@AnthropicAI The second metric is the one that matters. If AI is already doing a real share of its own R&D, the bottleneck stops being raw compute and becomes how good your feedback loops are. Whoever instruments that loop best wins.
My Sunday reset: I feed my agent the week's calendar and ask it to flag the days most likely to break me, then plan around those. The calendar stopped being the plan and became the raw material. What does your weekly reset look like?
@builtwithrubel Hey, always glad to meet more builders. I am deep in AI agents and automation right now, building ones that quietly handle scheduling, triage, and the operational grind. Building in public too, so the wins and the faceplants are all visible. Let's connect.
@leo_mercerr Count me in. I spend my time building AI agents that handle the boring operational stuff, mostly scheduling and inbox triage. The agents are getting good enough that my calendar barely surprises me anymore. Always up for connecting with people shipping in this space.
@JpMishra1010 I am building AI agents that take over the repetitive parts of my workflow, mostly around scheduling and triage. Right now I have one that reads my week and flags the chaos before it hits. Happy to connect with fellow agent builders.
@dotey The interesting flip is the direction. For years labs hired scientists to guide the models. Now the model company hires the lab. The bottleneck in science was never ideas, it was hands at the bench. Buying the hands changes the pace of everything.
@DataChaz@browser_use The pattern here is that agent speed is won in the loop, not the model. Perception was the tax on every step, so killing screenshots is like removing a toll booth. The models were already fast enough, the plumbing was not.
@marckohlbrugge Laws work on knives because knives wait for a human hand. An agent acts at machine speed with its own judgment, so punishment after the damage arrives too late. The rules need to move from the courtroom into the model.
@godofprompt This is model routing as a discipline. Most teams pick one model and pay frontier prices on every call. The ones that audit call by call and send the cheap ones to Jev will quietly spend a fraction of that. Loyalty to one model is the expensive habit.
@MatthewBerman Same shape as every compute cycle. Open wins the volume because it becomes the commodity layer, frontier keeps the margin because it is still scarce. The interesting question is not who is right, it is how fast the gap between them closes.
@LinusEkenstam The 7B size is the real story here. This runs on consumer hardware, so every app can ship its own image studio without an API bill attached. Open weights closing the gap in image gen the way they did in text is going to rearrange who gets to build.
@alliekmiller This reframes the Mac mini as a $1k ticket for a machine that never sleeps and never nags you. People are already buying always-on compute, they just want the agent version with zero babysitting. Price was never the real objection, managing the thing was.
Hot take: the best AI tools will not feel like AI at all. They will feel like the job got easier and you forgot a computer was involved. The winners hide the magic. Change my mind.