Who Builds an AI Product Your Team Can Maintain

A media or tech team can maintain an AI product when the repo is theirs, one person can change a safe feature, and auth and payments stay with an engineer.

By Muhammad Saif and Aashan Khan · Techtribe co-founders

The question from a media or tech team is rarely "can you add a model." It is "can we still run this after you leave." Editors change headlines. Producers swap a show. A product manager wants a field renamed before a launch. A staff engineer has to answer for the repository in a review they did not attend. If the only person who can make those changes is the agency that prompted the app into existence, the team does not have a product. It has a dependency.

This note is the handoff question. What it means to export a Lovable project so you are not stuck on the builder is who owns your Lovable app code. Read that when the question is portability. Stay here when the question is whether your people can change the product without starting a ticket for every sentence. What the work costs is how much an AI MVP costs.

What "maintain it ourselves" has to mean

It means a named person on your side can do four things without us.

They can run the app the way production runs it, from the repository, with secrets that your company controls. They can change copy, a prompt, or a content rule, and roll that change back. They can tell a safe change from a dangerous one. They can point at the model job in one sentence and say what a wrong answer looks like.

If that person does not exist, no repository layout will save the handoff. The first hire is not a second agency. The first hire is one owner, even if that owner is a producer who can only change the safe list, plus an engineer you call when the change is not safe. A media team that expects every editor to deploy is designing an incident. A tech team that expects every contractor to hold production credentials is designing a leak.

"Maintain it ourselves" does not mean you will never call anyone. It means the default change, the one you make every week, does not require a meeting.

What your team should be able to change

Copy and labels. The prompt text, when you also have a handful of examples you can recheck. Which document or episode a feature points at, when that choice is data and not a code change. A feature flag that turns the model off. The layout of a page that does not collect a secret.

A newsroom can own the examples. The questions readers actually ask, the answers an editor would accept, the answers an editor would refuse. Those examples are the asset. A model without them is a tone. A tech company's product team can own the same list for support replies or for a changelog summary. The builder should leave the list in the repo, next to the prompt, not in a chat transcript.

What your team should not change alone is smaller and more important. Sign-in. Who can read which row. Anything that moves money. A migration. A new dependency. A production key. Those are how a "small copy change" becomes a weekend. The product is maintainable when those paths are obvious, so a non-engineer does not wander into them, and an engineer can review them in one sitting. Technical debt that sticks in generated code is what it looks like when the paths are scattered through files nobody named.

What the builder has to leave behind

The repository in an account your company owns, from the first commit, not as a zip file at the end. A README that says how to run it, how to deploy it, and which environment variables exist, by name, without the values. The model job in one paragraph. The examples. The place the prompt lives. The feature flag. A note of what the thirty days of support covers, and what it does not.

Design files or a short description of the design system, if the package included design. If it did not, the interface is the design system the package named, and pretending otherwise creates a handoff argument. The Basic MVP uses our design system. The Production MVP designs in your brand. Either can be maintained. A mix that was never written down cannot.

Access to the cloud project in your name. The model provider key in your name, with a spend alert your owner can see. Billing emails that do not go only to a founder who is about to go on leave. A media company that puts the key in a producer's personal account will lose the key when the producer leaves. That is an ownership bug, and it is visible on day one.

None of this is a 40-page architecture binder. A binder nobody reads is not a handoff. The test is whether your owner can make a safe change in a week when we are not in the room. If the engagement cannot survive that test, it is not finished.

Why media teams and tech teams feel this differently

A media team often has the editorial standard and not the deploy pipeline. The failure mode is an editor pasting a prompt into a builder that also holds the production database, because that was faster than filing a ticket. The product has to make the safe change easier than the unsafe one. A prompt file in the repo, reviewed like copy, beats a builder session that can also drop a table.

A tech team often has engineers and a habit of moving fast. The failure mode is those engineers inheriting a generated codebase they will not touch, so every model change still goes back to the agency, or worse, each engineer prompts a different fix onto main. The product has to be boring enough that your engineers will accept it. Typed boundaries around the model. One way to call it. Tests on the path that charges or publishes. If your senior engineer looks at the repo and refuses to be on call for it, believe them. That reaction is the review.

Both teams share the original-prompter problem. The person who built the first version in a weekend understands the chat, not the code. Six months later they are on another project. The team that has to maintain the product was never in the chat. Buy the handoff as if that person has already left, because they will.

What to ask before you hire

Ask whose GitHub or GitLab organization receives the commits. If the answer is theirs, you are not maintaining the product.

Ask what a producer or a PM is allowed to change without an engineer. If the answer is "nothing" or "everything," the split has not been designed.

Ask to see a prompt change go out and come back. Not a slide. A commit, a check against two examples, a revert. An agency that cannot show that loop has not built a maintainable feature. They have built a demo.

Ask who gets paged when the model bill spikes or the deploy fails, during the engagement and after it. "We monitor it" is not a name. Your owner needs to know the name, even if the name is ours for the months you keep us.

Ask what happens to the repo if you stop paying. The answer should be that nothing happens to the repo, because it was always yours. Support stops. The code does not lock. If stopping payment locks the product, you are renting, and this article is about a different purchase.

Fixed scope, then your team, or a team that stays

The Basic MVP and the Production MVP end. Thirty days of support, then your people own the next change, and new features are a new quote. That is the right shape when you want the handoff. It is the wrong shape when you already know the newsroom or the product squad will need engineering every week. Pretending a fixed scope includes that week is how the relationship sours in month two.

