We build apps with AI now. Here's what that changes for your budget and timeline

A few years ago, building an app meant a team typing every line by hand. Today a lot of first-draft code is written by AI coding agents, and the job of an experienced engineer has shifted towards deciding what to build, checking what the AI produced, and owning the hard parts. We build this way every day. Here is what it means for you as the person paying for the software.

What genuinely got faster

• Screens and forms. Standard UI — lists, detail pages, settings, admin tables — now takes hours rather than days.
• Boilerplate. Project setup, API endpoints for standard operations, data models and migrations.
• Tests. AI is good at writing the first version of tests, which means more of the code is tested than before.
• Prototypes. A clickable prototype you can put in front of users in days, so decisions are made on something real instead of a document.
• Documentation and handover. Setup guides and API docs are no longer the thing that gets skipped.

In practice this means a well-run team can deliver the same scope noticeably faster than before — and you should expect that to show up in the timeline and the price.

What still needs experienced engineers

AI makes the easy 70% faster. The other 30% is where projects succeed or fail, and it hasn't got easier:

1. Understanding the business. Which workflow matters, who approves what, what happens when a payment fails halfway. AI can't sit in your operations meeting.
2. Architecture and data design. Get the data model wrong and every later feature is harder. This is a judgement call, not a typing task.
3. Security. Authentication, permissions, protecting personal data, handling secrets. AI-written code regularly contains plausible-looking security mistakes.
4. Money and edge cases. Rounding, refunds, partial failures, retries, reconciliation. The cases that happen once in a thousand transactions are the ones that cost you.
5. Integrations. Real-world payment gateways, SMS providers, WhatsApp, government portals and hardware behave differently from their documentation.
6. Shipping. App Store and Play Store review, performance on cheap Android phones, monitoring, backups.

The risk of "vibe-coded" apps

It is now possible for someone with no engineering background to produce an app that looks finished. It might even work in the demo. The problems show up later: no real access control, secrets inside the app, no tests, no backups, and code nobody can maintain. If you are buying software, these are the questions that separate a real product from a demo:

• Who reviews the AI-generated code, and how?
• How are passwords, OTPs and API keys stored?
• What is tested automatically before each release?
• Where does the data live, who owns the cloud account, and how are backups restored?
• If you change vendors, do you get the full source code and can someone else run it?

A good vendor will answer all five without hesitation.

Adding AI features to your own product

The other half of "AI" is putting it inside your product: reading documents, answering customer questions, grading submissions, summarising conversations. A few practical points:

• Start with a narrow, checkable task. "Extract these 12 fields from a shipping invoice" beats "an AI assistant for logistics".
• Measure accuracy on your own data before promising it to customers. Build a test set of real examples on day one.
• Keep a human in the loop wherever a mistake costs money or trust, at least until the numbers prove otherwise.
• Budget for usage. AI model APIs are billed per use. Costs are usually small per request but grow with volume, so they belong in the running-cost estimate alongside servers, SMS and email.

How this changes the way we work

Every project still starts with a written scope and a fixed price. AI changes what happens after that: working software in staging every week, more of the code covered by tests, and engineers spending their time on architecture, security, integrations and the edge cases that matter. The result is the same product, sooner.

If you are weighing whether an idea is worth building, now is a good time to ask again. Things that were too expensive two years ago often aren't any more.