LAYOFFs - The Org Chart got smaller. The Job Didn't

WEEKEND READ: What real 2026 data says about AI layoffs, shrinking teams, and who's left holding the WIP limit.

Based on known, published market data, by September 29, 2026, 519 layoff events had occurred that year. Approximately 225,000 jobs were lost. Quite obviously, AI was the single most-cited reason for five straight months. What's more interesting than the scale is how shaky the story gets once you look at what the executives doing the cutting actually said.

Amazon's Andy Jassy told employees in an internal memo that AI would let the company do more with fewer people. As a result, the corporate workforce will shrink. Later, on an investor call, he called it culture.

A Resume Genius survey of 1,000 laid-off workers found 53% believe automation contributed to their job loss. In the same research, nearly 60% of companies admitted, internally, that they sometimes frame cuts as AI-driven because it plays better with investors.

Gartner ran the numbers on whether any of this is actually paying off. Surveying 350 executives, they found roughly 80% of organizations piloting autonomous AI report workforce reductions. The reduction rate is statistically identical whether a company is seeing real ROI from AI or seeing none at all. Their own line on it: “Workforce reductions may create budget room, but they do not create return.” The organizations that actually benefit, per their VP analyst Helen Poitevin, aren't the ones eliminating people. The ones benefiting are the ones that restructure how the remaining people and the AI work together.

That distinction is the whole article. Cutting headcount and redesigning how the smaller team operates are two different projects, and many companies right now are only doing the first.

What It Costs the People Still in Job

Coinbase is one of the more candid examples of a company doing the second thing. CEO Brian Armstrong executed a 14% cut in May 2026 and eliminated around 700 roles. This restructuring explicitly eliminated “pure managers”. He pushed toward “AI-native pods.” A single person can direct agents that cover what used to be separate engineer, designer, and product roles. It's a real, stated strategy, and it's the clearest current answer to “what does the org actually look like after this.”

Gallup's 2026 State of the Global Workplace, with over 141,000 respondents, found global employee engagement fell to 20%, with manager engagement specifically dropping from 31% in 2022 to 22% in 2025.

Perceptyx found that after an actual layoff round, trust in leadership drops by 10 points and confidence in the company drops by 17.

None of that is unique to AI-cited layoffs specifically. What is specific to this wave is a phenomenon Amy Edmondson and Jayshree Seth named directly in Harvard Business Review this year: “trust ambiguity”.

Trust Ambiguity: A workplace where people are implicitly expected to trust the AI to do more of the work, but the drop in actual team trust is hard to pinpoint or discuss openly because nobody's sure whether the discomfort is about the tool or about each other.

Lencioni's model: trust is the base of his pyramid for a reason, and “trust ambiguity” is a genuinely new way for that base layer to crack.

What the Frameworks Are Actually Saying

Given how much is being asked of smaller teams, it's worth checking what Scrum, SAFe, and Kanban's own institutions have said about running that way. The answer is uneven.

Scaled Agile Inc. has moved the furthest, in writing. Their “AI-Native SAFe” material, published this year, frames it directly as the operating model for what they call the Age of AI, with one line worth keeping:

“Human judgment remains the final, non-negotiable loop for value, safety, and purpose.” — Scaled Agile Inc., “AI-Native SAFe”

That's a real commitment from the body that owns the framework. A specific kind of human judgment doesn't get automated out of the loop, no matter how small the team gets.

Kanban University has published nothing on adapting Kanban or WIP limits for teams that include AI agents.

As it drifts away from the official bodies, independent practitioners are offering more concrete thinking. Kanban consultant Yuval Yeret has written about Rakuten and Ramp running multiple parallel AI coding-agent sessions inside their engineering workflows, and proposes one actively guided feature per pod, treated as the WIP limit for a human now supervising several agents at once rather than writing code directly.

AWS's own prescriptive guidance goes further and reframes the vocabulary entirely. Planning becomes “Intent Design,” Testing becomes “Behavioral Evaluation”. The guidance states plainly that story-point velocity and fixed sprint planning lose their usefulness once agents are doing the execution.

The old ceremonies don't map cleanly onto this shape of team, and what replaces them is still being figured out in public.

