Anthropic’s Magnum Opus: Anthropic released Claude Opus 4.6 with a beta 1M-token context window, 128k-token outputs, enhanced agentic coding and debugging, achieved top benchmark scores versus GPT-5.2 and Opus 4.5, and token pricing at $5/$25 per million. - Anthropic

Claude Code offers experimental agent teams, disabled by default and enabled with CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, which spawns a lead plus multiple overcaffeinated teammate agents in a virtual sweatshop that will max out your token budget in short order. - Claude

For my next trick: casually building Linux in Rust like it’s a weekend project: Anthropic researcher Nicholas Carlini used 16 parallel Claude instances to build a 100,000-line Rust C compiler that builds Linux 6.9 on x86, ARM and RISC‑V, passes ~99% of several test suites, and cost just under $20,000. - Anthropic

Claude plugins unplug market: Anthropic once looked like an also-ran in the chaotic race for AI supremacy. Oops, this week, its industry-leading products crashed the stock market. - The Wall Street Journal

Anthropic released free open-source plug-ins this week, triggering a software-stock sell-off that pushed the S&P 500 into negative year-to-date territory while Amazon, Alphabet and Meta outlined roughly $200B, $185B and $135B AI capital plans. - The New York Times

Goldman Sachs has spent six months co-developing autonomous AI agents with Anthropic's Claude, embedding Anthropic engineers, to automate trade accounting and client vetting/onboarding, which CIO Marco Argenti calls a "digital co-worker" and expects to launch soon. - CNBC

In other news, MBA’s can’t find jobs - WSJ

Anthropic released a new AI model designed to perform financial research. - Bloomberg

Anthropic's Claude Code saw rapid holiday adoption, Caliber reported a 13-point rise in word-of-mouth exposure from Dec. 29 to Jan. 26, Claude Code exceeded $1 billion revenue in November 2025, and Anthropic released Opus 4.6. - The Verge

We rebuilt your unicorn for $15: CNBC used Anthropic's Claude Code to build a working Monday.com-like prototype with multiple boards, calendar and email integration in under an hour for about $5–$15 in compute credits, and investors warned some software tools are vulnerable. - CNBC

Anthropic's efficiency stems partly from a mission-focused culture that reduces the organizational friction that slows many Big Tech companies. - Bloomberg

OpenAI releases GPT-5.3-Codex - OpenAI

Single pane of glass to herd all your robot cats nuking prod: OpenAI launched Frontier, an enterprise platform for autonomous, agent-based AI built on open standards that hosts in-house or third-party agents and counts early adopters Uber, State Farm, Thermo Fisher and Intuit. - Business Insider

OpenAI launches Frontier, a platform to manage all your AI agents. - The Verge

AI threatens Wall Street's revenue from financial and legal data sales. - The Wall Street Journalg

Long-running AI agents have emerged to run continuously and handle extended, multi-step tasks over time. - The Wall Street Journal

Big Tech companies are undertaking a combined $660 billion AI-related spending spree, while stocks tumbled despite strong earnings, reigniting investor and analyst fears of an AI-driven investment bubble. - Financial Times

Amazon will invest $200 billion, mainly in 2026, to double AWS datacenter capacity by the end of 2027 after adding 3.9 GW in the past year, while expanding its Trainium and Graviton chips to support large AI clusters. - The Register

TSMC will manufacture advanced 3-nanometer semiconductors at its second Kumamoto, Japan plant (under construction) for AI, smartphones, robotics and autonomous driving, and plans to raise 2026 capital spending to $52–56 billion from $40 billion. - ABC News

Long dark winter of the SaaS: Michelle Miller of AlixPartners says AI has driven a plunge in software valuations, with over 30% of tech workflows using AI, about 75% of proofs-of-concept expected to fail, and predicts 2026 consolidation as M&A rises 30-40% YoY. - Business Insider

Welcome to the agent Thunderdome: GitHub is adding Anthropic's Claude and OpenAI's Codex as selectable AI coding agents in Agent HQ across GitHub, GitHub Mobile and Visual Studio Code in a public preview for Copilot Pro Plus and Copilot Enterprise subscribers. - The Verge

Event horizon: SpaceX and xAI fast-tracked their $1.25 trillion merger after SpaceX achieved a technical breakthrough in orbital data centers last fall. - The Wall Street Journal

I’m sure this is not a totally made-up breakthrough to justify a big IPO, and there won’t be a huge crash making Musk and Trump supervillains.

