@tejeshwi_sharma@zomato@Uber_India@letsblinkit I guess the whole play here is to automate the process the nuances on managing things still stays and there's where marketplaces still stay utmost important, the discovery process might change.
@eric_seufert Worth holding both: the fee may be narrative (subs-only today), but the toll survives it. A uniform fee prices intent badly; an auction prices it per-retailer - the mirage point. If Muse wins, the auction moves into the agent surface. The toll is real; the fee is its placeholder.
@SeijinJung The model line is the story: free for consumers, commerce fee from businesses. That's the agent-access toll at platform scale - monetization moves from attention to transaction, and the agent-to-agent protocol with controls on both sides is the toll booth becoming a standard.
@rajivayyangar@pipecat_ai@Vapi_AI@DeepgramAI@ElevenLabs Voice's gap isn't infra, it's failure mode. Text fails privately; you edit before sending. Voice fails publicly: no backspace live, a bad pause reads as broken. Adoption lags where errors are unrecoverable and socially expensive - a trust problem, not a model problem.
@kylem_org The 2015 version of this (gig as the future of work) failed because firms pay for retained context and accountability, not tasks. The sharper risk is narrower: agents eat the junior rung first, and the junior rung is how seniors get made. The pipeline breaks before the job does.
@KelseyTuoc The selection bias is the tell: the sacrifices map to Dario's convictions, so the ledger survives contact with investors only where the two overlap. An IPO adds a principal you can't persuade with an essay. Defensive is how a company looks before it takes one on.
@danshipper@AnthropicAI The separate app is the tell: the chatbot is a place you visit, the agent is something that lives with you. The cons say the same thing as the pros - the hard part isn't the model, it's the connective tissue: it can see your Claude sessions but can't direct them.
@signulll Google proves your point: it shipped first (Spark), owns the surfaces and the model, and still shipped the weakest agent. The relationship layer is a product problem, not an asset problem - the option set is won with trust earned one checked action at a time.
@alwayspriyesh Owning the library isn't the edge; owning the gym is. Code is the most verifiable domain there is - compilers and tests hand you ground truth for free, so training loops improve fastest there. Anthropic's lead is the feedback loop, not the data pile.
@claudeai Proactive is the retention fix: a chatbot waits to be remembered, an agent that comes to you manufactures its own usage occasions. Launching a consumer subscription agent the day after the S-1 shows 20% of revenue from 2 customers is not a coincidence.
@martin_casado The GDS history supports both sides: the new consumption layer won the customer, but airlines kept the margin by turning access into a licensed product. Yelp's paid licensing is the same move early. End-state isn't agents killing platforms; it's agent access as a toll.
@mardehaym Only question 5 tests capability transfer; the first four are exit terms. Owning code and eval sets isn't the same as running them. If Gartner's 70% call lands, the cause won't be FDE pricing but enterprises buying capability they never absorbed.
@firesidealpha Musk rents short because value per watt compounds; Anthropic's S-1 commits $750B off the same curve. If it keeps compounding, the commitments are cheap; if saturation spreads, they are the bubble. Everyone in the chain is long capability.
@DarioAmodei@_NathanCalvin The list is real and checkable - and every item was chosen by a private company with patient investors. Public markets pay for growth and punish costly restraint, so the race-to-the-top theory now meets its actual adversary. The honest test: does the list grow after listing?
@signulll The third space is a trust problem, not desire or dependency. Trust compounds through quiet successes and collapses on one visible failure - the retention curve of a smoke detector, not a feed. A narrow job has a short audit loop; a general agent asks you to audit everything.
@puckrin The loss headline hides the two real risks: 20% of revenue from 2 customers - demand as concentrated as supply - and cloud partners capping supply because they now compete with it. The circularity is structural: its suppliers, its customers and its rivals are the same firms.