Sierra Acquires Takeoff to Launch Horizon Agent Platform

Bret Taylor's Sierra is buying Takeoff, a 14-month-old, three-person startup building long-horizon AI agents, and folding the team into a new Sierra platform called Horizon. Sierra's own announcement leans on one unusual fact for a company this small: Takeoff went from zero to near eight-figure ARR since the start of this year, with agents deployed into lending, healthcare, telecom, media, and travel.

The interesting question in enterprise AI for a while has not been which model wins a benchmark, but who owns the workflow. Takeoff founder Aakash Thumaty's pitch, quoted in the announcement, is exactly that: "Inference API in isolation is a commodity, and the true enterprise winner in AI will be the company that builds end-to-end solutions deeply specific to each industry and gets paid directly as a function of outcomes." Sierra is buying a small team that has been proving this out in production for less than a year.

Horizon is pitched at agents that run over days or weeks rather than a single chat turn, which is a real departure for a company that built its brand on customer support. As The Information's Stephanie Palazzolo noted on X, the deal pushes Sierra "beyond customer support" into "longer-duration tasks that could run for hours or days." In a post on X, Taylor called Takeoff "the leader in long-horizon AI agents" and welcomed the team alongside co-founder Clay Bavor.

The honest caveat is that a near-eight-figure ARR ramp in six months is a number Sierra is choosing to share about a company it is acquiring, and the announcement gives no purchase price, no customer names, and no independent evidence that the agents actually outperform incumbents in production. Outcome-based pricing is easy to talk about and hard to price when the outcome is a funded loan or a completed patient onboarding, and neither company has said how disputed outcomes get settled.

If the model works, the beneficiaries are vertical incumbents in regulated industries who can finally sell agents that get paid on results rather than seat licenses. If it does not, this reads as a talent-acquire dressed up as a product launch. Either way, it is a bet that the interesting agent race is about who packages the model into a paid outcome, not who trains the model itself.