Guillermo Rauch, chief executive of Vercel, a $9.3 billion cloud deployment platform founded in 2015, has a message for anyone who spent years perfecting their individual contributor craft: you are now an agent manager whether you like it or not. In a March 2026 podcast appearance, Rauch said the individual contributor role as it existed is effectively over. "My mental model right now is that the IC is the agent," he said. "No one is an IC anymore." The framing he offered instead: everyone is now a "mini CEO."
The reality on the ground at Vercel is more concrete than the rhetoric. In October 2025, the company reduced its 10-person inbound sales development team to a single person managing an AI agent. The agent was not built from a product roadmap or a prompt engineering exercise. Vercel engineers shadowed the team's top-performing sales representative for six weeks, documenting every step of their workflow, then built an agent to replicate that process. The agent now qualifies leads by querying internal databases and researching company details through Exa.ai for web search. The one remaining human handles escalations and reviews what the agent flagged for human attention. Vercel has six AI agents deployed in production and aims to deploy hundreds within the next six to 12 months, all modeled on top performers in specific roles.
The dependency graph is the story. Rauch calls it a "mini CEO." What it actually describes is a human rubber stamp at the end of a pipeline that has already done the work. The agent processes the lead, qualifies it, and routes it. The human reviews what the agent couldn't resolve or what policy requires a human signature for. The risk embedded in this architecture isn't obvious until you sit in that review seat long enough: the cases that reach you are by definition the ones the agent couldn't handle, which means you're making high-stakes decisions on the tail distribution of a process you no longer fully understand.
"The halcyon days of the IC are over," wrote Moloch in a widely circulated Substack post on agent work. "Not because AI codes better than you, but because maximizing your productivity necessitates focusing your time on all the things that are, at the end of the day, manager tasks." The implication is not that management is beneath engineers. It's that the work that remains for humans after agents handle the execution layer is precisely the work that doesn't scale: contextual judgment, exception handling, relationship repair. These are, historically, the skills that separate senior contributors from junior ones. Now they're the entire job.
IBM is deploying this model at a scale that suggests the pattern is not a Vercel experiment but an emerging template. The company says it has built thousands of AI agents now deployed across nearly 300 client projects. IBM Consulting, which generated $21 billion in revenue in 2025, used this agent-assisted approach to complete 52,000 investigations in January 2026 alone. The cases that reach human investigators at IBM are increasingly the ones the agents flagged for review — a structurally identical setup to Vercel's sales team, just with higher stakes and more complex legal review cycles.
McKinsey is openly restructuring its hiring model around the same premise. Alex Singla, a senior partner at the consulting firm who co-leads its AI arm QuantumBlack, told Business Insider in December that McKinsey is looking for what he called "5Xers" — people deeply skilled in one domain who can manage three or four agentic workflows simultaneously. The firm isn't looking for people who can do the work faster. It's looking for people who can oversee the work being done faster, across more domains simultaneously, than any individual contributor could execute alone.
The honest version of Rauch's "mini CEO" pitch is that the role of senior individual contributor is being bifurcated. The execution layer — writing the code, qualifying the lead, drafting the contract — is being offloaded to trained agents. What remains is oversight: setting priorities, reviewing exceptions, allocating resources when the agent's confidence scores don't resolve cleanly. This is real work, and some people will be very good at it. But it is not what anyone who spent a decade becoming a strong IC trained to do. It's a different skill set, closer to a distributed operations manager than to a practitioner.
The companies shipping this transition have an incentive to frame it as empowerment. "Mini CEO" sounds better than "approval workflow with pattern-matching on the front end." The reframe is polished enough that it might even be true for a subset of workers who adapt quickly. But the structural dynamic — the agent processes, the human approves — doesn't change because you rename the role. The dependency graph is the dependency graph. Who sits at the end of the pipeline makes the decisions, and who processes the pipeline makes everything else.
