Articles by Richard L. Wells
-
OpenAI Codex Becomes Desktop Agent: Controls Mac Apps, Watches Screen, Runs on Mobile
Over a six-week stretch in spring 2026, OpenAI rebuilt what its Codex product actually is. On April 16, the company released a major Codex update titled "Codex for (almost) everything,"
-
Anthropic and Microsoft Negotiate Maia 200 Chip Deal: Claude Could Become Custom Silicon's First Frontier Test
Anthropic and Microsoft are in early talks for Anthropic to rent Azure servers running Microsoft's custom Maia 200 AI accelerator — a deal that would make Claude the first frontier model to validate the chip externally and give Anthropic a fourth custom silicon option to reduce per-token inference costs.
-
Anthropic Moves Closer to Public Claude Mythos Release: 10,000 Critical Bugs Found First
Claude Mythos vulnerability research has surfaced over 10,000 critical flaws across 50 organizations in one month. Anthropic signals a future general release, Claude Security launches in public beta for Enterprise, and open source software security vulnerabilities are piling up faster than maintainers can patch them.
-
Vibe Coding for Non-Developers: 63% of Users Now Have No Coding Background, Breaches Follow
Vibe coding for non-developers now accounts for 63 percent of all users, with writers, students, and investors shipping apps in hours. But the 2026 Moltbook security breach — which exposed 1.5 million API keys in a vibe-coded platform three days after launch — shows exactly where AI app building without coding becomes dangerous.
-
AI Agent Business Models Split Four Ways: Open-Source Infrastructure, Token Distribution, SaaS, Acquisition
AI agent business models in 2026 diverge sharply: OpenClaw and Hermes Agent win GitHub stars and inference tokens, Genspark crossed $200 million in annual revenue, and Manus saw its $2 billion Meta acquisition blocked by Beijing — four projects, four paradigms, none competing on the same metric. Here is what that means for developers, investors, and builders.
-
Google DeepMind Picks EVE Online as Next AI Frontier: Game That Never Ends
Google DeepMind EVE Online AI research partnership pairs the lab's most ambitious agent work with a 23-year-old player-shaped universe that never resets — targeting the long-horizon planning and continual learning gaps that no existing AI benchmark can expose.
-
AI Agents for Solo Founders: Genspark, Manus, Devin Raise Billions Before Proving It Works
AI agents for solo founders are the fastest-growing startup category of 2026 — but with seven months left on Dario Amodei's billion-dollar one-person company prediction, no platform has yet published the task-success or retention data that would confirm agentic workspace tools deliver what they promise.
-
Fei-Fei Li's ESI-Bench Catches Frontier AI Failing 3D Space: Seeing and Acting Diverge
A Stanford-led research team that includes Fei-Fei Li published a benchmark that documents a specific, measurable failure: when forced to actively move through a 3D environment rather than interpret a pre-composed image, state-of-the-art systems consistently choose the wrong actions, skip viewpoints that would correct their mistakes, and commit to wrong answers with high confidence.
-
NextEra's $67B Dominion Bet Is Not About Chips: It's About Electrons
When NextEra Energy announced on May 18 that it would acquire Dominion Energy in an all-stock deal valued at roughly $67 billion, the headline number obscured what the deal was actually buying.
-
China's AI APIs Cost 90% Less and Run Significantly Slower: The Tradeoff Most Builders Miss
China's largest AI developers are no longer selling scale alone. Parameter counts still matter, but the more important launch metric is increasingly what it costs to run a model after it has been built.
-
WordPress 7.0 Ships AI Agent Infrastructure: API Key Theft Risk Surfaces on Launch Day
WordPress 7.0 "Armstrong," released May 20, 2026, arrived without the real-time collaborative editing feature that had been its stated centerpiece for months — and within two days of launch, a security researcher was warning that the AI infrastructure it shipped in place of that feature has given hackers a new category of target: paid API credentials worth thousands of dollars now stored in WordPress admin dashboards across 43% of the web.
