Articles & Research

Technical writing from NSS — research notes, analysis, and arguments we found worth making in public.

Date	Category	Title
November 4, 2026	RESEARCH	A Better INT4 Qwopus We Decided Not to ShipNEW An improved INT4 build of a 27B coder model got closer to full precision on every quality measure — and still did not ship, because it was slower and broke the fast drafter most people use.
November 2, 2026	RESEARCH	Baking a 27B Coder Model Down to INT4 on One Arc Pro B65 Quantizing a 55 GB coder model to fit one 32 GB Intel card, with quality bars written down before measuring — and the tool-calling regression those bars caught before release.
September 1, 2026	RESEARCH	We Shaped the Report Template Around One Benchmark A data-only publishing pipeline means the template decides what a result can say. Why Launch80 tuned theirs for BetterBench, and what that costs every other kind of result.
July 24, 2026	RESEARCH	One Model, Every Local Stack Every local inference server and four coding agents, pointed at the same 35B model on one Mac. Several could not load it; the rest differed more than their marketing suggests.
July 23, 2026	RESEARCH	Two Optimizations That Were Never Allowed to Run Together A stock inference server made them choose between fast first responses and fast streaming. Forking it removed the choice — and a re-run exposed a benchmark that had flattered the result.
February 27, 2026	AI	Claude Cowork: Seven Use Cases, Examined Anthropic's Claude Cowork pushes AI into every department — financial analysis, legal review, HR onboarding, sales automation. A look at which of those use cases hold up in practice.
February 26, 2026	AI	Mapping Where Agentic AI Actually Pays Off Twelve workflow opportunities across three departments and four priority tiers — a framework for working out where autonomous agents earn their keep, and where they do not.
February 25, 2026	AI	Agentic AI and CAPEX: Running the Numbers Legacy workflows lock capital in depreciating assets. Autonomous agents are pitched as the way out, with 30–60% CAPEX reductions claimed. Here is where those numbers come from and what they assume.
November 19, 2025	INFRASTRUCTURE	Own or Rent: AI Infrastructure Economics Over Five Years The total economics of AI infrastructure over five years produce a less obvious picture than either vendor claims. A breakdown of the real costs — cloud versus owned hardware.
November 19, 2025	SECURITY	Four Regulated Industries Rethinking Cloud AI Healthcare systems, financial institutions, defense contractors and law firms share one requirement: certainty about where data sits and who can reach it. How each is evaluating its AI posture.
November 18, 2025	INFRASTRUCTURE	Vendor Lock-In When You Build on Someone Else's Models Everyone built on the Google Maps API, and then the pricing changed overnight. The same dynamics are forming in enterprise AI. How to measure vendor dependency before it becomes structural.
November 3, 2025	SECURITY	Where Your Data Actually Goes on a Cloud AI Request A single AI query can touch infrastructure across several jurisdictions. Tracing that path — and being able to demonstrate it — is a prerequisite for handling sensitive data at all.
October 15, 2025	SECURITY	Is Your AI Vendor Training on Your Data? Every interaction with a hosted model leaves a trail of proprietary knowledge somewhere. How to establish what your vendor is actually doing with it, and what your options are.
September 26, 2025	SECURITY	The Anthropic Copyright Settlement and Training-Data Liability Anthropic's copyright settlement raises real questions about liability for training data. What the ruling establishes, and how to think about your own exposure.