AGI as Team-Scale Capability
Hoffman defines AGI not as a single moment of machine consciousness but as a threshold of practical capability: the point at which one person, augmented by AI agents, can do what previously required a team. By this definition, he argues, we are at or close to AGI in 2025. The definition is deliberately deflationary. It strips away the metaphysical drama of machines waking up and replaces it with a practical, observable metric — how much team-scale work can one person now direct?
The definition has the virtue of being testable. You can look at specific workflows — drafting a legal brief, designing a marketing campaign, writing a research paper, building a software prototype — and ask whether one person with current tools can complete them at a quality level that previously required a team. In many cases the answer is now yes. In others it is not yet, but the trajectory is clear. By this measure, AGI is not a future event to be feared or anticipated. It is a present condition to be navigated.
The deflationary move is strategically important for Hoffman's broader argument. The maximalist definitions of AGI — superintelligence, machine consciousness, autonomous goal formation — produce anxiety that can paralyze policy and investment. The deflationary definition reframes the question. It says: forget the philosophical horizon and look at what is happening now. What is happening now is that capable agents are being deployed in workflows, and the question is how to design those workflows so that the human remains the conductor rather than the spectator.
There are good reasons to be skeptical of the deflation. The maximalist definitions are not just rhetorical. They point to genuinely different scenarios — scenarios in which AI systems set their own goals, evade their own constraints, or develop capabilities their designers did not intend. These scenarios are not made less plausible by being uncomfortable to discuss. Hoffman would argue that the right response is to keep those scenarios in research-mode, to address them through alignment research and interpretability work, but not to let them dominate the practical conversation about how to integrate currently capable systems into actually functioning economies.
The team-scale capability framing has practical consequences for organizations. If one person can do team-scale work, what is the team for? Hoffman's answer is that teams remain valuable for the kinds of work that benefit from genuine diversity of perspective — strategic decisions, ethical judgments, creative work that requires multiple sensibilities — but that the volume of work that requires teams shrinks. Organizations become smaller, more agile, more dependent on the integration capacity of their senior people, and more reliant on agents for the work that used to be done by junior staff. This has consequences for how junior staff acquire experience, which is one of the genuinely difficult problems Hoffman has acknowledged but not fully solved.
The biological analogy returns here. Just as Manas AI is betting that biological complexity can be compressed by AI-augmented research, the team-scale capability argument is betting that organizational complexity can be compressed by AI-augmented individuals. Both bets are placing capital on the proposition that the cost of doing complex things falls dramatically when the right tools are in the right hands. If the bets pay off, the world that emerges is one in which more is done by fewer people, with consequences for employment, for institutional design, and for the political economy of cognitive labor that are still being worked out. Hoffman is not pretending the consequences are simple. He is arguing that they are coming, and that the alternative to navigating them deliberately is being navigated by them.
