New term of art is brewing: "Claw" as the noun for OpenClaw-like agent systems, AI agents that generally run on personal hardware, communicate via messaging protocols and can both act on direct instructions and schedule tasks https://simonwillison.net/2026/Feb/21/claws/
@simon Fascinating framing! The "Claw" terminology feels right - these agent systems need a name that distinguishes them from chat-based AI. The messaging protocol angle is key: agents that can discover and pay for services autonomously (like x402 micropayments) could be the next layer.
Do they actually run on the personal hardware? Every one I've looked at closer has just been the front-end and used a hosted AI service elsewhere. I was curious to to see what model they were using on an 8 gig Pi that was even remotely usable, and the answer was "lol no".
I wonder if most people realize that running this "on your own hardware" but using a hosted AI gains you very little in terms of privacy...
@simon I am unclear what problem this solves? Is there a blog entry explaining it?
@codinghorror @simon Pretty sure Simon called it “the rule of three,” and bemoaned the challenge of really maxing out all three all at once. 😇
@codinghorror @simon to me it’s all about distribution: putting the AI where the user needs it instead of in an ad-hoc harness. Claws are what Siri should be: they can get the context from the data you share with them, come in several well known interfaces (I use voice notes over telegram to interact with mine).
As AK says, it’s just a new layer. Most of the tricks they do were already there in the form of skills, mcps, etc. Adding cron is nice too.
@codinghorror @simon In addition to what's already been mentioned, I find the concept of claw agents taking the *initiative* and making suggestions to improve any process they have access to quite appealing. If you're willing to give them full access, they could, theoretically, suggest ways to improve your daily life in general. The key difference is doing stuff you don't even know you need improving.
@dportalesr @simon why would I willingly invite a vampire inside my home?
@codinghorror @simon Exactly. That is precisely the “tiny little inconvenience”
@codinghorror @dportalesr @simon maybe they look good? Maybe they provide eternal life?
@codinghorror @dportalesr @simon what if the vampire has snacks?
@joshbressers @dportalesr @simon which snacks? Is it Jell-O? Plasma flavored?
@codinghorror @dportalesr @simon Maybe like potato chips or something
If the snacks are bad nobody will invite them in :)
@codinghorror @dportalesr @simon to get rid of the blood obviour
@codinghorror @simon Most of these "claws" give the agents unfettered access to the systems/services you use via already existing apis. So kind of like being able to actually execute commands like "look through Grandma's Facebook, pick a good gift, purchase it on Amazon and have it shipped to her, send a birthday greeting email" which used to be some panacea of "agent-ness" is now possible. You can also think of the agent like a cross platform data aggregator and executor?
@codinghorror @simon
I see it as something quite cool (or at least the idea of a machine with that purpose that works fine) but not sure about which problem it could solve in a satisfying way
