Why hire humans?
Setting aside regulatory reasons, in a free market humans hire other humans for three reasons:
- Task skill
- Access to resources (equipment, connections, etc.)
- Demonstrated judgement in a field
Especially for knowledge workers, once computer-use AGI dramatically lowers the cost of task outputs for an enterprising user, the value of collaborating with other humans accrues to 2 and 3. 2 is straightforward.
For 3, it's the reason a 40-year-old executive, even with AGI, would probably still hire a Gen Z to run their company's TikTok. The executive rents trust and judgement from the Gen Z employee to deliver better results in expectation than the AGI alone with no supervision. The same goes for specialists (e.g. an expert in soft-body physical simulation, or ClickHouse hiring Andy Pavlo; I pick fields with no regulatory requirements to remove that confound), and at almost every level, from schools all the way to countries and MNCs, because the math stays the same.
The corollary of 1 is that most things that can be readily abstracted into the shape of a Task, with clear goals, constraints, and verification methods, face automation pressure. This is certainly true at the consumer level, and in many entry-level blue-collar jobs, like truck driving, but also in many entry-level white-collar ones. As model capabilities increase, the positive expected value of expert supervision on a task of fixed difficulty decreases, to the point where it's cheaper to simply have the model complete the task autonomously. This is already the case for many of the mundane tasks involved in software engineering.
The consequence of this rising bar is that it is no longer enough to stay competent at a task of fixed difficulty, which only builds trust in a task where human supervision has eroding value. Instead, it's important to stay on a trajectory of increasing task (and goal) difficulty, and to develop a legible body of work in a field (or, to take a riskier bet, in really whatever interests you, because it might just be true that the high dimensionality of how AI represents information somehow returns a person as an expert in a certain subspace, regardless of how the human enterprise is currently organised), such that one continues to be trusted with the rising tide.
Having a public body of work is also what makes otherwise "economically irrational" pursuits of task competency, beyond the customer's level of discernment, rational: artifacts showing "the making of" become a strong trust signal for the capabilities of the person. This effect, however, is still upper-bounded by the expected uplift of human supervision on that task (e.g. being in the top 10% at household cleaning may become less valuable over time with respect to household automation), so the sustainable resolution is to produce that body of work in fields where the goalposts and desires continue to evolve over time rather than stagnate (e.g. visual development), and/or fields driven in large part by competitive pressure to find an edge (e.g. marketing, trading).
Inspired by the recent launch of GPT-6 Astra and Bad Decisions Studio's video impression of it.