Everyone Says Fire the Humans. Your Customers Just Said Don't.

Read the business headlines from May 2026 and the verdict looks settled: AI means fewer humans.

I think that verdict is half right. Automate the work your customers never see, and keep capable people on the surfaces where customers decide whether to trust you, because that is where the cost of cutting arrives late and lands hard.

On May 5, 2026, Coinbase cut roughly 14 percent of its staff and did something more interesting than the number. Its chief executive, Brian Armstrong, did away with the “pure manager” (the person who oversees work without doing any) in favor of player-coaches. He capped the leadership structure at five layers below his own. And he set out plans for what he calls AI-native pods: small teams, possibly a single person, directing AI agents that do the work of an engineer, a designer and a product manager. Two weeks later, on May 20, Meta cut about 8,000 jobs. According to an internal memo reported by Bloomberg, it also moved roughly 7,000 people into new AI groups with names like Applied AI Engineering and the Agent Transformation Accelerator. Different companies, same move: shrink the human layer, point what remains at the machines. My worry is less these two firms than the leaders who copy the move onto the surfaces customers touch.

I want to be fair to this view, because it is the smart-money position. If an agent can do the work of a designer, an engineer, and a product manager at once, then a one-person pod is a margin, not a gimmick. The companies making these cuts are reading the same capability curve everyone in my line of work is reading, and they are acting on it before their competitors do. On the cost side of the ledger, the math is real.

But there is a second ledger, and the customers keep it.

The customers are walking the other way

That same spring, while the market decided AI means fewer humans, the people who actually pay for things moved in the opposite direction.

Start with the independent data, because it is the cleanest. It measures how people feel about AI in general, so read it as the backdrop; the service question comes next. Prophet's 2026 AI-Powered Consumer Report, a survey of about 2,000 AI users in five countries published in April, found that overall excitement about generative AI has fallen about 7 percent since its 2024 study. More tellingly, the belief that people will come to rely on AI for most of their decisions dropped by 30 percent. I read a fall that size as consumers scaling back the role they expect AI to play in their lives. Gallup found the same current running through the youngest, most fluent users: among Gen Z who use AI every day, excitement about it fell 18 points in a single year. In that generation, the heaviest users are the ones growing more wary.

A separate survey points the same way, and I will tell you exactly what it is so you can weigh it honestly. AnswerConnect (a company that sells human answering services, so treat it as evidence from an interested party) commissioned a OnePoll survey of 6,000 consumers across the US, UK, and Canada and published it in May. It found that 85 percent now say they would rather reach a real person, up from 83 percent, and that 57 percent would trust a business less if it relied mainly on AI for customer service. They are also the only figures here that ask about service directly, so treat them as a signal to test on your own customers before you treat them as proof.

Put the two ledgers side by side. The market is pricing humans as cost. The customer is pricing humans as trust. Both are right about their own column. The mistake is assuming the first column is the only one that bills you.

Why the human surface becomes a moat

It would be easy to read all of this as nostalgia: people grumbling about robots, the way they once grumbled about self-checkout. That reading is comfortable and I think it is wrong.

Here is the part that should interest a leader rather than annoy one. My bet is that when every competitor automates the same customer surface at once, the brand that keeps a capable human there becomes the premium option, provided customers pay for what they tell pollsters they want. Scarcity does the pricing, the same logic that re-scores Porter’s five forces once intelligence is cheap. If reaching a person is the rare thing, reaching a person is the valuable thing; trust has switching costs that a chatbot will never have. People will forgive a slightly higher price from a company they trust. What drives them out is being made to feel processed.

This is why I keep saying it.

“AI without trust is just expensive software.”

It runs, it scales, it impresses the board, and it quietly teaches your best customers that you are no longer worth the loyalty.

The cruelty of it is the timing. The headcount saving books this quarter: clean, visible, applaudable. The churn it causes can take two, three, four quarters to surface, by which point it reads as “market conditions” or “a tough renewal cycle” rather than as the bill for a decision someone made in a planning meeting last spring. You automated the wrong surface, won the cost argument, and lost the customer on a delay long enough that nobody connects the two events. That delay is the whole danger: it lets a bad trade look like a good one for exactly as long as it takes to become irreversible. The guard against it is cheap. Before you cut a customer surface, baseline its churn and complaints, then check them every quarter.

