, or a rewrite |
The third question is the one that separates readers from rebuilders. Somebody who answers with a rewrite has told you they intend to start again, which is sometimes right and is always a different purchase from the one you came for. Somebody who names one place and says why has just proved they can read the thing you own.
If you would rather buy the reading before you buy the hire, that is a purchase of its own: paying somebody to read the code first is how most owners get an opinion without committing to a person, and fixing code a tool wrote covers what usually comes back.
Keep this to three questions on purpose. Proving that a candidate is genuinely the person their profile describes, without being able to read any of the code yourself, is its own sequence and it runs alongside these rather than inside them. The full list of questions to put to somebody before you agree anything is longer than the three here.
Name the job, then hire for it
If your customers will use the AI, hire for the first job and judge the candidate on things they have built and still run. If a tool built the app and you need somebody to take it over, hire for the second job, and judge the candidate on whether they can read your repository while you watch.
Then give whoever you pick a running start by sending the access and the account details up front, which is set out in what to send whoever you hire. Doing that before the first paid hour is the cheapest thing on this page. One question outlives the hire while you are handing access around anyway: what an assistant can reach once it is connected to your accounts is a permissions question of its own, separate from who is holding the keys this month.
Which shape of seller will take on AI-built app work at all, a freelancer found through a marketplace, an offshore shop, or a small team that works on nothing else, is the map this page sits inside. AxonBuild is one point on it: Bilal fixes and extends AI-built apps that already run. The video call is 20 minutes and free, and it is where you show him the app and what you have tried. If you want him to make the change, he checks the app and gives you a fixed quote, and you pay after you see it working. It is not an agency, not an AI development company, not a dedicated team and not a consultancy, and it is the wrong call for a new AI product build, a data-science hire or a machine-learning platform.
The five questions in the reading test cost you nothing to ask and take about twenty minutes. The candidates who can answer them about your repository are a much smaller group than the search results suggest, and they are worth finding before you agree a price rather than after.
Common questions about hiring an AI developer
What does an AI developer actually do?
Two different things, depending on which job you are buying. In the first, they build AI into a product your customers use: a model integration, a chat surface, a scoring or classification pipeline, and the account and cost that come with it. In the second, they take over an application an AI builder generated, read the code it produced, and repair or finish it. The same job title covers both, which is why the search returns pages that do not match what you want.
Is an AI engineer the same thing as an AI developer?
In practice they are used interchangeably by sellers, and both usually mean the first job. On the pages that ranked for this search on 29 August 2026, hire an AI engineer and hire an AI developer led to the same skills grid and the same contact form. Neither title tells you whether the person has ever opened a codebase a tool generated, so the title is not the thing to filter on.
Do I need one if my app was built with Lovable, Replit, Base44 or Bolt?
You need the second job, which is a developer who can read generated code, not somebody who builds AI features. The distinction matters because the marketplace filters and the seller pages are all aimed at the first job. Ask any candidate to name your builder, say what its export contains and say what it leaves behind in the vendor’s account, before you talk about anything else.
Does the AI part make the hire more expensive?
It changes what you are buying more than it changes the arithmetic, and the price still comes from the size of the named result rather than from the letters AI. Every cost question in this family, including hourly rates and what different routes actually charge, is answered on what this kind of work costs. No rate is printed on this page, in either direction, because the pages that publish rates are selling the first job.
Can I hire someone who only knows the builder, not the code underneath?
For small changes inside the tool, yes, and it is often the cheapest way to get them. For anything that touches who is allowed to do what, what the database permits or where a key lives, no: those live below the level the builder shows you, and somebody who only knows the prompt surface cannot see them. The test is question one of the reading test above.
What should I ask a candidate to show me?
One place in your own repository where the server checks who is asking. A portfolio, a certificate and a marketplace percentile all tell you less than that one file does. A candidate who can open your code and point at that in a few minutes is doing live what the job is; a candidate who answers with general principles has told you they have not read it yet.
Can I hire an OpenAI or ChatGPT developer specifically?
You can, and people search for it in those words, but the vendor name describes the model account rather than the skill. Someone who has shipped against one model vendor’s API can usually ship against another, and the model vendors’ own documentation is the primary source for what any of them currently supports. If your app was generated by a builder and simply calls a model, the harder half of the job is still the generated code around the call.
Why does my search keep returning app builders instead of people?
Because the bare phrase ai app developer means something else to Google. Of the eight organic results at depth 10 for it on 29 August 2026, seven were AI app builders or listicles of app builders, and the eighth was a forum thread asking which AI tool can create an app from scratch. Not one of them offered a person for hire. The two searches with a hiring word in them, hire an AI developer and hire an AI app developer, returned sellers and staffing shops on the same day.
Built it with AI. Can’t get the last part right?
That’s the normal state of an AI-built app, and it’s fixable. I trace what the app actually does, explain what needs changing, and build it if you want me to.
Talk about your app → 
Free 20-minute video call with Bilal.