GPU Starships and Chill: Elon Musk said concentrated AI compute will move to space within about 36 months, targeting terawatt-scale GPUs over years supported by SpaceX/Tesla solar production, tens of thousands of Starship launches, and large-scale chip manufacturing. - The Potential Surface

Gerber: X was out of money. Merged with xAI. xAI out of money merge with spaceX. SpaceX out of money. Merge with .... tesla. When they are all out of money.... $tsla - X (formerly Twitter)

Human responsibility over killer AI weapons? US, China, other countries say 'Pass.' - Reuters

Waymo planned to launch its robotaxi service in Washington, D.C., this year, but D.C.'s local regulatory requirements have created hurdles that could delay the rollout, increase compliance costs, and constrain near-term U.S. expansion and revenue. - WIRED

An AI Incident Database analysis finds deepfake fraud has become 'industrial', cataloguing over a dozen impersonation-for-profit cases and noting a Singapore victim lost nearly $500,000 while UK consumers lost an estimated £9.4bn in the nine months to November 2025. - the Guardian

Reports from TollBit and Akamai show training-related and live-data AI bot traffic rose since last July, prompting publishers including Condé Nast to deploy stricter defenses and sue AI firms as experts warn an escalating arms race could reshape the web. - WIRED

OpenClaw's OpenDoor problem is so bad that installing malware yourself might save time - The Decoder

“How about, you give me half the money And then we go out back, I kick you in the nuts and call it a day. Vegas Vacation (1997)

Thirty-three employees of state rail operator CFR Călători face trial for allegedly blocking sleeper/couchette seats to resell, and court filings say at least two consulted ChatGPT for legal advice amid a 700-plus-page prosecution alleging bribery, abuse of office and fraud. - The Register

“Is you asking AI about a mother****ing criminal conspiracy?” - Stringer Bell

Large-scale data signals converging with anecdotal productivity gains suggest AI agents are reaching a 'takeoff' moment. - Financial Times

EY surveyed 15,000 employees across 29 countries and found 88% use AI daily but only 5% are any good and qualify as advanced users, who gain roughly 1.5 extra days of productivity per week. - Inc.

PwC on Thursday launched the Learning Collective, a firmwide training initiative focused on 30 key skills (15 AI-centric, 15 human-centric), using AI agents, adding a tech engineering career track and cutting new-associate start locations from about 60 to 13. - Business Insider

Mitchell Hashimoto's AI adoption playbook recommends verifying coding agents by manually reproducing tasks, scheduling 30-minute end-of-day agents, and outsourcing routine work to boost developer productivity and automate repetitive coding, maintenance, scaffolding, and background progress. - Mitchell Hashimoto

AI is transforming test automation with ML-driven test generation, self-healing UI tests, predictive bug detection and autonomous QA pipelines, improving coverage and speeding releases while raising governance needs; experts predict generative AI will be adopted by end of 2026. - Hacker Noon

BrainIAC, a self‑supervised brain‑imaging model pretrained on nearly 49,000 MRIs and validated on 48,965 scans across seven tasks, estimates brain age, predicts dementia risk and brain‑cancer outcomes, and outperformed three task‑specific AI frameworks in a Nature Neuroscience study. - Harvard Gazette

Apple nerfed Project Mulberry, shifted health oversight to Eddy Cue after COO Jeff Williams' retirement, and will microdose AI health features like gait analysis, an AI health chatbot, and expanded Siri in iOS 27. - 9to5Mac

Axiom says its AxiomProver AI produced solutions to four previously unsolved algebraic-geometry problems involving differentials and resolved a conjecture posed by Dawei Chen and Quentin Gendron five years earlier. - WIRED

Tamagotchi singularity: AI-powered virtual pets in China provide conversational companionship to many young people and are being deployed by major tech companies for engagement, monetization and conversational data collection, raising privacy and regulatory concerns. - The Washington Post

Emotional affair with your screen? The New Family Intervention: We Need to Talk About Your AI Habit - The Wall Street Journal

How to maintain mental health while acquiring AI skills. - The Wall Street Journal

Human intelligence is social, embodied and collective, while large language models—trained on about 80% of online content in ten languages—lack embodiment, intentions and ethical reasoning and are useful tools, not superior minds. - The Conversation

AI helps funders evaluate more ideas more fairly. - Fast Company

Just kidding, it pattern matches on academic and Big Tech pedigree and then ghosts everyone else.

On this day, a “hellish broth of machine-driven AI slop and bad human choices.” - Ars Technica

Darren Aronofsky YouTube Series Backlash Shows World Is Not Ready For 'Auteur AI' - Deadline

Nobody asked for Nano Banana Ken Burns.

Follow the latest AI headlines via druce.ai on Bluesky