Why this Barclays engineer says AI speed is a trap

This article summarizes an episode of Brave’s video series featuring Andy McMahon, principal AI engineer at Barclays.

Image credit: Timmy Loen

The least expensive AI projects in a company can become the hardest to justify once an employee actually has to manage them.

Andy McMahon, principal AI engineer at Barclays, argues that corporate AI tools should earn their independence through clear evidence and strict controls.

His view frames the real AI competition around determining which tools are truly ready for everyday business use.

Inexpensive early projects create AI distractions

Low-cost AI makes it easier to test new ideas, but it also makes companies less likely to reject bad ones. McMahon worries that teams often confuse technical possibilities with actual financial value.

The first review needs to happen before a simple test project gains too much internal support. McMahon argues, “A developer, an engineer, or a scientist will always ask, ‘Can I build this?’ But before that, they should ask, ‘Should I build this?’ especially now, because the barrier to entry is so low.”

This decision becomes harder when multiple teams discover the same obvious idea at the same time. “Everyone says, ‘I can build a bot that reads Jira tickets,’” McMahon adds. “What you end up with is 100 bots that read Jira tickets. Is that really the best use of your time?”

Duplicate AI tools compete for attention, require ongoing support, and waste money on software that may never be truly useful. Security reviews and employee training then consume even more resources.

A better approach asks which daily tasks will improve enough to justify managing the tool for the long term.

Writing code faster is the wrong way to measure success

New software tools can make mere activity look like actual progress. In many companies, work slows down during the approval process or when connecting different systems. Writing software faster might not solve the real bottlenecks, even if reports show more work getting done.

McMahon questions the belief that writing software faster automatically leads to better business results. He notes, “You’re seeing people say, ‘Writing the code was never the bottleneck.’ We weren’t sitting there as developers saying, ‘I wish I could write 10x faster,’ and we’re not seeing that translate to 10x more revenue.”

The term AI becomes less helpful as these tools turn into normal everyday software. For McMahon, “The goal is to stop saying ‘AI.’ We don’t run around saying, ‘I’m using a database. Aren’t I so innovative?’ You want [AI] to reach the point where it’s table stakes.”

If a company cannot process new software fast enough, the AI only provides a minor convenience for the individual worker. Companies should measure how well the work performs once it is live, looking at direct revenue and tracking error rates.

Internal testing builds necessary trust

Independent tools need strict boundaries