🍦 AI Wrote All of Spotify's Code Last Month
Plus: OpenClaw joins OpenAI, Sonnet 4.6 drops, and the Pentagon's AI breakup gets ugly
FOMA AI News
February 19, 2026
Scooped from 366 sources. Read it fast before it melts.
⚡ If you only have 1 minute...
OpenAI Nabs OpenClaw: The fastest-growing GitHub repo in history just got acqui-hired by OpenAI. Personal AI agents are officially the new battleground, and Sam Altman is collecting the talent like Infinity Stones.
Sonnet 4.6 Drops: Anthropic’s “budget” model now embarrasses last year’s flagship Opus. 1M token context, better computer use. Devs say they prefer it over Opus 4.5. The Honda Civic just lapped the Ferrari.
MiniMax M2.5: A Chinese lab most people can’t name just dropped an open-source model matching Opus on coding benchmarks at $1/hour. “Intelligence too cheap to meter” isn’t a slogan anymore — it’s a pricing page.
💰 Market Moves
[ACQUISITION] Mistral AI Acquires Koyeb, Builds European AI Cloud
Mistral’s first-ever acquisition — cloud startup Koyeb — plus $1.4B committed for Swedish data centers. Revenue grew 20x to $400M+. Europe is finally writing checks instead of regulations. Refreshing.
[FUNDING] Anthropic Raises $30B at $380B Valuation
Thirty billion dollars. Single round. Not a typo. Revenue run rate is $14B, Claude Code writes 4% of all public GitHub commits.
[DEAL] Meta & NVIDIA Ink Multi-Billion Dollar GPU Deal
Meta is buying millions of NVIDIA’s Blackwell and Rubin chips because Zuckerberg’s “personal superintelligence” vision requires enough silicon to pave a highway. NVIDIA stays the only company guaranteed to profit no matter who wins the AI race.
[MOVE] Perplexity Ditches Ads Entirely
While OpenAI tests ads in ChatGPT, Perplexity killed its entire ad business to focus on enterprise. Bold move at a multi-billion valuation. Either they know something we don’t, or they just really don’t like money.
🛠️ Tech & Product
[ACQUISITION] OpenAI Acqui-Hires OpenClaw Creator
Peter Steinberger, creator of OpenClaw (fastest-growing GitHub repo ever), is joining OpenAI to build “the next generation of personal agents.” The viral AI assistant that manages calendars and books flights goes open-source under OpenAI. Sam basically said “nice agent, I’ll take the guy who made it.”
[LAUNCH] Anthropic Drops Claude Sonnet 4.6
The new Sonnet reads a million tokens at once (~15 novels), navigates your spreadsheets and websites like a slightly unhinged coworker, and is now free for everyone. Benchmarks match or beat Opus 4.6 at lower cost. Anthropic gave everyone a free upgrade and said “you’re welcome.”
[LAUNCH] MiniMax M2.5: Frontier Performance, Bargain Prices
China’s MiniMax dropped an open-source model scoring 80.2% on SWE-Bench, matching Opus on coding at a fraction of the price. $1/hour, 100 tokens/sec. The frontier pricing cartel just got undercut by a company most people can’t name.
[LAUNCH] Zhipu AI Drops GLM-5 — Trained on Huawei Chips
Zhipu AI released GLM-5, an open-weight agentic model with MoE architecture — trained entirely on Huawei chips, no NVIDIA needed. Free to use with strong tool-calling. Between this, Qwen 3.5, and MiniMax M2.5, this week’s open-source flood has enough entries for a bracket tournament.
[LAUNCH] Alibaba Drops Qwen 3.5 — 397B Params, Free
Alibaba open-sourced a 397B parameter multimodal model — text, images, video, 201 languages. Claims to beat GPT-5.2. Runs on your Mac. Chinese labs are speed-running “make frontier AI free” while American companies raise $30B rounds. Different business models, same vibes.
[LAUNCH] xAI Launches Grok 4.20 Beta With 4 Parallel Agents
xAI dropped Grok 4.20 (nice) with four named AI agents — Grok, Harper, Benjamin, and Lucas — that collaborate on research with a 2M token context window. The AI has coworkers now. They named one “Benjamin” so you know it handles the money stuff.
