This week, Meta cut 8,000 employees and launched the new AI Muse model on the same day. Microsoft offered voluntary retirement to thousands of long-tenured staff. Twenty thousand jobs gone in seven days, from two of the most profitable technology companies in history. The news call it an AI replaces humans. This storyline is is wrong. Meta needs different skills, AI focused skills. They need people who know how to work with AI. Engineers and Product Managers are replaced with AI Engineers and AI Product Managers.
The Job Title Survived. The Job Did Not.
The word “engineer” still appears in job postings. So does “product manager.” But the people being hired for those roles bear little resemblance to the people being let go.
AI Engineers Are Needed
A traditional software engineer writes code to spec. A feature is defined, built, tested, shipped. The system does what it was told. Reliability comes from precise requirements and clean implementation.
An engineer building AI products today works in a fundamentally different environment. The system does not do what it is told. LLMs write code and that means it produces probabilistic outputs. The job is no longer writing features. It is designing loops: how does the agent decide when to stop? What happens when a tool call returns an invalid schema? How do you catch a hallucination midway through a ten-step workflow before it sends the wrong email to the wrong person? These are not software questions. They are system judgment questions.
AI Product Managers Are Needed
Product management is shifting the same way. A traditional PM defines requirements, manages a roadmap, and optimizes for conversion and retention. Take a SAAS platform. It had one UX and thus the jog of a PM was to find the most average workflow. The one that fits for everyone. That is changing. The Product Manager assumes that every workflow is feasible.
Success is no longer “did we build the feature?” It is “how often does the system behave correctly, and how bad is the failure when it does not?”
That shift changes the entire job. We see the rise of AI PMs. (Quick intro to AI Product Management) Evaluation is no longer a QA step downstream of launch. It is a core product function that happens before the first line of code is written. You define failure modes before you define features. You build golden datasets, not just user stories. You think about what happens when the model hallucinates midway through a user’s task, and whether the system recovers gracefully or makes things worse.
The interface is no longer the product. The workflow is. An AI PM is not optimizing screens. They are designing end-to-end outcomes across a system of agents, tools, and human handoffs.
Human Judgment Is Exactly The Point
To be clear those workflows are neither automatic and far from autonomous. Thus It needs new talent. Talent that can write code togeher with an LLM. Talent that can understand the new product design. Essentially while we see those layoffs I stay to my view that we will need humans. Humans that know how to work with AI. Humans that decide on options generated by an AI. This is the same argument I made for creative teams, we need them to work with AI tools but AI cannot replace human judgment.
Companies are not firing engineers and product managers because they want fewer humans. They are hiring different humans. Humans who bring judgment into AI systems rather than working around them. Meta’s departing employees are not being replaced by robots. They are being replaced by engineers who can define termination conditions for autonomous agents, and PMs who know how to build evaluation datasets before a model goes live.
Everyone Needs To Retrain. Now.
McKinsey found that demand for AI fluency in job postings grew sevenfold in two years, concentrated in management and business roles, not just technical ones. This is not a story about coders. It is a story about every professional who touches a workflow.
Retraining does not mean learning to code. I teach AI workflows at Cornell, and the most important shift I see is not technical. It is conceptual. Learning to ask: which step of this workflow does AI change? What does failure look like in a probabilistic system? How do I evaluate a system that does not behave the same way twice?
Silicon Valley is running a live experiment. The companies willing to retool their talent base will come out ahead. The ones still hiring the same profiles for newly redefined roles will keep getting surprised. The job title survived. The job did not. The question is which side of that line you are on.
