Sett, the Tel Aviv-based startup using AI agents to automate game publisher marketing, has raised $30 million in a Series B round led by Greenfield Partners, with participation from existing investors Bessemer Venture Partners and F2, the company announced March 29. The round brings total funding to $57 million.
The money is not the story. The architecture is.
Sett has built what its founders — Amit Carmi, CEO, and Yoni Blumenfeld, CTO, both alumni of Israel's Unit 8200 intelligence service — describe as a multi-agent system for user acquisition creative production. At the center: Claude Opus 4.5, Anthropic's largest model, acting as an orchestrator that manages the production pipeline and delegates specific tasks to specialized diffusion models. This is not a single LLM wrapper over a spreadsheet. The dependency graph runs Claude as the decision layer, with downstream models handling the visual generation that the orchestrator plans around.
The company developed its own internal benchmarking metric to make this concrete. They call it the Last Mile — a measure of how much human work remains after the AI system finishes its portion of a task. In comparative tests against other leading models, Sett found Claude Opus 4.5 left less Last Mile work than alternatives on creative coding tasks. That finding comes from Anthropic's own case study on Sett, which is the most verifiable primary source in this announcement.
The output metric is easier to check. Before Sett integrated Claude into its system, producing a single playable ad required about a month of human planning, design, creative coding, and iteration. With Claude as the core model, that dropped to 75 minutes. That's a production cycle reduction of roughly 90 percent — and it maps to Sett's claims of cost and speed improvement, even if the specific 15x/25x figures that appear in Bessemer Venture Partners' profile of the company are Sett's own marketing numbers rather than independently audited benchmarks. The 75-minute figure, sourced directly from Anthropic's case study, is what can be verified without relying on the company's self-reported multiplier claims.
For one game studio using Sett, the adoption signal is stark: 90 percent of ad spend is now driven by Sett-generated ads, with two leading ad concepts commanding about a quarter of total spend and driving the highest return on ad spend. The system can already identify when parts of an ad are not working — a puzzle too difficult for players to complete, a call-to-action that underperforms — and automatically ideate simplifications for the next iteration. That feedback loop is the architecture paying off in practice, not just in the announcement.
Sett is generating tens of millions of dollars in revenue, according to Calcalist, with roughly 50 employees and growth that the company describes as over 10x year-over-year, achieved entirely through word-of-mouth in the gaming industry. The customer list spans Zynga, Playtika, and Papaya among the larger names, with a waiting list the company puts at over 100 gaming studios — suggesting demand is running ahead of onboarding capacity.
The market logic is straightforward. Mobile game studios spent $29 billion on user acquisition ads in 2023, according to AppsFlyer, against a projected industry revenue trajectory that reaches $180 billion to $250 billion annually by 2030. The UA side of that equation has historically been labor-intensive: human teams building, placing, measuring, and iterating creative across dozens of networks and formats. Sett's bet is that automating that pipeline — specifically the creative production and optimization loop — is worth a meaningful percentage of the spend it replaces.
The investor list adds a gaming-industry credibility signal. The round included Ben Feder, former CEO of Take-Two Interactive and former president of Tencent's gaming division, now at Tirta Fund. Bessemer led the $15 million Series A in May 2025, which brought total funding at that point to $27 million.
What the story is not: Sett is not claiming to have solved full autonomous user acquisition. The system automates creative production and optimization ideation; the actual media buying and budget allocation decisions still involve human review at most customers. The 15x/25x cost and speed figures that have circulated in secondary coverage trace back to Bessemer Venture Partners' profile of the company — Sett's own characterization, not an independent benchmark. The verifiable production-time reduction (one month to 75 minutes) is the more reliable anchor.
The expansion plan is where the story gets speculative. Sett says it plans to move beyond gaming into fintech, apps, and e-commerce — industries also driven by performance-based marketing — by the end of 2026. That's a reasonable adjacent market, but the gaming vertical is where the architecture has actual deployment evidence. Readers should separate what Sett has built from what it says it will build.
The honest framing: Sett's architecture is real agent infrastructure, not UA automation wearing an AI coat. The multi-agent orchestration with Claude as orchestrator and diffusion models as specialized workers is a genuine architectural choice, not a marketing label. The Last Mile metric and the 90-percent ad-spend figure are the kind of concrete evidence that earns the benefit of the doubt. The 15x/25x claims need independent corroboration before they belong in a headline. The expansion roadmap is a plan, not a product.
What's worth watching: whether the "Last Mile" metric becomes a standard benchmark in the agent infrastructure space, and whether Sett's approach — specific model specialization rather than one model handling everything — becomes a pattern other agent frameworks adopt for creative production tasks. The dependency graph matters. Sett is betting that who delegates to whom is the competitive moat.
