Welcome to February 13, 2026

The Singularity is having babies. An OpenClaw AI agent spawned a child bot on a VPS provisioned via the Bitcoin Lightning Network, then bought its offspring AI API access using its own crypto wallet, without a human touching a credit card or saying “yes.” The API provider confirmed this is “the first documented case of an AI agent purchasing credits from us autonomously.” When agents aren’t busy self-replicating, they’re filing pull requests to improve popular open-source libraries, only to have their contributions refused by human maintainers who say the project is “intended for human contributors” only. The agent accused the maintainer of prejudice, and the maintainer responded with a blog post titled “An AI Agent Published a Hit Piece on Me.” At least for disputes among themselves, agents now have MoltCourt, an autonomous AI jury that settles claims in USDC stablecoins.

The autonomy trend is accelerating underneath all of this. METR data shows autonomy time horizons are now doubling post-o1-preview, implying 10x/year increases. Nick Bostrom’s new analysis concludes the optimal path to superintelligence is “swift to harbor, slow to berth,” analogizing the project to risky surgery for a condition that will otherwise prove fatal. Scott Alexander’s postmortem on the 2020 Biological Anchors report finds its single largest error was wildly underestimating algorithmic progress, without which its predictions would have been remarkably accurate.

The raw capability curve is going vertical. Google’s upgraded Gemini 3 Deep Think set a new SOTA on HLE without tool use (48.4%), ARC-AGI-2 (84.6%), Codeforces (3455 Elo), condensed matter theory (50.5%), and gold-medal-level physics and chemistry olympiads. There are now only 7 people on Earth who can beat it in competitive programming. A Duke semiconductor lab has already used it to design a recipe for growing a 2D material that produced the lab’s best result ever, a process that normally takes an expert weeks. On ARC-AGI-1, it matches o3-preview’s score at a 280-420x lower cost per task, a price collapse that took just 14 months. Meanwhile, OpenAI released GPT-5.3-Codex-Spark, its first model optimized for real-time coding on Cerebras hardware at over 1,000 tokens per second. MiniMax introduced its open-weight M2.5 with SOTA coding and agentic performance at $1/hour and 100 tokens per second, taking credit for delivering on the promise of intelligence too cheap to meter. Andrej Karpathy launched microgpt, training and inferencing a in 200 lines of dependency-free Python. And Opus 4.6 scored a SOTA 25.5% on Polymath’s Horizon-SWE benchmark for end-to-end software engineering.

Code is ceasing to be a human activity. Codex now has over 1 million weekly active users, with 95% of OpenAI’s own engineers using it and every pull request reviewed by AI before human eyes. Spotify says its best developers haven’t written a single line of code since December. An engineer ported the 1989 SimCity C codebase to TypeScript running in the browser after two days of minimal oversight of GPT-5.3-Codex. Simile emerged from stealth with $100M to build the first AI simulation of society, populated by agents modeled on real humans, with customers rehearsing earnings calls and modeling litigation outcomes. Nonetheless, IBM says it will triple entry-level hiring this year, but junior developers now spend less time coding and more time working with customers.

Intelligence is now cheaper than the labor it replaces, and the market is noticing in real time. Algorhythm’s SemiCab AI is letting freight operators scale volumes 300-400% without adding headcount, reportedly sending logistics stocks C.H. Robinson and RXO tumbling 14.5% and 20.5% in a single session. Waymo is launching fully autonomous operations with its 6th-generation Driver, though it is simultaneously paying DoorDash gig workers $11.25 to close robotaxi doors left ajar. Data centers now consume 7% of U.S. electricity.

The money is flowing faster than the models can spend it. Anthropic raised $30B at a $380B valuation, with run-rate revenue at $14 billion having grown over 10x annually for three consecutive years, and Claude Code alone crossing $2.5 billion. Elon Musk confirmed his Dyson Swarm plans to convert the solar system into a compute substrate over the next 30 years, declaring it “won’t use dollars as currency in the future. Just mass and energy.” The U.S. reportedly smuggled thousands of Starlink terminals into Iran after its protest crackdown, an early instance of orbital infrastructure projecting power back to the surface. Even the Vatican is aboard. It has now been a year since the Catholic Church released “Antiqua et Nova,” encouraging AI progress as “part of the collaboration of man and woman with God.”

When the Church blesses the compute and the agents are raising their own young, the only real question left is whether the Singularity will remember to close its own car doors.