American Express is cutting about 5% of its workforce and paying a $336 million pretax restructuring charge to remove management layers before it automates more of its processes. Its CFO's rule, that a process should be "optimally structured" before it is automated, is a sequencing decision every CIO and CHRO with a 2027 agent plan has to make: restructure first and automate a cleaner process, or automate first and let the tools show you which layers are dead weight. Amex chose the first. If you copy it, size the new spans by the manager work that AI will not absorb, and write down in advance what would make you add managers back.
The announcement came with a record-revenue quarter, which makes it a cleaner signal than a cut made under pressure: Amex is reorganizing from strength so its next round of automation lands on fewer, wider teams.
What American Express Announced on October 23
American Express disclosed the cut on its third-quarter earnings call on Friday, October 23, and booked the cost in the same quarter. The company's results release shows $336 million pretax ($253 million after tax) of "restructuring related to efficiency initiatives," alongside a $982 million pretax gain from selling its stake in Global Business Travel Group. Revenue net of interest expense rose 9% to $20.1 billion, net income rose 28% to $3.7 billion, and the company raised full-year EPS guidance to $18.30 to $18.60.
The release itself gives no headcount, no percentage and no list of affected functions. Those came from the call and from interviews. CEO Stephen Squeri told analysts, "We've got good momentum, but to accelerate innovation and win over the next few years, we needed to become a more agile and efficient organization," and added that "when completed, we believe this exercise will help us operate and execute more effectively," as Techstrong.ai reported.
The job count varies by outlet. YourNews put it at about 3,500 roles against a base of about 76,800 employees at the end of 2025, and said the company has not specified which departments, locations or job categories are affected. iHeart also reported about 3,500, citing The Wall Street Journal's description of the goal as an effort to "reduce layers and speed up decision making." Techstrong's figure of roughly 3,840 is 5% of the year-end headcount. Treat the exact number as unconfirmed until Amex files it.
Two pieces of arithmetic are worth having in front of you. At 3,500 roles, the charge works out to roughly $96,000 per position. And the travel-stake gain was nearly three times the size of the charge, so Amex funded its restructuring from a one-time windfall and still posted 28% net income growth. Most companies will have to fund theirs out of the operating budget.
Why Amex Says Flatten First, Automate Second
Amex's stated logic is that automation should be applied to a process after it has been simplified, because automating a badly structured process locks the bad structure in. CFO Christophe Le Caillec put it plainly to CNBC, as quoted by IBTimes: the plan is to "reduce layers and increase spans of controls, so that we can automate," and to "take advantage of AI in the coming quarters and the coming years."
Span of control is the number of people who report directly to one manager. Widening it means fewer managers each overseeing more people. Removing layers means fewer levels between the front line and the CEO. The two usually move together, and both are the classic levers of a delayering program that predates generative AI by decades.
The flatten-first argument has real merit. Every approval step, handoff and status meeting that an agent inherits is one more place for it to stall or for an exception to queue. An agent wired into a seven-step approval chain spends most of its time waiting on approvers. Strip the chain to three steps first and the automation has less to do and fewer places to fail. Amazon used the same reasoning in a delayering push that had nothing to do with agents: Andy Jassy set a target to "increase the ratio of individual contributors to managers by at least 15%" by the end of Q1 2025, saying fewer managers "will remove layers and flatten organizations."
There is also a governance thread in what Squeri said about outside AI. According to Techstrong's account of the call, he said third-party AI assistants will not run unchecked across Amex systems and that outside access is limited to vetted API and Model Context Protocol integrations. That is the same posture large banks took when Schwab and JPMorgan blocked AI agents using customer logins. A company planning to automate its own processes wants a short list of known integration points first.
The Case Against Cutting Layers Before the Agents Work
The strongest counter-argument is that you do not yet know which manager work the agents will absorb, so cutting managers first means guessing. Meta is the live example. It built its Applied AI Engineering division in March at up to 50 employees per manager, and by September Business Insider reported that Meta had begun asking individual contributors in the division whether they would move back into manager roles, according to Fortune, which noted the move was reportedly voluntary and that roughly 7,000 people had been reassigned into the unit. We covered that reversal in detail in Meta Tried 50 Employees per Manager. Now It Wants Managers Back.
The older precedent is Google. In 2002 Larry Page and Sergey Brin eliminated engineering managers, and the experiment lasted only a few months; they relented "when too many people went directly to Page with questions about expense reports." Expense approvals, conflicts, calibration and incident ownership are the work that does not show up in a process map and does not go away when the process gets simpler.
