Rabbit-in-the-Headlights — AI stories that actually matter.
Update February 9, 2026
Welcome to Kernel Weekly, your short, sharp hit of AI news — the breakthroughs, bold moves, and occasional blunders worth sinking your teeth into.
🤖 1) Anthropic drops Opus 4.6 — and triggers a $285 billion “SaaSpocalypse”
Anthropic released Claude Opus 4.6 on February 5 with agent teams, a 1 million token context window, PowerPoint integration, and financial research capabilities. But the bigger story started days earlier: Claude Cowork plugins for legal, finance, and marketing launched January 30, triggering what analysts are calling the “SaaSpocalypse.” Thomson Reuters dropped 16%, Wolters Kluwer fell 10%, and software stocks shed $285 billion in market cap in a single day. The iShares Expanded Tech-Software Sector ETF fell for seven consecutive sessions — a 16% decline not seen since the COVID panic of March 2020. Analysts say the significance is that Anthropic is shifting from model supplier to the application layer and workflow owner. TechCrunch | CNBC | CNN Business
Kernel take: This isn’t just a model update — it’s a strategic land grab. When an AI lab ships plugins that directly replace legal review, financial analysis, and marketing workflows, the SaaS incumbents aren’t competing against a chatbot anymore. They’re competing against a platform. The $285 billion sell-off might be an overreaction, but the direction of travel is clear.
🧠 2) OpenAI fires back with GPT-5.3-Codex — released the exact same day
Minutes after Anthropic’s Opus 4.6 announcement on February 5, OpenAI launched GPT-5.3-Codex — their most capable agentic coding model yet. It merges the coding agent and reasoning model into one unified system, is 25% faster, and sets new highs on SWE-Bench Pro (56.8%) and Terminal-Bench 2.0 (77.3%). Most notably, GPT-5.3-Codex is the first model OpenAI says was “instrumental in creating itself” — early versions debugged their own training, managed deployment, and diagnosed test results. OpenAI classifies it as “High capability” for cybersecurity under their Preparedness Framework, deploying additional mitigations and access controls as a result. OpenAI | TechCrunch | MarkTechPost
Why it matters: The model wars have entered a new phase. It’s no longer about who has the best chatbot — it’s about who ships the best coding agent. A model that helped build itself sounds like science fiction, but it signals where agentic AI is heading: self-improving systems that compress entire engineering workflows into a single tool.
📉 3) Jensen Huang calls software selloff “the most illogical thing in the world”
At the Cisco AI Summit on February 4, Nvidia CEO Jensen Huang pushed back hard against the panic sweeping software stocks. His argument: AI uses software tools, it doesn’t replace them. “Would you use a hammer or invent a new hammer?” he asked. The comments came as the iShares Expanded Tech-Software Sector ETF added another 4% decline, bringing year-to-date losses to 22%. ServiceNow fell 7% (28% YTD), Salesforce dropped 7% (26% YTD), and Intuit tumbled 11% (34% YTD). Despite Huang’s reassurance, the sell-off continued into the following day. Bloomberg | Yahoo Finance | CNBC
Kernel take: Jensen’s analogy is comforting but incomplete. Yes, AI agents use existing tools — but they also eliminate the humans who used to operate those tools. The SaaS model depends on seat-based pricing. When one AI agent replaces ten seats, the economics break whether or not the software survives. The selloff isn’t about tools dying. It’s about business models dying.
🔮 4) Claude Sonnet 5 “Fennec” leaked in Google Vertex AI logs
Developers spotted the identifier claude-sonnet-5@20260203 in Google Vertex AI error logs, following Anthropic’s standard naming convention. The leak sparked intense speculation about an imminent release. Unconfirmed specs suggest Sonnet 5 could cost 50% less than Opus 4.5 while matching its benchmark performance (80.9%+ on SWE-Bench). A 1 million token context window and a “Dev Team” mode for multi-agent collaboration are also rumored. Anthropic has not commented. Dataconomy | DEV Community
Why it matters: If the leaked specs are real, Sonnet 5 at half the price of Opus 4.5 with comparable performance would be a pricing earthquake. The AI cost curve is collapsing faster than anyone predicted. Developers who locked into expensive API plans six months ago are about to feel very differently about their contracts.
🏭 5) TSMC posts record profit — raises 2026 capex to $56 billion
Taiwan Semiconductor delivered another record quarter with Q4 net income of $16.3 billion, a 35% year-over-year jump driven by insatiable AI chip demand. High-performance computing (AI + 5G) now accounts for 55% of sales. TSMC raised its 2026 capital expenditure guidance to $52–56 billion, up from $40.9 billion in 2025, with more than 70% allocated to advanced nodes. CEO C.C. Wei raised the AI accelerator revenue forecast to a mid- to high-50s% compound annual growth rate from 2024 to 2029, up from prior guidance of approximately 50%. CNBC | Fintool | Bloomberg
Kernel take: TSMC’s capex is the best leading indicator in AI. When the company that makes everyone’s chips raises spending by 40%, it’s not speculation — it’s pull-through demand from every major cloud provider. The AI infrastructure boom has legs.
