Agents that run for weeks made Temporal a $12.55 billion company
Two funding rounds on opposite ends of the AI stack today: one selling the plumbing that keeps agents alive, one selling music toys that refuse to prompt a model at all.
Temporal raised $550 million in a Series E at a $12.55 billion valuation, and the reason it gives is duration. A year ago the company says most of its customer conversations were about payments, onboarding and fulfillment; now customers want agents that run for days, weeks or months without losing state. That is the same problem Temporal's founders have worked on since Amazon's Simple Workflow Service and Uber's Cadence, but the demand curve has gone vertical: valuation roughly two and a half times the $5 billion it carried in February, headcount doubled to 570 in a year, and more than 4,000 customers on Temporal Cloud. The most revealing line came from OpenAI's VP of infrastructure Venkat Venkataramani, who said durable execution is now "a core requirement for modern AI systems" and one of the main reasons OpenAI built its own orchestration framework on Temporal. Lightspeed's Anoushka Vaswani put the commercial logic plainly: "the demo is easy, production is hard, because the systems around the models can't handle real-world execution." The read: as agent horizons stretch, the scarce asset stops being capability and becomes reliability — the boring layer where a step failing at hour 60 means a lost week of work. We watched the other side of that shift in July, when OpenAI said its agents now do 3.1 days of research per human day — long-horizon autonomy is only as useful as the machinery that lets it survive a crash.
A Vinyl Bar in Shibuya raised a $5.5 million pre-seed to make music apps with no AI prompting in them. The company — a self-described "music software label" founded by ex-Spotify head of innovation Máuhan M Zonoozy — ships playable tools instead of generators: a browser extension for mixing, a pocket DJ app, a mash-up machine, with a "musical sandbox mixer" due this year. Backers include Mantis VC, SV Angel, Boxgroup and former Spotify executive Dawn Ostroff as an advisor. Zonoozy's argument is an economic one rather than a purist one: if AI makes generated songs abundant, then taste, participation and human perspective get rarer and more valuable, so the company keeps AI out of the product while still using it internally on infrastructure. In a week where every music model launch leads the news, a funded bet on the opposite thesis is a data point worth having.
What to watch: whether more agent-infrastructure vendors publish enterprise names attached to long-running workloads — durability claims get testable when the customers are listed.
Would you trust an agent with a task that takes a week to finish, or does the failure mode scare you off? Tell us in the comments.
