Research & Deep Dives
Structured notes on industries, companies, papers, reports, and technical shifts worth understanding beyond the daily feed.
Start Here
The warehouse-versus-lakehouse framing is becoming less useful. This market map compares Snowflake, Databricks, ClickHouse, and open table formats through the architectural choices and control points that increasingly determine where analytics value accrues.
Company Deep Dive turbopuffer and the Cost Curve of Search — Company Deep Diveturbopuffer is a bet on infrastructure economics: object storage can reshape the cost curve of search when latency and architecture are designed around it. This company deep dive examines the product, technical tradeoffs, market position, and sources of advantage.
Research Explainer When the Odds Cannot Be CalculatedWhen probabilities and outcomes are impossible to predict, most investors flee. Richard Zeckhauser argues this domain of profound ignorance is precisely where the greatest returns are found—if you have the right framework.
Industry Deep Dives
The warehouse-versus-lakehouse framing is becoming less useful. This market map compares Snowflake, Databricks, ClickHouse, and open table formats through the architectural choices and control points that increasingly determine where analytics value accrues.
AI infrastructure ultimately depends on power availability, grid access, cooling, and data-center construction. This deep dive maps the energy value chain, its major constraints, and the companies positioned to benefit from growing compute demand.
The most capital-intensive pressure valve in AI infrastructure sits between constrained GPU supply and rising enterprise demand. This deep dive maps neocloud economics, customer segments, and the strategic position of providers between hyperscalers, model labs, and dedicated-compute buyers.
Moving enterprise AI from advice into execution requires a new coordination layer. This deep dive maps agentic workflow vendors, architectures, and control points, showing how software must manage models, tools, permissions, state, and human oversight to perform real work reliably.
Every business purchase passes through a control layer linking employees to accounting systems. This deep dive maps finance operations workflows and data models, showing how spending is requested, approved, recorded, reconciled, and made visible without sacrificing governance.
Production AI needs more than prompt monitoring. This deep dive maps the observability and evaluation stack required to trace model behavior, measure quality, diagnose failures, and improve systems whose outputs change across models, data, tools, and workflows.
Company Deep Dives
turbopuffer is a bet on infrastructure economics: object storage can reshape the cost curve of search when latency and architecture are designed around it. This company deep dive examines the product, technical tradeoffs, market position, and sources of advantage.
Research Explainers
When probabilities and outcomes are impossible to predict, most investors flee. Richard Zeckhauser argues this domain of profound ignorance is precisely where the greatest returns are found—if you have the right framework.
Jeremy Allaire argues that AI agents need economic infrastructure for payments, identity, and coordination. This explainer separates the functions that genuinely benefit from blockchains from those better handled by conventional systems, clarifying where onchain architecture earns its complexity.
OpenRouter's 100 trillion token usage study suggests AI demand is shifting from simple text generation toward reasoning, tools, code, and context-heavy workflows. This explainer maps what that change means for model providers, infrastructure, and the economics of serving AI.
AI Futures Project's AI 2040 scenario is less a prediction than a governance proposal. This explainer examines its case for compute verification, research transparency, and international coordination strong enough to slow a race toward concentrated control.
A study of 199 startup founders suggests AI adoption depends as much on managerial willingness to delegate work as on beliefs about model capability. This explainer shows why adoption is a management-design choice, not simply a technology-readiness test.
A field experiment with 515 startups suggests AI creates firm-level gains when teams learn where to reorganize work around it, rather than merely gaining access to tools. This explainer connects adoption outcomes to workflow mapping, experimentation, and organizational change.