A map for deciding what to automate

Of course you automate. The real question is where. I suspect automation gets scoped by department, by budget line, by whatever is easiest to rip out, and rarely by what the decision does to trust.

The tool I would hand any leader making this call is what I call the Trust Surface Map. It has two axes.

The first axis is how much trust a surface carries: from low (a status page, a shipping notification) to high (a billing dispute, a cancellation, a moment where something has gone wrong and the customer is scared or angry). The second axis is how visible the surface is to the customer: from invisible back-office work they will never see, to the front edge where they meet your brand directly.

That gives you four quadrants, and four different instructions:

• Invisible and low-trust: reconciliation, routing, data cleanup, internal forecasting. Automate aggressively. No customer will ever feel it, and the savings are pure.
• Invisible and high-trust: fraud scoring, credit decisions, anything where a machine judgment reaches the customer through a wall. Automate, but with a human audit. The customer does not see the process, but they live with its consequences, so someone accountable has to be checking it. This quadrant needs the strictest audit and a named owner, because the customer cannot see enough to object.
• Visible and low-trust: order tracking, appointment booking, the simple front-door stuff. Augment, do not replace. Let AI handle it, but keep an obvious escape hatch to a person, because a small problem becomes a trust problem the moment the customer cannot get out of the loop.
• Visible and high-trust: the complaint, the cancellation, the apology, the high-stakes purchase. Protect this. Keep humans here. This is the surface your competitors are about to strip-mine for savings, which is exactly why holding it becomes your advantage.

My guess: most of the cost saving sits in the quadrants no customer can see, and most of the trust in the one everyone tells you to cut first. The companies that win the next cycle will be the ones that automated the invisible work and defended the human surface, while everyone else discovers the bill on a delay.

The quieter mistake

Automating the customer surface is the visible error. It is the one your customers will eventually punish, on their own schedule, in churn you will struggle to trace.

There is a quieter one, and it is happening in the boardroom rather than the contact center. It is the decision to authorize an AI system you cannot actually switch off: the org has wired itself so deeply around the automation that turning it back into a human process is no longer an option anyone will sign for. The customer-trust mistake costs you revenue. That one costs you control. I take it up in Your Board Can’t Turn It Off.

“For now, hold one line, because it is the one this whole argument turns on: trust doesn’t show up on the balance sheet until it leaves it.”

Sources and Changes

• Fortune, "Coinbase didn't just lay off 14% of its staff due to AI", 5 May 2026, and Fortune on Coinbase's player-coaches, 5 May 2026.
• NPR, "Meta slashes 8,000 jobs as it pivots towards AI", 20 May 2026, and SiliconANGLE on the four new AI units, 19 May 2026.
• Prophet, "The 2026 AI-Powered Consumer Report", April 2026 (2,015 respondents in China, Germany, Singapore, the UK and the US, surveyed January to February 2026).
• Gallup, "Gen Z's AI Adoption Steady, but Skepticism Climbs", 8 April 2026 (1,572 people aged 14 to 29, surveyed 24 February to 4 March 2026, with the Walton Family Foundation and GSV Ventures).
• AnswerConnect on its consumer research, 13 May 2026, and the full AnswerConnect report (OnePoll, 6,000 consumers in the US, UK and Canada). Vendor-commissioned.

All sources checked on 6 October 2026.

Changes on 6 October 2026. New dek, summary and takeaways, and the claim now comes up front. I re-checked every fact. Coinbase announced its AI-native pods as a plan, so "started building" became "set out plans". Meta's 7,000 transfers come from an internal memo reported by Bloomberg, and the second team's name is the Agent Transformation Accelerator. The AnswerConnect poll was published in May, not run that month. Three lines that stated my judgment as fact, including one about the leaders I advise, now read as judgment. The Trust Surface Map is redrawn as a chart that reads on a phone.