Short answer
AI coding tools write code quickly, and we use them every day. They don’t decide what’s worth building, design an experience real users understand, or take responsibility when production breaks. A product team using the same tools supplies the judgment, UX, production quality and accountability the tools can’t.
Why this question comes up now
Building with AI has gone mainstream. McKinsey’s 2026 State of AI research found that a third of organisations had declined to buy software because they could build it themselves with agentic tools. So it’s a fair question: why not hand your team Cursor, Claude Code or Lovable and skip the agency?
What the tools are brilliant at
Writing code from a clear description. Prototyping an idea in an afternoon. Explaining unfamiliar code. Taking the drudgery out of tests and boilerplate. We wouldn’t build without them now.
What they don’t do
| An AI tool on its own | Fluff, using the same tools | |
|---|---|---|
| UX experience | Builds the screen you described | Designs the experience your users need, from years of watching real people use real software |
| Strategy | Says yes to every idea | Tells you what’s worth building, what isn’t, and in what order |
| Product knowledge | Starts from scratch every conversation | Learns your product, users and business, and gets sharper every month |
| Iteration | Quick to write, slow to get right | Ship it, see how people use it, adjust. In days, not quarters |
| Production | Looks finished in a demo | Tested, secure and ready for real users on a busy Monday morning |
| Accountability | Someone on your team still has to own it | A senior team that answers for every release |
The production gap is real
The clearest evidence is security. In May 2026, researchers at RedAccess reported around 5,000 apps built on platforms including Lovable, Replit, Base44 and Netlify exposing sensitive data with little or no authentication (Axios). Platforms responded that configuration is the builder’s responsibility, so someone in your business still has to own it.
The honest middle ground
If you have a strong senior developer, AI tools will make them faster, and that might be all you need. If you’re hoping a tool replaces the senior developer, you’ll get software that looks finished in a demo. Our guide to running AI coding tools in-house covers what to put around them if you go that way.