What Happens When the Loop Gets Too Thin

Here's the part of this research that should worry anyone treating “smaller team, same output” as a solved problem. Meta ran exactly that experiment, shrinking engineering groups from 10–20 people down to 3–5-person AI-augmented pods (reported by LeadDev this September)

Code output rose sharply; the outcomes that actually matter (features shipped) barely moved, while incidents and firefighting rose faster than anything else on the chart.

It's a story about what happens when measurement stays fixed on output while working software reaching users without breaking quietly falls behind. DX's Brian Houck put it plainly: “We are often falling in love with our output numbers and not paying enough attention to our outcome metrics.” Gather.dev founder Peter Bell's take, from the same reporting, is that the evidence still points to 5–8 humans as a functional team size.

According to CircleCI's 2026 delivery report across 28 million CI workflows, coding activity is up 59% year over year, but median team throughput actually declined 7%.

More code isn't more delivery. Meta's own numbers are the receipt.

One Person, Converging Role

This pressure also shows up in what a single surviving team member is now expected to know. Scrum Master hiring has dropped sharply.

Real data from Qualify Nation's job-market analysis. A separate dataset (Ravio) shows a 46% relative decline over roughly the same period using different methodology.

The two sources don't fully agree on why: one attributes it to AI absorbing facilitation work, the other to Agile practices simply getting embedded directly into engineering teams rather than sitting with a dedicated role. What's less ambiguous is where those responsibilities are landing. An analysis of nearly 9,700 active product manager postings found Agile skills now appear in almost a quarter of them, and when a PM posting mentions Agile at all, it's more than three and a half times as likely to also require Scrum specifically. This provides direct evidence that delivery-management skills are folding into the product manager's job description rather than staying separate.

Userpilot's CEO, Yazan Sehwail, said it as plainly as anyone has:

“You're no longer operating. The AI is operating. You're just basically evaluating and monitoring the agent workflow.” — Yazan Sehwail, CEO, Userpilot

His read is that the product manager and Scrum Master are emerging as one role, because both disciplines are offloading responsibilities/tasks like backlog grooming, status updates, and sprint reporting to the same agents.

It’s called a skill-stacking problem. Product strategy, stakeholder management, and delivery-system discipline used to be three different roles for three different people. Now one person is expected to hold all three. This creates a problem of decreased peer coverage; at the same time, the delivery system underneath them gets harder because half the “team” is agents.

Where This Leaves? A Smaller Team?

Nobody has a clean case study of a team that shrank and got measurably better at it through flow discipline alone. Teams that only cut headcount, without also redesigning how the remaining humans track WIP, produce more output, fewer real outcomes, and more firefighting.

If the gap on your team is the skill stack- one person now needing to know flow metrics, delivery discipline, and product judgment- we run the structured, accredited training built for such roles.

If the gap is the system itself - designing how a shrunk team actually holds WIP limits, ceremonies, and accountability with agents doing real execution work - that's what we consult and coach on.

SOURCES

1. CNN, Amazon layoffs and AI; Fortune, Jassy on “culture, not AI”
2. Fortune, Coinbase org-chart restructuring
3. Resume Genius, 2026 AI Layoffs Report
4. Gartner, “AI Layoffs May Create Budget Room But Do Not Deliver Returns” (May 2026)
5. Gallup, State of the Global Workplace 2026; Perceptyx, “The Layoff Aftermath” (May 2026)
6. Jayshree Seth & Amy Edmondson, “How to Foster Psychological Safety When AI Erodes Trust on Your Team,”HBR (Feb 2026)
7. Scaled Agile Inc., “AI-Native SAFe”
8. Yuval Yeret, “Calculating Kanban WIP Limits in the AI Age”
9. AWS Prescriptive Guidance, “Operationalizing Agentic AI: Software Delivery”
10. LeadDev, “Meta Tried to Shrink Engineering Teams Around AI” (Sept 2026)
11. CircleCI, 2026 State of Software Delivery
12. Ravio, LinkedIn post on Scrum Master hiring decline; Qualify Nation, “UK and Global Agile/Scrum Job Market: 2021–2026 Analysis”
13. InterviewStack.io, “Product Manager Skills in 2026”
14. Userpilot, “What Is Product Management” (May 2026)