AI coding tools for non-technical staff: what the research says, and how to roll them out safely

We have rolled out Claude Code, an AI coding assistant, to non-technical staff at a UK site of one of the world's largest manufacturers: people in operations, quality and finance rather than software developers. It was in-house work before Woerka, so we don't name the business or share its data. This article sets out why we think it's worth doing, what the research does and doesn't show, and what keeps it safe.

The argument: AI raises the floor

The usual pitch for AI tools is that they make your best people faster. The stronger case, and the one the research supports, is that they bring everyone else closer to your best people.

Most operational teams have someone who knows the process inside out but can't write the script, query or small tool that would save them hours each week. An AI coding assistant closes much of that gap. The person who understands the work can describe what they need in plain English and get working software, with a technical reviewer checking it before anyone relies on it.

What the studies show

Several well-designed studies point the same way: the biggest gains go to the least experienced.

• Customer support. In a study of 5,179 support agents, an AI assistant raised issues resolved per hour by 14% on average, and by 34% for novice and lower-skilled agents, with little effect on the most experienced. (Brynjolfsson, Li and Raymond, Generative AI at Work, Quarterly Journal of Economics, 2025)
• Consulting. In a field experiment with 758 Boston Consulting Group consultants, those using AI completed 12.2% more tasks, 25.1% faster and at more than 40% higher quality. Consultants below the average performance threshold improved by 43%, against 17% for those above it. (Dell'Acqua and others, Navigating the Jagged Technological Frontier, Harvard Business School, 2023)
• Professional writing. Among 453 college-educated professionals, ChatGPT cut the time taken on writing tasks by 40% and raised quality by 18%. Weaker writers gained most, so the gap between workers narrowed. (Noy and Zhang, Science, 2023)
• Programming. Developers given GitHub Copilot finished a set programming task 55.8% faster than those without it, and those with less programming experience benefited most. (Peng and others, 2023)

What the studies don't show

You will see claims that AI makes people ten times more productive. We haven't found a rigorous study that supports that, so we don't make the claim. The measured gains are large but not magic, and they depend on the task.

Two findings matter for anyone planning a rollout:

• AI can make people worse at some tasks. In the same consulting study, on a task chosen to sit outside what the AI could do well, consultants using it were 19 percentage points less likely to reach the correct answer. People need to know where the tool is weak.
• Experts don't always gain. A 2025 trial with 16 experienced open-source developers, working on large projects they knew well, found they took 19% longer with AI tools, even though they believed they were faster. (METR, 2025) The payoff is greatest where people lack the specialist skill, not where they already have it.

What a safe rollout involves

Giving non-technical staff a tool that writes and runs code needs guardrails. This is how we approach it:

• Company accounts, not personal ones. Use a business plan with settings IT controls, so company information never passes through personal accounts.
• Clear rules on data. Written guidance on what can and can't be given to the tool, in plain language, with examples from people's own work.
• Start small. A pilot group with real problems to solve, trained properly, before anyone else gets access.
• Limit what the tool can touch. Restrict permissions so it works on copies and approved folders, not live systems or production databases.
• Review before reliance. A technical person checks anything that will be used for real decisions or shared with others.
• Teach the weak spots. Show people where AI is unreliable, such as figures they can't check, and make checking a habit.

Where to start

Pick one team with a process they would love to automate and someone who understands it well. Give them the tool, the rules and a reviewer, and compare the result with how the work is done today.

If you'd like help planning it, see our practical AI service or talk to us.