Barclays gives itself three months to put half its developers on Claude Code

Both live deployments are retrieval and sorting jobs. In each one a person acts on what the model returns: staff answer the customer, and operations teams handle the routed request ^[7][8]. The assistant's million-plus searches across more than 16,000 users come to roughly 62 a head since 2025 ^[12]. At 120,000 emails a day, the Global Markets system gets through that many items in a little over eight days ^[13].

Claude Code is a different kind of deployment, aimed at the software itself. Its target counts people: half the developer population by the close of 2026, and a majority of software engineers during 2027 ^[2]. The announcement came on October 1 ^[1], so the first milestone is three months away ^[14]. The bank did not disclose its developer headcount, how many staff already use the tool, or what the expanded contract is worth.

The two milestones are closer together than the wording suggests. A majority is any share above half, so any figure above the 50 percent set for 2026 meets the 2027 goal ^[15]. The targets also use different nouns, developer population and software engineers ^[2], and those need not be the same group. Anthropic's chief commercial officer, Paul Smith, went further and summarised the footprint as reaching most engineers by 2027 ^[11].

I see three ways this plays out. One is distribution: licences go out on schedule, and use per engineer looks like the assistant's 62 searches a head ^[12]. A second is the outcome group co-COO Craig Bright described, with the tool doing real work on legacy refresh and software quality ^[4]. The third is that the governance, security controls and human oversight the bank says remain central ^[9] limit how much of the build-and-run cycle agents may touch.

I'd expect distribution to dominate through year-end. Three months is enough time to hand out access and too little to rewrite a legacy platform. The counter-case is that Barclays set the target after internal coding results it has kept private.

Both co-COOs say saved time will go to harder problems ^[4][5]. Anne Marie Darling's stated aim is a simpler organisation inside a controlled environment ^[5]. As described, then, the plan keeps the engineers and changes what they work on. A year-end figure for Claude Code output, defects or technology cost, reported as precisely as the email count, would support the second outcome and prove this view wrong.