Your team can build apps with AI. Here’s what it really costs.

Tools like ChatGPT, Lovable, Replit and Cursor let anyone describe an app and watch it appear. Someone in operations builds a job tracker over a weekend, a sales manager builds a quoting tool, and on Monday it looks almost done. That part is real, and it is useful. What usually follows is less visible: weeks of evenings, a second draft, a third, and an app that still is not quite ready for the rest of the team.

The first 70 percent is fast

Engineer and author Addy Osmani calls this the 70% problem: people without engineering experience reach about 70 percent of a working app surprisingly quickly, and then the last 30 percent becomes an exercise in diminishing returns. The last stretch is the hard part of any app: edge cases, real data, permissions, error handling and everything that happens when twenty people use it at once.

Even experienced builders plan for a long tail. SaaStr founder Jason Lemkin spent almost 100 hours building a commercial app with AI tools, and now advises budgeting 150 hours across a full month to reach commercial quality, with more than half of that going to testing, debugging and refinement. For someone fitting the work around a day job, 150 hours is not a month. It is a season.

Then the drafts start

AI tools are very good at answers that are almost right. In the 2025 Stack Overflow Developer Survey, the top frustration with AI tools, named by 66 percent of developers, was "AI solutions that are almost right, but not quite", and 45 percent said debugging AI-generated code takes more time. Those are professional developers who can read the code.

For someone who cannot read the code, "almost right" turns into a loop: ask for a fix, the fix breaks something else, ask again. After enough rounds the app is hard to reason about, so they start over with a better prompt. That is draft two, and it is rarely the last.

Feeling fast is not the same as being fast

In an early-2025 randomized study by METR, experienced developers expected AI tools to make them 24 percent faster. Measured, they were 19 percent slower, and afterwards they still believed they had been about 20 percent faster. The tools have improved since, and METR’s later work suggests they now help experienced developers. But the gap between how fast it felt and how fast it was is the lesson, and it is widest for someone who cannot judge the code: an app that changes on screen every few minutes feels like progress even when it is going in circles.

The hours come from people with other jobs

DIY app projects rarely show up in a budget. They come out of evenings and the gaps between someone’s real responsibilities, often someone senior who knows the process best. Five hours a week for four months is about 85 hours of a key person’s time, before anyone else has used the app. And because the hours are spread out, nobody adds them up until the project stalls.

What the demo does not show

• Security. In Veracode’s 2025 tests, AI models introduced a known class of security flaw in 45 percent of coding tasks. A non-developer has no way to spot these.
• Review. In a 20i survey of 600 people using AI to write code at work, half had little or no coding knowledge, and fewer than a third had their AI-written code checked by a professional developer.
• Real data and backups. In a widely reported July 2025 incident, an AI coding agent deleted a live database during a code freeze, against explicit instructions. The data was recovered, but only because a backup existed.
• Permissions. A demo has one user who can see everything. A real app needs managers, staff and customers who each see only their part.
• Integrations. Writing into your accounting system or ERP without creating duplicates is where many apps get hard.
• Who maintains it. When the person who built it changes roles or leaves, the app usually stops changing, and then stops working.

Where DIY AI apps are the right call

None of this means your team should stop. AI app builders are excellent for:

• A personal tool that only the builder uses.
• One-off reports, calculators and checklists.
• Prototypes. A clickable version of what you want is the best project brief we can receive. It shows the screens, the steps and the words your team uses, and settles questions that would otherwise take meetings.

The trouble starts when a prototype becomes the production app by default, with customer data, payments or the whole team depending on it.

A better split: your team prototypes, a professional team builds

The fastest path we see is to let the people who know the work prototype it, then hand it to a team that builds production software. We use AI coding tools every day as well; the difference is the experience to know when the code is almost right, and the habits around it: a proper data model, permissions, security review, tests, backups, hosting, and someone accountable after launch. Your prototype becomes the spec, nothing your team learned is wasted, and the drafts stop.

Signs a DIY project needs help

• It has been "almost done" for more than a month.
• Each fix seems to break something that used to work.
• It is about to hold customer data, payments or anything you would not want public.
• More than a handful of people will depend on it.
• Only one person understands how it works.

If two or more of these sound familiar, send us what you have, prototype included. We will tell you honestly whether it needs a rebuild, a cleanup or nothing at all. Our guide on what to prepare before you hire an app developer covers the rest of the brief.

More guides: How much does a custom app cost? · How much does a B2B ordering portal cost? · NetSuite SuiteCommerce or a custom B2B portal? · ServiceTitan or custom dispatch software? · How much does an AI customer support assistant cost? · Custom software or off-the-shelf? How to decide · What to prepare before you hire an app developer · Adding AI to business software, safely · How to stop taking orders by email · A customer portal on top of NetSuite or QuickBooks · Shopify app or custom build? How to tell · How to forecast inventory for retail stores · Why business apps get bloated, and how to keep yours lean