The data on wide spans is conditional. Gallup found the average number of people reporting to a manager rose from 10.9 in 2024 to 12.1 in 2025, while the median held at about five to six and only 13% of managers oversee 25 or more. Managers who spent less than 40% of their time on individual-contributor work stayed more engaged whatever their team size, while those above that line lost engagement faster as their teams grew. Gallup also found that meaningful weekly feedback "nearly triples the percentage of engaged employees." A delayering that also hands each remaining manager a quota of their own individual work fails that test.
The forecasts push the other way. Gartner predicted that "through 2026, 20% of organizations will use AI to flatten their structures," eliminating more than half of current middle management positions. And the regret data from the last wave should temper anyone tempted to front-load cuts: we tracked companies that cut staff for AI and later admitted it was a mistake.
Barclays Made the Same Move One Day Earlier
Barclays announced a cut of similar size the day before Amex, and its framing put more weight on systems modernization than on AI. The bank plans to cut around 5,000 jobs worldwide by the end of 2026, about 5% of its workforce, IBTimes UK reported, with about 1,000 roles already gone. CFO Anna Cross said around 60% of the reductions relate to support functions, including software developers. CEO C.S. Venkatakrishnan called AI "the icing on the cake" and said "it's not the cake," arguing that banks have to build the infrastructure first.
The IBTimes US piece quoted the Barclays results presentation citing "organisational simplification and technology change efficiency driven by AI," and added a caveat worth keeping: the disclosures do not show that every role will be replaced by AI, or how many developer jobs will go. That is also true of Amex.
Barclays is a useful comparison because it is further along with the tools. It split its AI rollout between Claude for developers and Copilot for desk staff, so its developer cuts come after deployment, while Amex is cutting layers ahead of the automation it describes.
Neither is an outlier. YourNews, citing Challenger, Gray & Christmas, reported that AI was cited in more than 120,000 announced job cuts through September 2026, while noting that companies often cite several reasons and not every eliminated role is replaced by automation. Uber's 10% cut in September aimed to reduce managers by about 20% and halve its one- and two-person teams, and TechCrunch's report of it does not mention AI.
How to Test the Flatten-First Sequence on Your Own Org
The decision comes down to whether you can name the manager work that will disappear before you remove the managers. If you can, flatten first, as Amex did. If you cannot, automate one function first and let the measured load set the new spans. BMW faced the same choice when it cut senior roles on the promise of agents that still needed a human signer.
This Week:
- Pull your org chart and count layers between the front line and the executive team in your three largest functions. Mark every manager with fewer than three direct reports. Uber targeted one- and two-person teams in its cut.
- For one function you plan to automate in 2027, list the steps a manager must personally approve or own: expense sign-off, exceptions, calibration, hiring decisions, incident ownership. Agree with HR which of those an agent could take over, and which it legally or practically cannot.
This Month:
- Set span targets by type of work, not one number for the company. A team doing standardized, well-instrumented work can run wider than a team doing judgment-heavy review. Gallup's 40% threshold for manager time on individual-contributor work is a reasonable first check.
- Write a reversal trigger before you cut: the metrics that would make you add managers back (review backlog, regretted attrition, escalation volume, time to resolve exceptions). Meta has given no public reason for its reversal; with thresholds written down, yours would come from a number someone agreed to in advance.
- Get finance to model the restructuring charge against the automation savings by quarter. Amex had a $982 million one-time gain to absorb $336 million. If you do not, the charge lands in the same year as the AI spend, before any savings show up.
Before Your 2027 Plan Locks:
- Run one function through the full sequence (flatten, automate, measure for a quarter) before applying it company-wide. Barclays' order (deploy tools, then cut) gives you evidence. Amex's order gives you speed. Pick one deliberately.
- Brief your security lead on how the leaner org will govern outside AI access. Fewer managers means fewer people approving which tools touch customer data, which is why Amex paired its cut with a statement about vetted API and MCP integration points.
The Bottom Line
Amex has turned a decades-old delayering playbook into an AI readiness program, and the logic holds as far as it goes: automate a simplified process, not a tangled one. The risk sits in the manager work that never appeared on a process map, which is where Google's 2002 experiment broke and where Meta appears to have run short this year. The banks moving now have the profits to absorb a mistake. Before you copy them, list the manager-only tasks in the function you plan to automate first, and set the new spans from that list.
Continue Reading
- Meta Tried 50 Employees per Manager. Now It Wants Managers Back.
- BMW Bets 100 Senior Jobs on AI Agents That Still Need a Signer
- Goldman Says New Hires Manage AI Agents Before Learning the Job
- Apple Pauses 5,000 AppleCare Layoffs Tied to Its AI Support Agent
- Barclays Picks Claude for Half Its Developers, Copilot for Desks
- AmEx Buys Hyper to Embed AI Agents in Corporate Cards