🔍 6) Google DeepMind goes on acquisition spree — three deals in one week
Google DeepMind made three strategic moves in quick succession. First, Google acquired Common Sense Machines, a startup specializing in 2D-to-3D AI models (closed January 24). Second, DeepMind hired the core team from Hume AI, known for emotion and voice analysis — CEO Alan Cowen and approximately seven engineers joined, while Hume continues operating independently with a licensing deal. Third, Google partnered with Sakana AI, a Japanese lab focused on scientific AI research. All three moves strengthen Gemini’s multimodal capabilities against Claude and GPT. The Decoder | TechCrunch | 3D Printing Industry
Why it matters: Google is assembling Gemini’s multimodal stack through acquisitions rather than building everything in-house. 3D generation, emotional voice, and scientific reasoning are the gaps that keep AI assistants from feeling truly intelligent. Google is buying its way to closing them.
🇪🇺 7) EU publishes first AI transparency Code of Practice
The European Commission released the first draft Code of Practice on transparency for AI-generated content. The rules require AI providers to mark generated and manipulated content — including deepfakes — in machine-readable, detectable, and interoperable formats. Deployers of generative AI for professional purposes must clearly label deepfakes and AI-generated text published on matters of public interest. The second draft is due mid-March 2026, with the Code expected to be finalized by June and rules becoming binding on August 2, 2026. European Commission | Jones Day
Kernel take: August 2, 2026 is now the compliance clock. If you’re building AI tools that generate content for European markets, watermarking and labeling aren’t optional features — they’re legal requirements. Start engineering for transparency now.
🎓 8) Free AI training targeting 10 million workers by 2030
The UK government expanded its AI Skills Boost programme to every adult in the country, with a target to upskill 10 million workers by 2030 — nearly a third of the workforce. The initiative is being called the largest targeted training programme since the creation of the Open University. Courses are available online, some as short as 20 minutes, with completion recognized through a government-backed digital AI foundations badge. The programme has already delivered 1 million course completions since June 2025. Additional measures include £27 million in new TechLocal funding and a new AI and the Future of Work Unit to advise across government. GOV.UK | Computing
Why it matters: This is the most ambitious national AI upskilling programme anywhere. The bet is clear: if AI is coming for jobs, the best defense is making sure the workforce can work alongside it. Other governments will be watching closely.
📊 9) The rise of vertical AI — generic chatbots are dead
The consensus emerging from enterprise AI deployments in early 2026 is that vertical, domain-specific AI is winning. Tempus is processing clinical and molecular data across cancer, cardiology, depression, and infectious diseases with models trained exclusively on medical data. Seventy-one percent of CPG leaders adopted AI in at least one business function in 2025, with measurable revenue growth and cost reduction. Analysts say user experience will replace model intelligence as the primary sustainable differentiator in 2026, with companies competing on workflow integration rather than benchmark scores. The AI & Big Data Expo in London reinforced this shift, with enterprise leaders focused less on large language models and more on infrastructure: data lineage, observability, and compliance. AIMultiple | Aforza | AI News
Kernel take: This is the thread connecting the entire week. Anthropic’s plugins are vertical AI for legal and finance. The market panic is about vertical AI replacing horizontal SaaS. And the enterprises actually deploying AI are choosing domain-specific over general-purpose. The generic chatbot era is over. The workflow era has begun.
📬 Final Thought
This was the week AI stopped being a model race and became a market event. Anthropic shipped plugins that wiped $285 billion off software stocks. OpenAI shipped a model that helped build itself. Jensen Huang told everyone to calm down, and nobody listened. Meanwhile, Claude Sonnet 5 leaked, TSMC bet $56 billion on continued AI demand, and the EU set a compliance clock for August.
The SaaSpocalypse may be overblown. But the message is unmistakable: general-purpose SaaS is no longer safe from AI disruption. The question isn’t whether AI replaces these tools — it’s how fast the business models underneath them unravel.
Want this in your inbox every week?
Subscribe at thekernel.io
Sources: TechCrunch, CNBC, Bloomberg, CNN Business, OpenAI, Anthropic, Yahoo Finance, Dataconomy, DEV Community, MarkTechPost, Fintool, The Decoder, 3D Printing Industry, European Commission, Jones Day, GOV.UK, Computing, AIMultiple, Aforza, AI News