The Dedicated Team is the stay: two engineers, a project manager, founder oversight, your repo, your cloud, and 30 days' notice. What that engagement costs is in how much an AI MVP costs. Your editors and PMs still need the safe list. A dedicated team that becomes the only path for a headline change has not built a maintainable product either. They have become the bottleneck you hired to avoid. The split remains. They take auth, money, and migrations. Your side takes the changes you listed.

Some buyers want the fixed build first and the dedicated team only if the pilot is real. That sequence works. Say it at the start so the handoff notes are part of the fixed scope and not a favor at the end.

A handoff you can rehearse

In the last week of a fixed engagement, your owner makes one change we did not make for them. A prompt sentence. A label. A document added to the set the model can see. They deploy it, or they watch the deploy and can repeat it. They revert it. If they cannot, the engagement is not done, and the calendar should say so while we are still on it.

Write the dangerous list on a page in the repo. Auth. Permissions. Payments. Migrations. Keys. Your owner does not need to be afraid of the product. They need to be unwilling to improvise on that list. The page is what you show a new producer on their first day, so the rule does not live in someone who might leave.

Keep the examples next to the prompt. When an editor says the model sounded wrong, add the example and the acceptable answer. That is maintenance. Rebuilding the feature from a new chat is not.

If the change they want is on the dangerous list, the path is a review by an engineer, yours or ours, not a prompt from the editor's laptop. The AI MVP development service is how we take that kind of change when you want it scoped. Use the path that matches the frequency you actually have, not the frequency you hope for. Weekly engineering is a team that stays. A rare change is a new scope. Neither decision is a number on this page.

What a good first month looks like

Week one after handoff, your owner ships a copy or prompt change and reverts a bad one. The model bill is visible to them. Nobody pastes a production key into a side tool.

Week two, an editor or PM files a change that is on the safe list and sees it go out without a project being opened. A change that is on the dangerous list waits for a review. The wait is a success. It means the split is real.

Week three, you look at one wrong answer and add it to the examples, or you decide the job is wrong and you stop. Stopping is allowed. A maintainable product is also a product you can turn off. The flag is part of the handoff.

If none of that happened because everyone is afraid to touch the repo, call. The handoff failed, and the fix is a short engagement to make the safe path real, not another year of prompting in the original chat.

The builders who fit this question are the ones who will pass that month. Ask them to describe it back to you before you sign. If they describe a retainer with no split, they are selling presence. Presence is sometimes what you want. It is not the same sentence as "our team can maintain it."

A note for the buyer who is also the only engineer

You can be the owner and the reviewer. The handoff still has to be written down, because the version of you that built the chat will not be the version of you that gets paged in six months. The page in the repo is how future-you, or the first hire, starts from the code instead of from a memory of the prompt.

Do not skip the rehearsal because you trust yourself. Make the safe change on a branch, deploy it, revert it, on a Tuesday when nothing is on fire. The rehearsal is fifteen minutes. The alternative is discovering during an outage that the deploy lives in a builder you no longer have a login for.

If you already know you will not be the one who maintains it, put the actual owner on the scope call. A call with a founder who will not touch the product, and no one else, produces a beautiful demo and a stranded repo. We would rather meet the producer or the staff engineer now, while the safe list can still be designed, than meet them later when they are angry at a codebase they did not ask for.

The same rule applies if the owner is a vendor of yours, a freelance editor, or a sister team in another city. Name them. Give them the account. A handoff to "the team" is a handoff to nobody.

What we will not pretend

We will not pretend a weekend prototype is maintained because the file is on your laptop. We will not pretend your editors are engineers. We will not hide the dangerous list in a polite phrase like "ongoing optimization." Optimization that touches payments is engineering. Say the word.

If the product you have in mind cannot be split into a safe list and a dangerous list, it is not ready for this kind of handoff. Write the two lists anyway, even if the safe one is short. A short safe list is a real product. A product with no lists is a chat log. It may still be ready for a dedicated team that stays. Those are different yeses. The call is where we pick one, or tell you the product needs a simpler feature before either yes is honest.

If this is your app, send the repo.

A 30-minute scope call. We read what is live, name what is dangerous, and quote a fixed scope.

How Techtribe hands a product to your team

A fixed Basic MVP or Production MVP ends with the repo in your account, the model job written down, and thirty days of support on what shipped. A Dedicated Team stays when the roadmap does not end. We will say which one matches the way your editors or engineers actually work. We will not call a locked-in builder a handoff.

The scope call is 30 minutes with a founder. Bring the person who would own the product after we leave, and one change they must be able to make without us. If the app already lives in a builder, Lovable, Bolt, v0, and Replit describe what you can and cannot take with you. The service page is AI MVP development. This note does not answer what it costs. That is how much an AI MVP costs.

Questions

Who should build an AI product a media or tech team can maintain?

A team that puts the code in your repository, documents the one model job, and leaves a person on your side able to change a safe part of the product. If the only people who can ship a copy change are the original builders, you do not maintain it. You rent it.

What can our own team safely change?

Copy, prompts that you can test against known examples, content, and layout that does not touch sign-in, permissions, payments, or migrations. Those four stay with someone who can explain the diff. A product that mixes them into the same file fights that split.

Is this the same as the cost of maintaining an AI app?

No. This article is the handoff: whether your editors, PMs, and engineers can live with the product after we leave. What the build costs is a different article. What it costs to keep running is another. Neither number is decided here.

What if we want you to stay?

That is the Dedicated Team: two engineers and a project manager on your repo, monthly, with 30 days' notice. It is a different purchase from a fixed MVP that ends in a handoff. Say which one you want on the scope call.

Book a free call

Have a vibe-coded app in production?

30-minute free call. We'll read what you have, name what is dangerous, and quote a fixed scope.