American Express cuts 5% of jobs to prepare for AI and limits agents to API and MCP

American Express cuts about 5% of roles to get ready for AI

American Express is eliminating about 5% of its positions globally and has taken a $336 million restructuring charge, chief executive Stephen Squeri told analysts on the company's third-quarter earnings call on 23 October 2026. In an interview with CNBC, chief financial officer Christophe Le Caillec said the restructuring was meant to make the company "leaner, more agile, and also to prepare for AI."

The charge landed in a strong quarter. Amex reported record revenue of about $20 billion, up 9%, beat profit expectations and raised its full-year earnings guidance. Squeri framed the cuts as a choice to spend part of a one-off investment gain on making the organisation more efficient and better prepared for the future, rather than as a response to weakness in the business.

The same call carried one of the most detailed statements yet from a large card network on how it will treat AI agents that try to shop, book and pay for customers. Squeri said Amex will not let assistants roam freely through its systems and will only admit agents that come through approved channels.

Why the CFO wants fewer layers before more automation

Le Caillec's explanation is the clearest window into how the cuts relate to AI. "You don't want to automate processes that are not optimally structured," he told CNBC. "So what we're doing here is to reduce layers and increase spans of controls, so that we can automate and take advantage of AI in the coming quarters and the coming years."

In plain terms, the company is flattening its management structure first. Fewer layers means each manager oversees more people, and the processes those managers run become simpler to map and hand to software. The cuts are not described as AI already doing the work of the people leaving. They are described as clearing the ground so automation lands on a cleaner organisation.

That distinction matters for how the numbers are read. Studies such as Goldman Sachs economists' estimate that AI could displace up to 9% of US workers model direct substitution. Amex is describing something earlier in the chain: a restructuring justified by planned automation, with the productivity gains still to come.

How many people the cuts affect

Amex did not give an exact headcount on the call. CNBC reported that the company had almost 77,000 employees at the end of 2025, which would put 5% at a little under 3,900 roles. Squeri told Bloomberg Television that the company has 78,000 employees, and a Bloomberg report syndicated by Moneycontrol said the move would affect about 3,500 jobs.

The gap between those figures likely reflects positions that are already open or will not be refilled rather than people losing jobs, but Amex has not broken that down. It has also not said which functions, countries or levels are most affected beyond the stated goal of removing management layers.

On the earnings call, Le Caillec quantified the charge at $0.37 per share. It was partly funded by the sale of Amex's stake in Global Business Travel Group, which produced a gain of $1.11 per share, according to the call transcript.

The five places Amex says AI changes its business

Squeri spent a large part of his prepared remarks on AI from the customer's side rather than on cost savings. He said Amex is already using AI to increase productivity and automate processes, but that "Agentic transaction volumes are still small, but the direction is clear." He then laid out five areas where Amex is applying AI, according to the transcript.

The first is value propositions, where a dining companion uses card spending insights to recommend restaurants through Resy. The second is customer acquisition, including how Amex products are represented by AI platforms and reaching prospects through language models and agents. The third is engagement: Amex is deepening its partnership with OpenAI, has launched Resy discovery and reservations in ChatGPT, is working on an Amex plugin for discovering and enrolling in Amex Offers, and offers Perplexity-based tools for eligible business card members.

The fourth is payments, the most relevant area for AI developers. Squeri pointed to the Agentic Commerce Experiences, or ACE, developer kit published in the second quarter, the recently released Intent Intelligence framework for agent purchases, and Agent Purchase Protection, which is in development and would cover eligible purchases when registered agents that share intent data make mistakes. The fifth is service, where Amex recently launched its first few conversational AI voice agents.

Agents get in only through approved API and MCP channels

Asked about AI and rewards later in the call, Squeri gave a blunt account of the company's stance on outside agents. "We're just not going to let assistants run rogue through our systems. That's not going to happen," he said, per the transcript. "We need to protect our customers, merchants, and we're only going to allow agents access to our systems that are either approved through API or MCP channels."

He illustrated the concern with Resy, the reservation platform Amex owns. Amex does not want AI agents "pinging Resy every four seconds and disadvantaging our other card members," he said, adding that the company will be selectively looking at agents and how they act. He also noted that some banks have been blocking assistants and said Amex would take a risk-based approach too.

For developers building shopping or booking agents, the message is that screen-scraping and unattended browser automation against Amex properties are unwelcome, while sanctioned integrations are the intended route. That mirrors a wider shift in finance toward formal agent access, such as Mercury letting ChatGPT and Claude request payments through MCP write actions. Amex did not say which MCP servers or partners are approved today.

The quarter that paid for the restructuring

The financial backdrop explains why Amex could absorb the charge without changing its outlook. Reported earnings were $5.42 per share and adjusted earnings $4.68, ahead of the $4.48 average analyst estimate compiled by LSEG, CNBC reported. Amex reaffirmed revenue growth guidance of 10% and raised its full-year earnings guidance to $18.30 to $18.60 per share on a reported basis, or $17.60 to $17.90 adjusted.

Squeri also said adjusted results included an expense for legal matters covering the unreserved part of a recently announced civil money penalty from the Office of the Comptroller of the Currency, and that the fine does not affect 2026 guidance, according to the transcript. Billed business grew 8%, US consumer spending rose 9.4%, and restaurant spending was up 9%.

Investors were less enthusiastic than the beat suggested. Shares fell roughly 4% in afternoon trading on Friday, CNBC reported.

Banks keep tying job cuts to AI

Amex's announcement came in the same week that Barclays said it would eliminate 5,000 jobs and named its first chief AI officer. The Bloomberg report noted that HSBC is planning reductions in its UK wealth business and that Citigroup has also been lowering headcount, with some financial firms attributing changes to greater use of AI.

The pattern is consistent: large financial companies are pairing headcount reductions with explicit AI language, while describing the AI benefits as future rather than current. That makes the next few quarters a test of whether the promised automation shows up in expense lines and service quality, or whether AI becomes a convenient label for routine restructuring.

What to watch next

For staff and investors, the open questions are where the cuts fall and whether Amex reports savings attributable to automation. The company has not said when the eliminations will be completed or how much annual expense they will save, and it did not split the $336 million charge by region or function.

For the AI industry, the more consequential details are in the payments roadmap: when Agent Purchase Protection ships, which AI companies integrate the ACE developer kit, and how Amex decides which agents count as approved. A card network that explicitly limits agents to API and MCP channels gives agent builders a clear rule, and makes Amex's partner list one of the more important documents in agentic commerce once it is published.