Bits and Bobs 3/9/2026

Anthropic released this fascinating chart that shows a large gap between what AI could automate and where it is actually being used today. The biggest adoption so far is in software and quantitative fields because the work is digital, rule-based, and mistakes are cheap.

The next wave will not necessarily come from where AI is most capable, but from where institutional friction starts to break. Legal, healthcare, engineering, and education all show high theoretical coverage but low real usage. That gap signals where the next generation of AI-native companies will emerge.

Physical-world jobs remain relatively protected because they depend on robotics, not just reasoning. So near-term disruption will concentrate in knowledge work before physical labor.

The broader pattern: digital intelligence work → structured knowledge work → physical work

AI capability is no longer the primary constraint. Trust, regulation, and workflow integration are. In Enterprise, I'd focus on where AI capability is high + workflow pain is high + outsourcing already exists.

The earliest openings are not health or finance. Those may be bigger markets later, but they are harder to enter because trust is lower, mistakes matter more, and users want more control.

The real early wedge is in high-frequency, low-stakes, preference-rich decisions where feedback is fast and people already outsource judgment to algorithms, friends, and influencers, like entertainment, media, food, lightweight shopping, and parts of travel. These matter not because they are the biggest markets, but because they generate the richest signal about the person.

In personal intelligence, the winning path is where there is:

So the strategy is not just to build a better media-specific layer. It is to use repeated, taste-driven decisions to build a cross-domain model of the person, then move up-stack from information assistance → decision assistance → life automation.

The wedge is not the final market. It is the path to owning the memory layer that future personal computing depends on.

The emergence of personal intelligence follows a sequence, but its value accrues in a stack.

Layer 6: Personal autopilots (agents that act on your behalf)

Layer 5: Insight & reasoning layer (queries, micro-apps, analysis)

Layer 4: Personal context graph (taste model, behavior history)

Layer 3: Identity & permissions layer (ownership, access, sharing)

Layer 2: Data ingestion layer (APIs, imports, manual input)

Layer 1: Source platforms (Spotify, Netflix, Amazon, etc.)