The Spinning Plateau

The wild elevator ride of winter 2025

Winter 2025 to spring 2026 were a wild elevator ride in AI coding and agents. Claude Opus 4.5 was released in late November and ushered in the era of “I barely look at code anymore” when developers stopped using coding agents inside of their editors and instead just talked to them while the agents wrote the code and tests. Opus 4.6 cemented this behavior change in February, and GPT 5.5 made it pretty clear in April 2026 that we’re unlikely to go back to spending a lot of time in IDE’s.

Meanwhile, after incubating for a couple of months and renaming itself nearly every day for a week, OpenClaw launched in late January as the first personal agent that promised to do all of your work and run your life for you. By March 2nd, it had reached a quarter million GitHub stars and everyone rushed out to buy Mac Minis as domiciles for their crustaceans. OpenClaw was quickly followed by other claws: NanoClaw, PicoClaw, Hermes Agent, and Nvidia’s NemoClaw.

Claws were tremendously exciting in February and are no doubt going to evolve into something genuinely useful, but having run one for a few months, it has not been transformative yet. As life-changing technologies go, my 3D printer has probably made a bigger and more durable dent. It just doesn’t think yet.

This week Anthropic released Fable, a Mythos-class model with a jump in capability for long-running autonomous work, as well as an increased scope of refusals — it’s shadow-banning AI researchers.

What about jobs?

Q1 ’26 also saw a spike in layoffs at tech companies, which are being justified somewhat dubiously by productivity increases from AI. It’s dubious because those productivity gains have not shown up convincingly in the data yet, and may not materialize until companies embrace new workflows. However, a decline in job opportunities for new college grads does seem to be real and is worrying. Overall, software engineering hiring is still trending up, mostly. Update: Stanford Digital Economy Lab launched a collection of AI Economy indicators this week that is worth bookmarking.

I have no doubt that new model architectures and new approaches to memory and continuous learning will deliver more breakthroughs and wild elevator rides. Things definitely still feel like they are spinning. For now (this week? today?) it feels more like a spinning plateau than a spinning elevator. The earth spins through space at 107,000 km/h and we’re used to it. The challenge remains: Making sure the spinning AI elevator raises everyone up, not just a few.

Featured Image: Vermillion Cliffs National Monument, photo by: Bob Wick, U.S. BLM