Article archive
Published news and blog articles, organized by category. Browse older coverage by month or search for a topic. Undated blog guides appear after dated news.
48 articles · Newest first
The Collapse in the Price of Inference: Why Intelligence Getting Cheap May Matter More Than Intelligence Getting Good
Anthropic and OpenAI are racing to ship more capable models at lower per-token prices. The economic story is not the capability curve — it is the cost curve, and who captures the surplus as the marginal price of machine reasoning falls.
Training the Machine That Replaces You: Toyota, Demonstration Data, and the Economics of Self-Substituting Labor
Toyota is having human workers teach humanoid robots their jobs. The deep question is not robotics but labor economics: when a worker's demonstration data becomes capital that substitutes for their own labor, who should be compensated, and…
Crime at Machine Scale: The Microsoft Disruption and the Industrial Economics of AI-Assisted Account Compromise
Microsoft's disruption of an AI-assisted phishing platform reportedly tied to roughly 12,000 account compromises is a case study in offense-defense economics: when generative models collapse the marginal cost of a convincing lure, the…
The Margin Migration: Open-Weight AI and the Commoditization of Intelligence
Xiaomi's MiMo models are the latest entry in an increasingly competitive Chinese open-weight ecosystem. N43 examines whether the commoditization of foundation models transfers economic value from the labs that train them to the companies…
A Million Tokens of Working Memory: Substitution, Cost, and What Long Context Actually Solves
DeepSeek's million-token-context architecture prompts a deceptively simple question: do enormous context windows substitute for persistent memory, retrieval systems, and even further training? N43 examines the attention economics of long…
The Network That Tunes Itself: AI Control Comes to the Radio Access Layer
Samsung reports substantial downlink improvements from AI-assisted virtual radio-access networks. N43 examines the control-theory transformation behind the number — cellular infrastructure shifting from human-planned systems to closed-loop…
From Stochastic Parrots to Stabilization: The Robot That Learned to Fly 450 Percent Faster
MIT researchers report an AI controller making a tiny flying robot roughly 450 percent faster. N43 examines what the result actually demonstrates — real-time physical intelligence as a research frontier distinct from language models — and…
Cloud Wetware: Biological Computing's First Commercial Beachhead and the Question of a Third Paradigm
A company is making biological neural-system-inspired computing available through AWS. N43 examines what organoid-based systems actually demonstrate, why cloud availability is an institutional milestone rather than a technical one, and…
Light Between the Chips: Nanolasers, Optical Interconnects, and the Energy Bill of Computing
Researchers are developing tiny lasers that could replace some electrical chip interconnects with light, potentially cutting computing's electricity appetite. N43 examines interconnect energy as a structural constraint on AI-era compute,…
One System Now: AI, Electricity, Inflation, and the New Geometry of Geopolitics
AI capital expenditure, power infrastructure, inflation, interest rates, and great-power supply-chain competition now form a single feedback system — this capstone analysis maps its loops, names its stabilizing and destabilizing arms, and…
The Hotline Problem: Designing AI Crisis Communications Between Washington and Beijing
Washington and Beijing are exploring an AI crisis-communications mechanism. Not the proposal itself — the design problem: what would a working AI incident channel actually require, and why is it harder than the nuclear hotline?
Does the Future Internet Need Identity Verification for AI Agents?
Nearly half of all internet traffic is already automated, and agents are the fastest-growing slice. The industry is converging on cryptographic agent credentials — TLS for machines — before unauthenticated agents get treated as hostile by…
Could AI Safety Become an Insurance Product?
Cyber insurers learned to price hacking only after catastrophes gave them loss data; AI liability has no actuarial history, correlated failure modes and models that change under the policy. Lloyd's and specialist underwriters are writing…
Could AI Agents Make APIs More Important Than Websites?
Agents consume APIs, not pixels — and protocols like MCP plus agent-payment schemes are turning machine-readable access into a revenue line. The website is becoming the human fallback while the API becomes the storefront.
Accenture Is Putting $2 Billion Into Anthropic AI Evaluation — Is AI Testing Becoming Big Business?
Accenture is committing roughly $2 billion to train 50,000 people and build an AI-evaluation practice around Anthropic's models. The deal signals that testing frontier AI — like auditing financial statements — is turning into a…
The Death of the Dashboard? Why Executives May Soon Ask AI Instead of Opening Business-Intelligence Software.
Every major BI vendor now ships a natural-language copilot, and a generation of executives is asking questions instead of opening dashboards. The dashboard is becoming legacy UX — the question is what replaces its governance.
AI Spending Isn't Slowing Down Despite Safety Warnings — Who Is Financing the Next Compute Boom?
The four hyperscalers are on track to spend well over $300 billion this year on capital projects, much of it AI compute, and the debt is increasingly coming from private credit. The safety warnings are on the record; so is the money — and…
Could AI Agents Replace Traditional Enterprise Software?
Enterprise software has always charged by the seat. Agents do not sit in seats — they call APIs, and the biggest SaaS vendors are racing to sell them before agents commoditize the dashboards humans used to click.
Could an International AI Incident Database Work Like Aviation Accident Reporting?
Aviation made crash investigation independent, blame-free and public — and flying became the safest form of travel. AI has a volunteer database of 3,000+ incidents, no investigators and no mandate; the question is whether the NTSB model…
Could Autonomous AI Agents Eventually Need Something Like Digital Passports?
Autonomous AI agents are increasingly acting on behalf of institutions across networks that were designed for humans with credentials. The idea of a machine-readable passport for software agents is moving from thought experiment toward…
Will Companies Eventually Need 'AI Liability Insurance' Before Deploying Autonomous Agents?
Insurers have begun writing dedicated AI liability lines just as the EU's high-risk regime starts to bite. The open question is whether proof of coverage becomes a deployment gate — and whether anyone can price it.
U.S. and China Consider an AI 'Emergency Hotline' — Could AI Incidents Become the New Nuclear-Crisis Diplomacy?
U.S. officials have proposed an AI incident-notification mechanism at the latest senior-level talks with Beijing, borrowing the logic of the Cold War hotlines. The idea is easy to state and hard to build, because an AI emergency may move…
Could AI Agents Eventually Negotiate Prices With Other AI Agents?
The tech exists today: agentic-commerce protocols from PayPal, Google and OpenAI let a shopping agent query a merchant bot and pay without a human. The policy question is older than software — will machine-to-machine haggling restore the…
AI Coding Agents Are Getting Better — Does Software Become Cheaper to Produce?
The agents write most of the new code at many companies now, and the studies converge: per-task output up sharply, code volume way up — but defect rates and maintenance burden rise with it. The policy question is whether software cost is…
Also explore blog articles and guides or search all DutyStation pages.