Tokens & Signals for 5/13/2026. We scanned ~1,200 Twitter accounts (1071 tweets), 13 subreddits (45 posts), Hacker News (7 stories), 6 newsletter posts, 5 podcast episodes, 184 Discord messages, and leaderboard data for you. Estimated reading time saved: ~10 hours.
* Anthropic has officially overtaken OpenAI in business adoption, hitting 34.4% market share to OpenAI's 32.3% per Ramp data. x.com/AndrewCurran_/status/2054582686698848294
* OpenAI is fighting back hard, offering two months of free Codex access to new business customers. x.com/sama/status/2054626219858293128
* Figure Robotics proved humanoid autonomy with an 8-hour livestream of their Helix-02 system working with zero human intervention. x.com/adcock_brett/status/2054603963996278786
* Anthropic's Mythos Preview became the first model to solve the "Cooling Tower" cyber range end-to-end. x.com/bcherny/status/2054617810253615147
* @GaryMarcus on the "jobpocalypse" discourse: "The long-term impact on coding roles is up for debate, but widespread disruption is starting to feel less like a theory and more like a timeline." x.com/GaryMarcus/status/2054364773438104044
* LangChain launched SmithDB, a database built specifically for the massive observability needs of complex AI agents. x.com/LangChain/status/2054658661776244936
* Devin AI can now autonomously test Android apps — a big deal for anyone tired of manual mobile QA. x.com/cognition/status/2054639069230198934
* Recursive AI emerged from stealth with $650M+ to build self-improving, research-conducting AI. x.com/chrmanning/status/2054579049197146469
* Condé Nast's CEO is telling teams to prepare for zero search traffic, betting on a future where AI answers users before they ever click a link. x.com/tbpn/status/2054316964731040207
* Local LLM fans are successfully running Qwen 3.6 27B on solar-powered home setups using M1 Max chips. reddit.com/r/LocalLLM/comments/1tbfcfe/solar_po...
Best to Build With Today
* Coding — gpt-5.4-xhigh (Best-in-class for agentic tasks).
* Reasoning — claude-opus-4-6-thinking-auto (Top pick for complex logic).
* Chat — gemini-3.1-pro (Current Chatbot Arena leader overall).
* Open-source — Qwen 3.6 27B (Community favorite for efficiency and raw power).
Deeper Dives
💼 Industry & Business
Anthropic Overtakes OpenAI in Business Adoption
Fresh data from Ramp shows Anthropic now powers 34.4% of businesses, edging out OpenAI at 32.3%. The shift comes down to companies preferring Claude 3.5/3.7's steerability and low hallucination rates for high-stakes API workflows — basically, when it actually matters, people are trusting Claude more. 📱 Twitter
OpenAI Offers Two Months Free Codex
OpenAI is offering two months of free Codex access to new business customers. It's a pretty clear response to enterprise teams drifting toward competing APIs, and a sign that the business market is genuinely up for grabs right now. 📱 Twitter
Condé Nast Signals End of Profitable Search Era
CEO Roger Lynch has told Condé Nast brands to plan as if organic search traffic will eventually hit zero, forcing a pivot toward direct audience relationships and licensing deals. That's a pretty stark internal admission about where the web is headed. 📱 Twitter
Perplexity Enterprise Momentum
Perplexity is quietly evolving into a serious research platform — PayPal alone is processing 74,000 weekly tasks through its secure internal RAG pipelines. That's not a pilot program anymore. 📱 Twitter
🧠 Models & Research
Mythos Preview Cracks Cyber Ranges
Anthropic's Mythos model navigated the "Cooling Tower" cyber range end-to-end, posting a 40% improvement in identifying exploit chains over traditional automated tools. Cybersecurity is quickly becoming one of the more interesting proving grounds for frontier models. 📱 Twitter
Researchers Debate Multi-Agent Efficiency
A study across 22,500 scenarios is pushing back on the "more agents = better results" assumption. Turns out, multi-agent systems frequently drown in their own overhead compared to a single, well-designed model. More isn't always more. 📱 Twitter
Recursive AI Emerges from Stealth
Recursive raised over $500M to build AI that autonomously runs its own experiments. The bet is that the next big leap in AI capability won't come from scaling data — it'll come from recursive research loops where the model improves itself. 📱 Twitter
🚀 Products & Launches
Figure Robotics' 8-Hour Autonomous Milestone
Figure ran its Helix-02 humanoid for 8 straight hours without a single human stepping in — performing unscripted warehouse tasks using end-to-end neural networks. An 8-hour unscripted livestream is a very different kind of demo than a carefully choreographed highlight reel. 📱 Twitter
LangChain Launches SmithDB
LangChain's new SmithDB is purpose-built for agent observability, designed to store the messy, nested execution traces — tool calls, reasoning steps, the works — that standard databases just weren't built to handle. 📱 Twitter
Devin's Mobile QA Capabilities
Cognition has expanded Devin to support Android Virtual Device (AVD) testing, meaning it can now build, launch, and validate mobile apps on its own. If you're doing mobile development, this one's worth keeping an eye on. 📱 Twitter
Funding & Deals
* Recursive AI: Raised $650M+ from AIX Ventures and others to automate AI research cycles. x.com/chrmanning/status/2054579049197146469
Launches
* SmithDB: LangChain's new database for agent observability.
* Devin Android Support: Full autonomous mobile testing integration. x.com/cognition/status/2054639069230198934
Closing thought: The Condé Nast news is a canary-in-the-coal-mine moment. When a major media CEO tells their teams to assume search traffic goes to zero, the "link-click" economy isn't dying slowly — it's being scheduled for demolition. If you're building in content or software right now, you should be optimizing for agents, not browsers.