Paradigm GP @danrobinson explains the 99% automation paradox, where eliminating almost all of someone’s tasks can still leave their job intact: "You could imagine that 99% of someone's job gets automated. You might imagine that means we have demand for only 1% of their time. But in fact, that 1% of their job basically becomes their entire job." "We've seen this with software engineering. We don't really see humans being 100% replaced. 99% of the work they did was automated, and now they're spending all of their time doing the other 1%." "Output has gone up by some decent amount, but the labor share of output is still actually relatively constant." "Chad Jones, who recently joined Anthropic, and Chris Tonetti at Stanford, which sort of covers this. So I've been working on a game that might explore that concept." @paradigm

A new scaling paradigm for robotic model training: from raw trajectories to reusable, composable and distillable skills. Read the full blog: Three findings from our latest experiments: • Embodied RSI: the deployed policy generates real-world data that trains its stronger successor (22% to 52%). • A lightweight proxy model scores trajectories by utility and coverage, so a filtered subset can beat the full pool. • Expert models reach near 100% success for $5 to $10 in compute per task and can be chained into long-horizon behaviors. What Axis scales is a continuously expanding library of robotic capabilities that can be produced, composed and distilled into general-purpose models.

A fundamentally new computing paradigm is emerging where the user interface is just an AI agent, and the agent uses the computer. It’s not hard to imagine a future where the vast majority of software users are AI agents, not humans. The agent can interface with the user naturally, speaking their language, and translate that into the correct input to get what they want. This solves the problem of the user having to learn to how to use various apps and services. Even the biggest power users won’t know how to use the software as well as these agents as computer use models improve. Services may end up looking more like APIs, with less emphasis on a single fixed GUI. Because these models can write code, they can create custom user interfaces on the fly perfectly suited to a given task, integrating multiple data sources, etc. The design team may think of many of these use cases and try and design around them, but users will be able to extend and remix the first party components with their own. With today’s computing paradigm, the leaders of the world are Apple, Microsoft, and Google. They may not be the kings in this new computing paradigm. Because if you’re just talking to an agent that’s generating UI on the fly and dispatching computer use tasks in the cloud, who cares if the client is Windows, Mac, Linux or some vibe coded kernel? OpenAI, Anthropic, and SpaceX are just as likely as Microsoft, Apple and Google to lead in this new paradigm IMO. Maybe even more likely, because the tech giants of the last few decades are all awkwardly fumbling their way into the AI era. The dominant OS will increasingly be Linux, from the looks of it.

One real paradigm shift in engineering: with sufficiently good agentic tools, constant change stops feeling like a threat. You can batch incoming changes, hand the bundle to agents, and let them surface conflicts, negotiate constraints, and return a coherent resolution. While they work, you stay available for the next hard problem. The human effect is larger than the workflow. Good agentic tools cut burnout because the most exhausting part of engineering has always been the politics of conflicting requirements. Egoless systems now absorb more of that load than people do.