Does AI Make Software Development Cheaper?

In April 2026, Google said that around 75% of its new code is now generated by AI. In late 2024 the figure was about 25%. The detail most headlines skip: every line is still reviewed and approved by Google's engineers before it ships.

Founders read those numbers and ask us the obvious question. If AI writes most of the code, shouldn't my app cost a lot less?

It costs less, but not as much less as the headline suggests. Our team works with Claude and Cursor every day, and we get through roughly 30–40% more work in the same time than before. Below is where that time comes from, why your quote doesn't shrink by the same amount, and what we do so that AI-written code doesn't turn into a codebase nobody can maintain.

What AI actually speeds up

AI is fastest at work where the answer is already known and the job is typing it out correctly:

• Boilerplate. Screens, forms, API endpoints and data models that follow a pattern the project already has.
• Tests. Unit tests for logic that's already written, and the edge cases a tired developer skips.
• Migrations. Moving code to a new library version or a new API, file by file.
• Documentation. Comments, READMEs and the notes the next developer needs.

On a typical web or mobile project this is a large share of the hours, which is why the overall gain is real.

Why your quote doesn't drop by 30–40%

A good part of any project isn't writing code, and AI speeds those parts up much less:

• Deciding what to build. Discovery, scope and the trade-offs behind them are conversations with you, not code.
• Architecture. How the data is structured, where the logic lives, what talks to what. A developer designs this before AI writes a line.
• Integrations. Payment providers, health records, a client's legacy system. AI can write the call; it can't tell you how the other system behaves when it fails.
• Hardware. For our Misty the Cloud app, every Bluetooth feature had to be tested against the real night light on iOS and on Android. No model does that for you.
• Review and release. Code review, device testing and App Store and Google Play review take as long as they did.

So the build costs less and arrives sooner, but by less than the productivity gain. Treat any agency that promises to halve the price "because of AI" with caution: either the estimate was padded before, or something is being left out now.

How the saving reaches your invoice

This depends on how the agency bills.

We fix the price after discovery, then bill every two weeks for the hours the team actually worked, and the total never goes over the fixed price. When AI shortens a task, it shows up in the hours, so you pay for the faster work, not for the old estimate.

With a classic fixed price, the agency keeps the time it saves. With a pure hourly contract, you get the saving only if the agency is honest about how it works. Either way, ask how the use of AI is reflected in the price.

Without architecture, AI multiplies the mess

AI writes code that looks right. Without a clear structure to follow, it also writes the same thing three different ways in three files, adds a library for something the project already does, and fixes one bug by moving it somewhere else. Each change looks fine on its own. After a few months, nobody can change anything without breaking something.

We see the result when founders bring us apps built quickly with AI tools such as Lovable or Cursor. The common problems:

• data that anyone with the public key can read, because database access rules were never set
• secret keys sitting in the frontend code
• no tests at all
• sign-in edge cases that were never handled: password reset, expired sessions, deleted accounts

None of this means the prototype was a mistake. It proved the idea. It just needs a review before real users and real data go into it.

How we use AI without losing control of the code

Here's how it works on our projects:

1. A developer sets the architecture first. Structure, data model, conventions and boundaries are decided by a person before AI writes code. AI then works inside that frame, so the project stays consistent as it grows.
2. A developer reviews every change AI produces. The developer who asked for the code is responsible for it, as if they had typed it.
3. Tests check it automatically. Code that doesn't pass doesn't get merged.
4. Your code stays in your repositories, from the first commit, whoever or whatever wrote it.

Google works the same way at its scale: AI writes, engineers approve. The share of code AI generates is going up, and the review stays.

Questions to ask an agency about AI

• Which AI tools does your team use, and for which tasks?
• Who reviews code that AI generates before it's merged?
• Who designs the architecture, and when?
• What tests run before a change goes in?
• How does your use of AI show up in the price?

Clear, specific answers mean the team has a process. Vague answers mean AI is either not used or not controlled.

Frequently asked questions