SLAPUP

For web developers

AI code is almost right, and that's the expensive part

Broken code tells you it's broken. Almost-right code looks finished, passes a quick glance and fails later.

By SlapUp · · 4 min read
A code editor flags a line that looks right but has no error check, next to the Kodo mascot.

AI assistants are now a normal part of web development, but trust hasn't kept up. In Stack Overflow's 2025 Developer Survey, with more than 49,000 respondents, 80% of developers said they use AI tools, while only 29% said they trust their accuracy.

The top frustration, named by 45%, was AI answers that are almost right but not quite. And 66% said they spend more time fixing almost-right code. That matches what many of us see day to day: writing got faster, but checking became the real job.

Use AI tools
80%
Spend more time fixing almost-right code
66%
Trust AI accuracy
29%
Stack Overflow 2025 Developer Survey, 49,000+ respondents.

Where almost-right code hides

A simple way to work with it

Ask for small changes

A 30-line change is easy to review. A 900-line change usually gets skimmed. Break the work down, and commit each step.

Describe the result before you generate it

Write what should happen, including the failure cases. Better still, write the tests first. It gives you something objective to check against.

Run it, then read it

Run the code, the tests and the type checker. Then read every line as if a new teammate wrote it. If you can't explain a line, don't ship it.

Check what the browser actually does

Open the page on a phone, use it with only a keyboard and run a performance check. Plenty of problems only appear in the real page.

Keep humans on the risky parts

Be especially careful with login, payments, personal data and anything hard to undo. Use AI to help, but review these yourself, line by line.

The short version

AI can speed up the typing. You're still responsible for the result, and that judgment is what clients are paying for.

Need a fast, well-built website?

Talk to Kodo