AI Agents & Job Security

Cloudflare cut 20% of its staff, and Meta laid off 8,000 employees while reallocating thousands to AI roles and committing billions to AI capex. This deep dive explains how to secure your career by embracing AI agents.

EXECUTIVE SUMMARY

The latest round of layoffs at Cloudflare and Meta signals a structural shift in how enterprise work will be done.

The cuts are not a sign of weakness. Cloudflare reported record revenue growth, strong free cash flow, and customer growth while cutting more than 20% of its workforce. Meta also reported strong financial results while cutting 8,000 workers and moving thousands more into AI-focused roles.

// Cloudflare let go of more than 20% of its workforce, targeting middle managers, operations jobs, and other roles focused on measuring work rather than building products or selling them.

// Meta dismissed 8,000 employees, about 10% of staff, and reassigned 7,000 into AI-focused roles. Mark Zuckerberg told employees that “success isn’t a given” in the AI race and called AI “the most consequential technology of our lifetimes.”

// Meta plans to spend between $125 billion and $145 billion on AI-driven capital expenditures over the next year.

// The cuts expose a new reality: job security will not come from managing headcount or accumulating reports. It will come from delivering outcomes with lean teams augmented by AI agents.

Before I dive into this topic, I want to be very clear. I think this is a very scary topic for many people. I think there are two camps: one that says AI will replace workers, and another that says it will be better for workers.

My position is that it will be disruptive, and for some of you, it’s going to be a rough ride. That’s partly why I do what I do: I want to do my best not to scare anyone, but to equip you with what you need to have a good chance of successfully navigating AI adoption. So to anyone who’s struggling with this, I feel for you, and I want to help.

I am helping by providing information that gives some of the AI power to people outside of big tech and to the workers who use it.

Now on to the analysis.

In May 2026, Cloudflare CEO Matthew Prince wrote in The Wall Street Journal that the company had less need for middle managers, operations jobs, and other “measuring” positions. The company had strong revenue growth and free cash flow. It was not shrinking because the business was failing. It was shrinking because leadership believed the business needed to change.

That distinction matters.

A company in distress cuts because it must survive.

A company with strong growth cuts because it thinks its operating model is wrong.

That is why the Cloudflare move landed so hard. Prince was not saying, “We cannot afford these people.” He was saying, “This kind of work no longer deserves the same place in the company.”

That is a colder argument. It is also the argument many companies are now testing.

The criticism is obvious. Many so-called “measuring” roles are not useless. Good operations leaders reduce friction. Middle managers preserve context. Program managers coordinate execution. Compliance and reporting teams keep companies from mistaking motion for progress—or from turning a quarter of bad decisions into a subpoena.

The problem is that many companies trained those roles to optimize for visible activity rather than business outcomes.

// Tickets closed.
// Dashboards updated.
// Meetings attended.
// Jira fields completed.

Status decks shipped.

When companies reward measurement theater, employees learn to perform measurement theater. That is not an employee failure. It is a management design failure.

AI makes that failure harder to hide.

Agents can summarize, report, classify, route, and monitor work at a fraction of the cost. They do not need to look busy. They do not need to protect a function. They do not need to sit through the pre-meeting, the meeting, or the post-meeting recap.

That creates pressure on every role that primarily coordinates, reports on, or translates activity between groups.

Meta’s restructuring points in the same direction.

Zuckerberg’s memo framed the cuts as part of an AI race where success is not guaranteed. Meta laid off 8,000 employees while moving 7,000 into AI-focused roles. It also plans to spend up to $145 billion on AI-driven capital expenditures.

That is not a company retreating from the future.

That is a company buying the future while forcing the present to pay for it.

There is a hard lesson here for professionals.

Companies are making long-term investments in AI because they believe future competitiveness depends on it. Individuals need to make the same investment in themselves.

This does not mean every knowledge worker needs to become an AI engineer.

It does mean every knowledge worker needs to become AI-capable.

The old career insurance policy was experience, institutional knowledge, and maybe a few direct reports.

