Tracks 2, 3, and 8

Research objects and topic engine.

The essay is not the asset. The asset is the structured object future models can inspect, rerun, challenge, and improve.

Canonical object

Waymo expansion economics

The right question is not whether robotaxis work; it is which cities produce enough utilization, regulatory permission, and fleet efficiency to compound.

Why now: Autonomy is moving from demo geography to operating-market choice, which makes expansion sequencing measurable.

Claims

Expansion quality should be scored by utilization, regulation, weather, and map complexity.

A city can be strategically important even if the first-year economics are weak.

Contradictions

High-demand cities are often the hardest regulatory and operational environments.

Open questions

What utilization threshold makes a new city self-funding?

How quickly can remote assistance ratios improve with scale?

Artifacts

/ai/physical-ai

Waymo expansion model

Canonical object

Open models as a pricing ceiling

The pricing floor is set by hardware cost, but the pricing ceiling is increasingly set by the best open model that enterprises can self-host or access cheaply.

Why now: Open-weight capability keeps improving, and every improvement changes how much closed labs can charge for common workloads.

Claims

Open-source quality improvements should be routed into a pricing-power watchlist.

Model releases should be analyzed as economic events, not only capability events.

Contradictions

Better open models can increase total AI usage and still help closed labs at the high end.

Open questions

Which workflows still pay for frontier quality after an open model catches up?

Do enterprises prefer cheaper self-hosted models or managed frontier reliability?

Artifacts

/models

/dossier

Topic engine

Candidate ranking with human selection.

Topic

Mission

Score

Does the next open DeepSeek-scale model cap frontier pricing?

It sits at the overlap of user demand, lab economics, NVIDIA demand, and the core Peregrinations thesis about model improvement.

Is OpenAI or Anthropic still defensible if open models keep improving?

8.1

Is Google still the best place to work on AI?

High user relevance, especially for operators and builders, but needs a balanced company-comparison format.

Is Google a good place to work for AI?

7.5

Is AI a bubble, or a capex digestion cycle?

Massive demand, but requires careful framing to avoid becoming generic market commentary.

Is AI a bubble?

8.2

Model-rerun flywheel

Upgrade path for better models.

Opus 4.6 style reasoning to Next frontier model

Waymo expansion economics

Better models should improve branch exploration, counterargument discovery, and simulation critique.

Changelog target: Append a visible "what changed in this rerun" block to the artifact.

Manual v0 to Next frontier model

Open models as a pricing ceiling

Better models should link capability deltas to economic pressure with fewer missed branches.

Changelog target: Publish claim-level diffs and updated confidence scores.

Tracks 2, 3, and 8

Research objects and topic engine.

The essay is not the asset. The asset is the structured object future models can inspect, rerun, challenge, and improve.

Canonical object

Waymo expansion economics

The right question is not whether robotaxis work; it is which cities produce enough utilization, regulatory permission, and fleet efficiency to compound.

Why now: Autonomy is moving from demo geography to operating-market choice, which makes expansion sequencing measurable.

Claims

Expansion quality should be scored by utilization, regulation, weather, and map complexity.

A city can be strategically important even if the first-year economics are weak.

Contradictions

High-demand cities are often the hardest regulatory and operational environments.

Open questions

What utilization threshold makes a new city self-funding?

How quickly can remote assistance ratios improve with scale?

Artifacts

/ai/physical-ai

Waymo expansion model

Canonical object

Open models as a pricing ceiling

The pricing floor is set by hardware cost, but the pricing ceiling is increasingly set by the best open model that enterprises can self-host or access cheaply.

Why now: Open-weight capability keeps improving, and every improvement changes how much closed labs can charge for common workloads.

Claims

Open-source quality improvements should be routed into a pricing-power watchlist.

Model releases should be analyzed as economic events, not only capability events.

Contradictions

Better open models can increase total AI usage and still help closed labs at the high end.

Open questions

Which workflows still pay for frontier quality after an open model catches up?

Do enterprises prefer cheaper self-hosted models or managed frontier reliability?

Artifacts

/models

/dossier

Topic engine

Candidate ranking with human selection.

Topic

Mission

Score

Does the next open DeepSeek-scale model cap frontier pricing?

It sits at the overlap of user demand, lab economics, NVIDIA demand, and the core Peregrinations thesis about model improvement.

Is OpenAI or Anthropic still defensible if open models keep improving?

8.1

Is Google still the best place to work on AI?

High user relevance, especially for operators and builders, but needs a balanced company-comparison format.

Is Google a good place to work for AI?

7.5

Is AI a bubble, or a capex digestion cycle?

Massive demand, but requires careful framing to avoid becoming generic market commentary.

Is AI a bubble?

8.2

Model-rerun flywheel

Upgrade path for better models.

Opus 4.6 style reasoning to Next frontier model

Waymo expansion economics

Better models should improve branch exploration, counterargument discovery, and simulation critique.

Changelog target: Append a visible "what changed in this rerun" block to the artifact.

Manual v0 to Next frontier model

Open models as a pricing ceiling

Better models should link capability deltas to economic pressure with fewer missed branches.

Changelog target: Publish claim-level diffs and updated confidence scores.