The Alliance and the End of Loyalty
The Alliance, Hoffman's 2014 book with Ben Casnocha and Chris Yeh, argued that the employer-employee relationship had broken and needed to be rebuilt as an honest, time-bounded alliance. The old contract — lifetime employment for lifetime loyalty — was dead, killed by globalization, software, and the volatility of markets. Pretending otherwise produced bad behavior on both sides. Employees padded resumes and jumped ship. Companies issued vague promises and laid off without warning. The Alliance proposed replacing the fiction of permanence with a series of explicit tours of duty, each with a defined mission, term, and mutual obligation.
In the AI era, that proposal becomes both more urgent and more strained. It becomes more urgent because the half-life of any given role is shrinking. A tour of duty oriented around a specific mission — say, building out a company's machine learning infrastructure — may now compress from years to months as the underlying models reshape what the mission even means. The transactional clarity Hoffman recommended is even more valuable when both parties are navigating capabilities that did not exist last quarter.
It becomes more strained because the very category of employee is unstable. If a single employee, augmented by AI agents, can do the work of a team, what is the team for? Hoffman has been candid that AI compresses headcount in many functions. He has also argued that the same compression frees humans for higher-leverage work and creates entirely new categories of role that did not exist before. The history of technology supports this on long enough horizons. The short horizon, though, is where the alliance has to hold.
The alliance framework also assumes a unit of trust between two parties who can each negotiate. AI agents complicate that two-party structure. When an employee is augmented by a system trained on the company's data and configured to act on the company's behalf, who is in the alliance? The person? The agent? The model provider? Hoffman's answer, drawn from his work at Inflection AI and his board roles, is that the answer is contractual and observable — you define which decisions the human owns, which decisions the agent owns, and which decisions require both signatures. This is messy. It is also the only honest description of how modern teams actually function.
One of Hoffman's most underappreciated contributions in The Alliance is the idea of the corporate alumni network. He argued that companies should treat departing employees not as defectors but as the start of a lifelong relationship. The PayPal Mafia is the canonical example: a generation of founders who left a single company and built much of the modern internet on the relationships forged inside it. In the AI era, the alumni network expands again. Each departing person carries with them not only their human relationships but a portfolio of agents, prompts, and workflows that encode institutional knowledge. The question of who owns that portfolio is going to define employment law for the next decade.
The Alliance was, beneath the practical advice, an argument about dignity. It insisted that workers and employers stop lying to each other and start negotiating honestly about what each owed the other. AI does not change that ethic. It raises the stakes. A workplace organized around honest, time-bounded missions is more resilient to AI shocks than a workplace organized around the comforting fiction of permanence. Hoffman has been pointing at this exit for a decade. The exit has now become the only door.
