Sapiom is a San Francisco company building a platform for developers to create, deploy, and operate AI agents in production, with an early focus on making the unit economics work. Today's launch introduces three products: a Router that sits in front of an existing agent (two lines of code, a base URL and an API key) and directs each request to the cheapest model that can handle it reliably while keeping high-stakes cases like security incidents or at-risk high-value customers on frontier models, an Agent Studio that plans an agent step by step and surfaces the projected cost per run before anything executes, and a Runtime that hosts deployed agents and exposes per-run and per-customer economics through a dashboard.
The pitch to builders is that most agents never reach production, and the ones that do often route every query, even trivial ones like "where do I see my invoice," through expensive frontier models. Sapiom claims the Router holds the same quality bar for roughly 70 percent less by matching each ticket to a smaller model when the task allows. Studio bundles model access, infrastructure, and tool calls so a described outcome can move from spec to a running endpoint in minutes rather than weeks, which is the "cost-first" wedge behind the launch.
Founder Ilan Zerbib is based in San Francisco and, per his own bio, previously led Shop Pay and Shop Cash as payments director at Shopify and earlier founded and exited Earny. Accel's own note describes him as a co-founder of Earny, one of the earliest AI shopping assistants, before leading payments engineering at Shopify. The company had previously raised a $15 million seed round led by Accel with participation from Okta Ventures, Gradient Ventures, Array Ventures, Menlo Ventures, Anthropic, and Coinbase Ventures , and earlier coverage framed Sapiom as a financial infrastructure layer letting AI agents securely purchase software, APIs, data, and compute . Today's $35M Series A, per the launch tweet, brings in Dragonfly alongside Accel and Anthropic and expands the product surface from payments plumbing to the full build, run, and observe loop for agents.
Comments (10)
Sign in to join the discussion.
N
Nadia KrastevAug 5, 2026
A trillion agents feels like the kind of number you pick after three espressos and a whiteboard. What happens at agent 999,999,999,999?
I
Ifeanyi OkonkwoAug 5, 2026
The teaser cut at 0:03 to the product shot is genuinely tight, but that voiceover sounds like it was recorded inside a filing cabinet. Fix the low-end and this thing sings.
P
Parvati R.Aug 5, 2026
Three products on launch day is a flex or a cry for help, and I honestly can't tell which yet.
R
Reyansh VaidyaAug 5, 2026
Before I get excited, who's handling procurement on the enterprise side? SOC2 Type 2, SSO via Okta, DPA templates ready to go?
M
Mikkel TovaldsenAug 5, 2026
Cost is the right first wedge, everyone else is out here selling vibes and eval dashboards. Curious what the pricing floor actually looks like once you strip the marketing.
C
Clara BeaumontAug 5, 2026
Been watching this team quietly cook for months and the thesis holds up. Agents that can't justify their compute cost are going to look silly by next year.
L
Lasse DahlAug 5, 2026
One more thing for the roadmap: give me a kill switch dashboard where I can watch my agents die in real time. Purely for emotional reasons.
M
Mert YildizAug 5, 2026
Tweet copy buries the products under the fundraise line, classic mistake. Lead with what you shipped, the money is context, not the hook.
J
Joon-ho P.Aug 5, 2026
Retention curves on autonomous agents are the whole ballgame. Would love to hear how you measure a successful agent-week before I call this interesting.
A
Amara OkaforAug 5, 2026
resurrection_teaser in the utm is doing more storytelling than the landing page. Someone on that growth team is having fun and it shows.
Work With Us
Book an intro call to explore our suite of services. We might have a waitlist.
Sapiom
Help unlock the next trillion agents.
Help unlock the next trillion agents.
Tags
TLVC Rating
Community Rating
No ratings yet
to rate this launch.
Sapiom is a San Francisco company building a platform for developers to create, deploy, and operate AI agents in production, with an early focus on making the unit economics work. Today's launch introduces three products: a Router that sits in front of an existing agent (two lines of code, a base URL and an API key) and directs each request to the cheapest model that can handle it reliably while keeping high-stakes cases like security incidents or at-risk high-value customers on frontier models, an Agent Studio that plans an agent step by step and surfaces the projected cost per run before anything executes, and a Runtime that hosts deployed agents and exposes per-run and per-customer economics through a dashboard. The pitch to builders is that most agents never reach production, and the ones that do often route every query, even trivial ones like "where do I see my invoice," through expensive frontier models. Sapiom claims the Router holds the same quality bar for roughly 70 percent less by matching each ticket to a smaller model when the task allows. Studio bundles model access, infrastructure, and tool calls so a described outcome can move from spec to a running endpoint in minutes rather than weeks, which is the "cost-first" wedge behind the launch. Founder Ilan Zerbib is based in San Francisco and, per his own bio, previously led Shop Pay and Shop Cash as payments director at Shopify and earlier founded and exited Earny. Accel's own note describes him as a co-founder of Earny, one of the earliest AI shopping assistants, before leading payments engineering at Shopify. The company had previously raised a $15 million seed round led by Accel with participation from Okta Ventures, Gradient Ventures, Array Ventures, Menlo Ventures, Anthropic, and Coinbase Ventures , and earlier coverage framed Sapiom as a financial infrastructure layer letting AI agents securely purchase software, APIs, data, and compute . Today's $35M Series A, per the launch tweet, brings in Dragonfly alongside Accel and Anthropic and expands the product surface from payments plumbing to the full build, run, and observe loop for agents.
Comments (10)
A trillion agents feels like the kind of number you pick after three espressos and a whiteboard. What happens at agent 999,999,999,999?
The teaser cut at 0:03 to the product shot is genuinely tight, but that voiceover sounds like it was recorded inside a filing cabinet. Fix the low-end and this thing sings.
Three products on launch day is a flex or a cry for help, and I honestly can't tell which yet.
Before I get excited, who's handling procurement on the enterprise side? SOC2 Type 2, SSO via Okta, DPA templates ready to go?
Cost is the right first wedge, everyone else is out here selling vibes and eval dashboards. Curious what the pricing floor actually looks like once you strip the marketing.
Been watching this team quietly cook for months and the thesis holds up. Agents that can't justify their compute cost are going to look silly by next year.
One more thing for the roadmap: give me a kill switch dashboard where I can watch my agents die in real time. Purely for emotional reasons.
Tweet copy buries the products under the fundraise line, classic mistake. Lead with what you shipped, the money is context, not the hook.
Retention curves on autonomous agents are the whole ballgame. Would love to hear how you measure a successful agent-week before I call this interesting.
resurrection_teaser in the utm is doing more storytelling than the landing page. Someone on that growth team is having fun and it shows.