- As LLMs get better, they will do more and more of the legwork autonomously, and a lot more of our work will rise up the abstractions into oversight and understanding. LLMs can help here too because as intelligence and code are increasingly abundant, you can ask for large, custom, discardable software artifacts (e.g. web apps, video explainers) that would have never made sense to create before. Push the boundaries here and you’ll be surprised. x.com/karpathy/status/2105819303471976479
- Running lots of coding agents in parallel can create a new bottleneck: the human supervising them. More agents mean more context switching, coordination, review, and mental overhead. Instead of optimising for maximum agent concurrency, we should optimise the human-agent interface so one or two agents can be steered, understood, and reviewed much more efficiently, using faster feedback, structured outputs, visuals, and richer ways to communicate intent. blog.exe.dev/etoomanythings
- MCP works best when you stop thinking of it as a bunch of tools to dump into context and start thinking of it more like OpenAPI with intelligent discovery. Tools should be discoverable through good descriptions, return structured data, and be exposed to a sandbox where the model can wire them together. That is basically why CLIs work so well today: models are good at composing small primitives with bash. There is no fundamental reason MCP cannot work the same way. At this point, the bigger limitation is less MCP itself and more how MCP servers and harnesses are built around it. earendil.com/posts/you-said-no-mcp
- We may start by calling what LLMs do “thinking” because it’s a useful shorthand, while still insisting it isn’t really thought. But if thinking means taking in information, building internal representations, weighing possibilities, reasoning through them, and arriving at new conclusions, then the distinction starts to get blurry. The interesting question may not be whether machines think exactly like us, but whether we’re watching a different kind of thinking emerge. x.com/paulg/status/2103135915195371973
- The real luxuries in life are time, health, a quiet mind, slow mornings, the freedom to travel, rest without guilt, a good night’s sleep, calm and boring days, meaningful conversations, home-cooked meals, people you love, and people who love you back. feld.com/archives/2026/09/the-real-luxuries-in-life
- The Sun is weirder than you think. youtube.com/watch?v=eNkxTjirY-Q
What I’ve been reading (W40, 2026)
How our work rises into oversight, why the human is the bottleneck for agents, MCP as OpenAPI, a different kind of thinking, the real luxuries in life