Should you run AI coding tools in-house?

Why companies are asking

Building with AI tools has gone mainstream. McKinsey’s 2026 research found around a third of organisations had declined to buy software because they could build it themselves with agentic tools. So the obvious question: why not hire one person, give them the tools and keep everything in-house?

When it works

• The person is properly senior: they’d be good without the tools.
• The work is mostly internal tools with low risk if something breaks.
• Someone else can review what they ship.
• You’re prepared to recruit for months to find them.

Where it falls short

• Judgment. Tools say yes to every idea. Someone has to say “not that, and not yet”.
• Experience design. A tool builds the screen you described. It doesn’t know what confuses real users.
• Production. Security, monitoring, recovery and edge cases need someone who’s been woken at 2am before.
• Review. One person reviewing their own AI-assisted code is no review at all.
• Dependency. You’ve rebuilt the one-developer problem, with faster output.

The cost comparison

A senior hire outside London costs about £88,000 in the first year once NI, pension and fees are added, more in London (the breakdown). A subscription using the same tools, with senior review and planning time, costs £48,000 a year on Embedded Plus at current pricing.

If you do go in-house

• Hire for judgment first and tool fluency second.
• Give them a reviewer, inside or outside the business.
• Run the production checklist before anything faces customers.
• Keep accounts, repositories and keys in the company’s name, not the person’s.

Questions