Insights

Research papers and tools

AI Demand, Capacity Commitments and Timing — A synthetic rack capacity planning model shows why a commitment's calendar, recovery capacity and cost horizon have to b
Rethinking AI Inference Costs and Pricing — A synthetic inference fixture calculates a per-request cost floor from bounded requests, fixed capacity and failover. Cr
Network capacity is sometimes compute capacity: an accounting framework and decision boundaries for infrastructure investment in large-scale AI training — An AI training cluster can report 89% effective training time while converting only 63% of purchased accelerator capacit
The Host-Side Ceiling on Agentic Inference Capacity — Installed compute has potential but effective capacity is what survives load imbalance, host contention, tool execution,
The Cross-Stack AI Architecture Review Guide — Templates, trigger classes, an eight-stage release lifecycle, and a fifteen-metric worksheet for reviewing the seams of
Where AI Systems Actually Break: At the Interfaces — The flagship made the argument; this is the tool. What actually goes on the page when you write a contract for a seam in
Who Owns the Interfaces Across the AI Stack? — Every consequential seam in a production AI stack needs a named owner, and your org chart is quietly deciding which seam
The AI Stack Is a System — For weeks in 2025, Anthropic's evaluations stayed green while three separate bugs degraded Claude in production. Users n
Ground Truth Is the Moat — The real competitive advantage in autonomous network infrastructure is not the sensors or the models, which anyone can b
The Infrastructure Schism — Every governance framework being built for autonomous infrastructure controls what an agent may do, not whether what it
The Sovereign NOC: A Hybrid Human-Agent Operations Case Study — This case study applies the hybrid human-agent squad blueprint to network operations, showing how NOCs can coordinate hu
The Network as the AI Grid: Edge Inference, Silicon Economics, and Sovereign Infrastructure for the Last Mile — In the generative era, the digital divide is shifting from a data transit problem to an intelligence access bottleneck.
AI Compute Economics: What Spending Alone Cannot Measure — Capital spending, delivered capability, and useful economic output measure different things. A preliminary AI-economy fr
The Two Workloads: Matching AI Infrastructure to Training and Inference — Training, prefill, and decode put different demands on a compute fleet. The capital decision is when specialization impr
Compute With Legal Identity: Sovereignty as an Architecture Constraint — Workload placement depends on more than data location. Legal jurisdiction, operational authority, provider structure, an
The Hardware Bet You Can't Actually Audit — At the scale of modern AI clusters, hardware failure isn't dramatic. It's silent. And most CTOs are approving infrastruc
The AI Workload Hierarchy: A Portfolio Framework for Usable Compute — A capacity plan becomes stronger when it connects workload demand with power, cooling, network, reliability, topology, a
The Maintainer Loop: Keeping Books on the Bookkeeper — An LLM reviewer once invented a paper that nearly landed in my related-work table, and a dead API key once produced an a
A Technical Framework for Frontier AI Capability Evaluations — Frontier AI models evaluated with reduced safety refusals exhibit multi-step reasoning, zero-day discovery, and persiste
Ontology Debt: A Consistency Ledger for LLM World-Models — A model a standard accuracy score would call 96 percent correct insisted a row of untouched dominoes had fallen, then ha
Guardrails Are Not Instructions — An instruction in a system prompt is a suggestion the model is statistically likely to honor, not a rule it must obey, a
Enforcing Runtime Policy: The Architecture of Trust in Clinical AI — Capability is not safety. In high-stakes clinical environments, bridging the gap between raw LLM reasoning and patient s
Audit-After-The-Fact Is Bankrupt. Governance Now Has a Latency Budget — Agents act fifty times faster than humans, but the policy layer most enterprises bought was designed for ticketed review
Measuring Judgment in an AI-Assisted World — Caliber — AI made fluency free, so the scarce asset is now judgment that holds up. Why the signal that told you who could actually
Building Hybrid Human-Agent Squads with Shared OKRs — A practical enterprise blueprint for designing and governing hybrid human-AI agent teams. The paper covers shared OKRs,
World Models and Zero‑Shot Planning: The Imagination Engine for Physical AI — How predictive, action-conditioned simulation enables grounded reasoning and planning—complementing language models with
Frontier AI Models Have a Blind Spot for Ambiguity — New benchmark results show where social cognition breaks—and it's not where you think. Complexity Preservation, the abil
The Quiet Crisis: Why Moving Fast on AI Is Breaking the Pipeline That Builds Your Next Leaders — The first real AI labor shock won't be the layoffs. It will be the juniors who never got the chance to become seniors.
From Pilot Purgatory to the Orchestrated Era: A Manifesto — Most companies aren’t failing at AI. They’re succeeding at the wrong version of it — and the gap is about to become irre
Hybrid Pricing for AI Support in B2B SaaS — Most B2B SaaS companies will get AI support pricing wrong unless they combine a subscription base, an outcome-tied varia
The Resolution Ledger: The Missing Product Behind Outcome-Based Pricing — Outcome pricing sounds simple until a customer asks why you billed them for 3,214 resolutions last month. A resolution l
The Sovereign Enterprise: Rewiring for the Agentic Era — Most organizations are currently stuck in "Pilot Purgatory," deploying chatbots that act as expensive bandaids on legacy
The Architecture of the Forkable Firm: Why the Future of Leadership is 'Org Code' and Vertical Co-Design — The model now dictates the org chart, not the other way around. Conway's Law has reversed, and most firms aren't ready.
Dirty Data to Clarity: How AI Transforms Raw Data into Insight — Transforming raw randomness into structured intelligence. Leveraging AI in data governance.