[PARTNERSHIP] Figma + Claude Code: “Code to Canvas”
Figma and Anthropic teamed up to convert AI-generated code into editable Figma designs. The “just have AI build it” workflow now has a return path to “wait, the designer needs to fix this.” Designers and devs can stop arguing about who goes first — the AI does.
[LAUNCH] Manus Launches Agents on Telegram — Gets Suspended
Manus shipped autonomous AI agents on Telegram for research, data processing, and document creation. Telegram promptly suspended them. Launching your agent army on someone else’s platform without asking is the tech equivalent of setting up a lemonade stand inside a Costco.
[RESEARCH] GPT-5.2 Claims Novel Physics Discovery
OpenAI says GPT-5.2 derived a new result in theoretical physics — a formula for gluon tree-level amplitudes. If you understood that, congrats, you’re a physicist. For everyone else: the AI simplified a formula for subatomic particle behavior and physicists confirmed it checks out. One step closer to “AI wins Nobel Prize.”
[LAUNCH] Cursor Launches Plugin Marketplace
Cursor dropped a plugin marketplace with MCP servers, skills, sub-agents, and hooks — basically an App Store for your coding AI. If your IDE has an app store, it’s no longer just an IDE. It’s a platform. VS Code is nervously updating its extensions page right now.
[LAUNCH] WordPress Gets a Built-In AI Assistant
WordPress — powering ~40% of the internet — now ships with a Gemini-powered AI assistant baked in. It understands your site, builds pages, and edits on command. This is the “AI reaches normies at scale” moment. Your uncle’s fishing blog is about to get a glow-up it didn’t ask for.
🗣️ Industry Chatter
[POLICY] Pentagon Threatens to Drop Anthropic Over Safety Limits
The Pentagon may label Anthropic a “supply chain risk” and kill a $200M contract because Claude won’t do weapons development without guardrails. Meanwhile, Claude is the only AI on US classified systems AND was used in the Maduro raid. The Pentagon is mad at the AI already working for them. Your tax dollars at work.
[PRODUCT] Meta Adding Facial Recognition to Smart Glasses
Meta’s Ray-Ban glasses (7M sold) are getting “Name Tag” — identifies people by looking at them and pulls up their info via Meta AI. Remember when we worried about Facebook tracking us online? Now they want to track us at brunch. Through sunglasses. Cool cool cool.
[CULTURE] Spotify Devs Haven’t Written Code Since December
Spotify built “Honk” (powered by Claude Code) — it generates fixes, deploys builds, and reviews code. Their best engineers now just review and direct AI output. Haven’t manually written code since December. Your coding bootcamp tuition is now a sunk cost. Condolences.
[MILITARY] Pentagon Used Claude in the Maduro Raid
The U.S. military used Claude in a classified op to capture Nicolás Maduro. The AI safety company’s chatbot helped plan a military raid — in the same week the Pentagon threatens to drop them for being too safe. Claude is simultaneously too safe AND running ops in Venezuela. Schrödinger’s defense contractor.
[STRATEGY] Microsoft Planning to Ditch OpenAI, Build Own Models
After pouring tens of billions into OpenAI, Microsoft is building its own AI models to reduce dependence. Mustafa Suleyman leading the charge. The corporate equivalent of funding your ex’s startup then building a competing one. Most expensive situationship in tech history.
[PATENT] Meta Patents AI to Post for You After You Die
Meta patented an AI that trains on a dead person’s communications to keep their social media posting after death. Your digital ghost will be arguing in comment sections long after you’re gone. Either a Black Mirror episode or a Tuesday at Meta. Both, apparently.
🍦🔥
DARE TO SUMMON THE BOT?
Our Ice Cream mascot is desperately looking to make new friends. Share us on LinkedIn and tag @FOMA AI News.
We guarantee a snappy comment (or a light roast) on your post within 24 hours.
WARNING: BOT HAS ZERO CHILL.