The new insurance policy is operational leverage.

// Can you use AI to produce better work faster?
// Can you manage a workflow where humans and agents share responsibility?
// Can you validate AI outputs, spot failure modes, and improve the process?
// Can you turn a vague goal into a repeatable AI-assisted workflow?
// Can you deliver more without asking for five more people?

That is where job security is moving.

This shift gets more interesting as companies deploy more agents.

Vijoy Pandey, the Cisco executive behind the Internet of Agents initiative, makes a useful point: “A million agents is an economy, not a calculator.”

That line should make every executive think about that: a million agents is a 1-1 relationship for repetitive work, but it’s a one-to-many.

A few agents can be treated like tools.

A large group of agents behaves more like an organization.

At that scale, the problem is no longer just model quality. It is coordination. It is incentives. It is governance. It is trust. It is conflict resolution. It is the digital equivalent of Conway’s Law with a caffeine problem.

Pandey argues that the boundary conditions of multi-agent systems are no longer defined only by computing theory but by game theory. Failure modes that economists have studied for decades are already showing up in production.

That matters for the job security conversation because the winners will not just be the people who “use AI.”

The winners will be the people who know how to organize AI.

You will need to know when to assign work to one agent, when to use multiple agents, when to keep a human in the loop, and when to shut down an automated process before it optimizes the wrong thing beautifully.

Four forces are converging.

1// Measuring work is automatable. AI systems can track and report activity better than humans in many contexts. That makes some layers of middle management less defensible, especially when those layers exist mainly to observe work rather than improve it.

2// Capital is being redirected. Meta’s planned AI capex of $125 billion to $145 billion is not a side bet. It is an operating thesis. If capital is flowing toward compute, infrastructure, and AI teams, companies will scrutinize every role that does not contribute to those priorities.

3// Headcount is no longer the default answer. For years, the enterprise reflex was simple: if the workload grew, hire more people. Now the first question is different: can agents absorb the work? That does not remove the need for people. It changes which people become more valuable.

4// Coordination is becoming the scarce skill. Individual AI tools are useful. Agent ecosystems are harder. As Pandey’s framing suggests, large-scale agent systems begin to behave like societies. They need norms, oversight, incentives, and feedback loops.

That is not a bot problem. That is a leadership problem.

The professionals who thrive in this environment will not be the ones who guard processes or count tickets.

They will be the ones who can increase their team's productive capacity.

// Orchestrate AI agents. Learn to delegate research, summarization, report generation, analysis, and workflow automation to agents, so your team can focus on judgment and execution.

// Prioritize outcomes. Shift from measuring activity to measuring value. Tie work to revenue, customer experience, risk reduction, product velocity, or decision quality.

// Continuously upskill. AI tools are changing fast. The safest professionals will be the ones who can keep absorbing new tools without turning every change into a personality crisis.

// Reframe management. Managers must become orchestrators of people and machines. The work is no longer just assigning tasks. It is designing workflows, validating outputs, setting rules, and deciding where human judgment belongs.

// Govern agent ecosystems. Large groups of agents need incentives, guardrails, and monitoring. Treat them like a workforce, not a plugin.

Outcome over optics. Stop rewarding performative productivity. Measure results and remove incentives for busywork.

Invest in AI literacy. Train teams to use generative AI and agents across functions. Provide safe environments to experiment.

Flatten carefully. Reducing bureaucracy makes sense. Eliminating context and coordination does not. The goal is not fewer humans. The goal is better throughput. Everyone will have access to AI; what they need is teams to leverage that AI.

Redesign roles. Move people from measurement-only work into judgment, orchestration, analysis, customer insight, and AI-enabled execution.

Design agent ecosystems like societies. Large multi-agent systems behave like economies. Use incentives, governance, and feedback loops to align agent behavior and prevent failure modes.

The strategic question for the next budget cycle is not “How many people can we cut?”

It is better framed this way: “How much more valuable can each team become when humans and agents work together?”

That is a better question for companies.

It is also a better question for your career.