June 19, 2026
Meta’s Employee Morale Crisis: Why Snacks Can’t Fix a Trust Problem
Meta’s own technology chief has admitted employee morale is near a 20-year low. The company’s response – capping manager workloads on one hand, raising the snack budget on the other – shows leadership still misreading the problem it created.
When CXM reported on the AI backlash building inside Meta, Google DeepMind, and Amazon, the discontent was largely bottom-up: petitions, bathroom-stall flyers, and union votes. The signal has now reached the top, and a senior leader is saying out loud what employees have been signalling for months.
At an internal “Tuesdays with Boz” session on 2 June, Meta Chief Technology Officer (CTO) Andrew Bosworth told staff that morale was “maybe not the worst it’s ever been in 20 years here, but it’s probably up there”, comparing the mood to the fallout from the Cambridge Analytica scandal, according to Business Insider.
In a follow-up memo first reported by Wired, Bosworth went further. He conceded that leadership had done an “atrocious job explaining the vision” behind the AI restructure, and admitted the company had “undermined the trust” employees placed in the value of their own expertise and career growth. Candour of this kind from a CTO is unusual.
A Rare Admission, and a Revealing Remedy
Bosworth’s memo also set out what Meta intends to do about it: cap managers at roughly 20 direct reports, move staff between teams less often during reorganisations, and let employees reassigned to AI work apply for other internal roles if they want out. Alongside those structural changes, Wired reports the company is increasing budgets for travel and social events, as well as improving office break areas stocked with snacks and drinks.
Some of these moves address real grievances. The manager-span cap and the loosening of transfer rules speak directly to the lack of agency employees described. The snack budget does not. It is the reflex industry analyst Josh Bersin warned about, noting workers “don’t want any more perks”, they want fair pay, flexibility, and some assurance of a long-term career.
Meta has also softened the surveillance policy that drew so much anger. After more than 1,500 staff signed a petition against software that logs keystrokes and screenshots to train AI agents, the company now lets employees pause the tracking for 30 minutes at a time, according to Reuters. A concession, but a narrow one, and one that leaves the underlying practice intact.
Meta’s AI Division a “Gulag”
This is the same dispute, further down the road. The Applied AI unit formed in the restructure pulled in around 6,500 engineers and product managers, many of whom had little say in the move. One worker described the division to Wired as a “gulag” and likened mandatory data-labelling work to being drafted. The bitterness has a clear root. The layoffs landed weeks after Meta posted $26.8 billion in net income for the first quarter of 2026, and asked employees to train the AI systems many fear will replace them.
What began as bottom-up rumblings has escalated into something more structural, and Meta’s leadership is now openly acknowledging the cost.
The AI Policy Whiplash Behind Meta’s Falling Morale
The trust problem is compounded by a second pattern: a leadership that keeps changing its mind. In 2025, Meta declared “AI-driven impact” a core performance expectation. In spring 2026, it ran an employee-built internal leaderboard, reportedly nicknamed ‘Claudeonomics’, ranking the top 250 employees by the volume of AI tokens they consumed.
Employees did exactly what the metric incentivised, spinning up parallel agents to inflate their rankings, and total token use jumped to 60.2 trillion in 30 days and reached 73.7 trillion before the company pulled the leaderboard.
Now the message has reversed again. According to The Information, Meta has told around 6,000 employees it is restricting internal AI usage as costs head toward billions of dollars in 2026. Staff who were ranking for maximum AI consumption are now being rationed.
For employees, the whiplash is breaking down trust even further. Being told to maximise a behaviour, then being penalised for it is not a strategy they can buy into.
Why Perks Won’t Fix Meta’s Employee Morale Crisis
“In a high-trust culture, free food, office dogs, and wellbeing gestures give off a ‘care signal’; it all feels like generosity,” says CX and EX Strategist Danny Seals.
“In a low-trust culture, however, they start to feel like bait. Employees judge perks with a wide lens, looking at what they sit next to. Care signals placed next to device tracking, reduced raises, and AI replacement get converted into a ‘consequence signal’.”
For Seals, that is why Meta’s response cannot work as intended. “No matter how good your perks are, they won’t fix this, because they are all being received in a low-trust culture. It’s like asking someone to trust the chef while they’re watching rats run across the kitchen.”
Culture and leadership expert Deborah Hartung is blunter still. “Meta is offering a masterclass in what not to do during AI restructuring,” she says. “The speed and impunity of these moves show they care more about winning than about their people, and the proof is in the response: employees reeling from mass layoffs, surveillance, and being told to train their own replacements, and the best Meta can offer is snacks and travel.”
The Lesson for Every Leader
Meta is an extreme case, but the dynamics are not exclusive to big tech. Any organisation restructuring around AI is making the same implicit promise and risking the same breach. Trust is built through consistency and structural commitment, not gestures, and once it is spent it is not easily bought back with benefits.
For CX-led organisations, the stakes are higher than headcount. A distrustful, disengaged workforce is the same workforce serving customers, and the discretionary effort that disappears first under these conditions is precisely what good customer experience depends on. Leaders who respond to a trust deficit with perks are answering a question their people did not ask.