-
Newsom Signs AI Workforce Order: Studies Required, No Worker Protections Yet
California Governor Gavin Newsom on Thursday signed what his office described as a "first-of-its-kind" executive order directing state agencies to prepare workers, small businesses and communities for the workforce disruption widespread artificial intelligence adoption is expected to bring. It is, according to Newsom's office, the first executive order of its kind, and one of the clearest state-level moves so far to treat AI-related workforce disruption as a distinct policy problem.
-
ChatGPT Finances Centralizes Your Net Worth Data: One Account Breach Exposes Everything, Security Experts Warn
OpenAI began letting ChatGPT Pro subscribers in the United States link bank accounts, investment portfolios, and credit cards directly to the chatbot on May 15 — two days after a federal class-action suit was filed alleging that OpenAI had already been secretly routing users' private conversations to Meta and Google through tracking code embedded in the ChatGPT website.
-
AI Took About 80% of Global Venture Funding Last Quarter: Thin Wrapper Apps Were Not the Winners
The largest U.S. technology IPO since Uber's 2019 debut closed its first trading day last Thursday at a 68% gain, and the company behind it — AI chipmaker Cerebras Systems — had almost nothing to do with large language models. The May 14 Cerebras debut, which raised $5.55 billion at a price of $185 per share, is the exclamation point on a Q1 2026 funding landscape that has already redrawn the map.
-
Qwen3.7-Max Wrote Its Own Chip's Software in 35-Hour Run: Alibaba's Full-Stack Bet
Alibaba's Qwen team unveiled Qwen3.7-Max at the Alibaba Cloud Summit in Hangzhou this week, positioning the model as its flagship "agent era" release — the company's most advanced foundation model for long-horizon, tool-heavy autonomous tasks. The launch claims are striking. They are also vendor claims, and the deeper news is not the model in isolation but the full Chinese AI stack it was built to showcase.
-
OpenAI Model Cracks 80-Year Erdős Conjecture, Verified by Its Harshest Previous Critic
On May 20, OpenAI said an internal reasoning model had produced a counterexample to Paul Erdős's 1946 unit distance conjecture — a result now presented in a human-verified companion paper by nine external mathematicians, including some of the same researchers who publicly corrected OpenAI's last major math claim seven months ago. OpenAI describes the work as the first time an AI system has autonomously solved a prominent open problem central to a subfield of mathematics. That detail — the verifying signatures, not just the claim — is what makes this different from October.
-
Utah Stratos Project Crystallizes AI Buildout's Dual Crisis: Transformer Shortages, Community Vetoes
The American AI infrastructure buildout is colliding with two constraints at the same time. The first, well-documented, is the electrical hardware that physically connects new data centers to power — transformers, switchgear and grid interconnects whose multi-year backlogs have begun erasing huge fractions of the announced 2026 pipeline.
-
AI Coding's Discipline Turn: Three Open-Source Frameworks Superpowers, gstack, GSD Outpace Model Upgrades
Six months after Anthropic shipped a plugin system for Claude Code, the most-installed extensions on the platform have nothing to do with new capabilities. They are opinionated rulebooks — and the three that have broken into mass adoption this spring share a thesis that has since spread to 14 different AI coding agents: the bottleneck in AI-assisted software development is not model intelligence.
-
Claude Code 98% Harness: Four Competing Teams Built Same Agent Harness, Pointing to Real AI Moat
A research team at Mohamed bin Zayed University of Artificial Intelligence published a finding that has gained traction in engineering circles for reasons that go beyond its headline number. By analyzing Claude Code v2.1.88, the four-author team at VILA-Lab dissected 1,884 files and approximately 512,000 lines of code.
-
TSMC Arizona Fab Posts $514M Year-One Profit: Q1 2026 Earnings Surpass Full 2025 Figure
The back-to-back figures mark the first time a data point has directly challenged the long-dominant expert view that advanced chipmaking outside Taiwan cannot be economically viable.
-
Cursor Composer 2.5 Matches Claude Opus 4.7 on Coding Benchmarks at One-Tenth Cost
Cursor released Composer 2.5 on May 18, its most capable in-house coding model, and built the launch around a single provocative claim: frontier-level agentic coding at roughly one-tenth the price of Claude Opus 4.7 or GPT-5.5. The headline is largely defensible. It is also selective, rests heavily on Cursor's own benchmarks, and carries caveats the marketing does not lead with.
