5 October 2026

Do You Still Need a Developer Now That AI Writes Code?

AI writes code in seconds. So do you still need a developer? Yes: for the right calls, and for code review so your webshop doesn't crash.

TL;DR

AI tools write code in seconds. But typing code has never been the hard part of a developer's job, and AI doesn't change that.
The real job takes experience: an informed opinion on what to build, what to leave out and what fits your business. You build that opinion by seeing what works in practice, and what breaks.
The other half is review: using that same experience to check every change before it goes live, so your webshop doesn't break on your busiest day.
The research agrees. AI makes teams ship faster, but releases also break more often. Speed without review just moves the problem to your customers.

Anyone can now ask an AI tool for working code and get it in seconds. So do you still need a developer now that AI writes code? My answer: the value was never in typing fast. It is in experience. Experience gives you an informed opinion about what to build, and the eye to spot in review what is going to crash. If you run a webshop or a brand website, you feel that difference in your revenue, and in how often something needs fixing.

Typing code has never been the hard part

Say you want a new feature in your Shopify store. A product bundle, a size filter, a banner that changes per country.

Writing the code is only a small part of that work. The rest is questions. What do you actually need? How does it fit with the apps and theme you already have? What happens when one of those apps updates? Can your team change it later without calling a developer?

An AI tool can write a size filter in a minute. It can't tell you whether you should build one at all. You only know that once you've seen how choices like this play out.

What the research says

Three sources are worth knowing. They don't all point the same way, and that is the point.

DORA (Google Cloud), 2024 and 2025. DORA surveys thousands of software teams every year. In 2024 it found that more AI use went with slightly slower delivery and noticeably less stable delivery: for every 25% increase in AI adoption, an estimated 1.5% drop in throughput and 7.2% drop in stability. In the 2025 report, the speed effect flipped: teams that use more AI now ship faster. But stability did not recover. More AI still goes with more releases that break something and need a fix. DORA's summary of 2025 is that AI is an amplifier: it makes strong teams stronger and makes existing weaknesses bigger.

Stack Overflow Developer Survey, 2025. 84% of developers use or plan to use AI tools. But more of them distrust the accuracy of the output (46%) than trust it (33%). Only 3% trust it highly. The most cited frustration, named by 66%: AI answers that are "almost right, but not quite". 45% say debugging AI-generated code takes more time. And the two tasks developers least want to hand to AI? Deploying and monitoring (76% don't plan to), and committing and reviewing code (59% don't plan to).

METR, 2025 and 2026. In an early-2025 study, 16 experienced open-source developers took 19% longer on real tasks when they were allowed to use AI. The striking part: they expected AI to make them 24% faster, and afterwards still believed it had made them 20% faster. METR now calls those results out of date. Its February 2026 follow-up suggests newer tools probably do speed developers up, but METR says its own new data is an unreliable signal. What still holds: how fast AI feels tells you very little about how fast it is.

Put together: the code comes faster now. Someone with experience still has to decide what to build and check that it works.

What having an opinion means in practice

An AI tool answers the question you ask. An experienced developer asks whether it's the right question. That opinion doesn't come from a prompt. It comes from earlier projects: what worked, what turned out unmaintainable a year later, what fell over on the busiest day.

Some examples from webshop and website work:

Saying no. An extra Shopify app for one small feature means more code on every page and another monthly bill. Sometimes a few lines in the theme do the job. Sometimes the app is the better choice. Someone who has seen both routes play out has to make that call.
Choosing the setup. A native Shopify theme, or a headless build where the store your customers see is a separate Next.js site connected to Shopify. Both are good, for different brands. I wrote about how to choose between headless and native Shopify.
Building for your team. The clever solution is not always the one your marketing team can edit in the CMS. I've seen often enough how quickly a site drifts when the team can't work with it themselves. That's why I'd rather build the version they can run without me.
Timing. Knowing what not to touch two weeks before Black Friday. You don't learn that from documentation.

None of these choices live in the code. They come before it, and they come from experience.

What review means in practice

Review means reading every change before it goes live and asking a few plain questions:

Does it do what we agreed?
What happens with unusual input? An empty cart, a sold-out variant, a discount code combined with a gift card.
Does it make the page slower?
Is it safe? No exposed API keys, no customer data where it doesn't belong.
Will the next developer understand it?

"Almost right" code is the risky kind. It works in the demo and fails on the edge case. In a webshop, that edge case is usually the checkout, a discount or a payment: exactly the places where a bug costs you money straight away. An experienced reviewer knows where those edge cases are, because they've seen them go wrong before.

What AI does well, and what still needs a person

Task	AI tool	Experienced developer
Write a first version of a component or script	Fast, often fine	Checks it and adjusts it
Decide whether a feature is worth building	Doesn't know your business	Weighs cost, upkeep and who will use it
Catch edge cases (sold-out variants, stacked discounts)	Often misses them unless asked	Tests them on purpose, knowing where things broke before
Keep the site fast	Adds code easily	Removes what isn't needed
Take responsibility when something breaks	No	Yes

Questions to ask your developer or agency

Use this checklist when you hire someone, or when you talk to the team you already have:

Who reviews code before it goes live, and how?
Do you use AI tools? What do you check by hand?
How do you test checkout, discounts and payments after a change?
What happens if something breaks after launch?
Can my team edit content without touching code?

If the answers are vague, that tells you something.

What this means for your brand

Fast code is now cheap and everywhere. What you pay a developer for is experience, judgement and responsibility: choosing the right thing to build, and making sure it works when your customers use it.

Want a second opinion on your Shopify store or website? Get in touch, see how I build Shopify stores or have a look at my work.

FAQ

Will AI replace software engineers? AI is changing how code gets written. It doesn't replace the parts clients care about most: the experience to decide what to build, and taking responsibility when it runs in production.

Is AI-generated code bad? No, it's often fine. But it needs the same review as any other code. In the 2025 Stack Overflow survey, AI output that is almost right, but not quite, was the most cited frustration among developers (66%).

Should my developer use AI tools? That's fine, as long as they review what goes live. Ask how they check AI-written code, especially around checkout and payments.

Does AI make development cheaper? Sometimes, for the writing part. The research is mixed: DORA's 2025 report links more AI use to faster delivery but also to less stable releases. Don't assume a lower price without asking where the time and the testing go.

Have a project in mind, or want someone to look at what you have now? Contact me.

Sources

← All insights