Sapiom

We're excited to welcome 23 startups to the inaugural Mastercard Start Path Agentic Commerce & Services cohort. As the largest Start Path cohort to date, these innovators are accelerating the next generation of technologies and services that enable autonomous, AI-driven experiences across the payments and commerce ecosystem. From agentic payments and AI shopping to identity, trust, merchant enablement and financial infrastructure, they're helping build the foundation for the future of commerce. Welcome to Start Path! Learn more here: https://lnkd.in/eb-eKqKv

The AI agent that hallucinates in front of a client is, contrary to popular belief, likely an architecture problem. We just shipped an integration with Sapiom that tackles both halves. The You.com Web Search API returns ranked results in under 300ms at 93% accuracy, and the Contents API strips any URL down to clean, LLM-ready text—no scraper, no browser automation to maintain. Sapiom enforces per-agent spend mandates before execution, not after. That's the difference between "the agent overspent and we caught it on the invoice" and "the agent couldn't." Chain youcom_search → youcom_contents and you get retrieval with a budget attached—metered and audited per call, with zero API keys to rotate. Run multiple sub-agents (retriever, synthesizer, fact-checker) and each gets its own identity and spend policy, visible in real time. Full breakdown below. 👇

Eleven months ago, I founded Sapiom. Six months ago, we announced our $15 million seed. Today, we are announcing our $35 million Series A, led by Dragonfly, with participation from Accel, Anthropic, Gradient, Coinbase, Operator Collective 🔆, Formus Capital, and VanEck, alongside continued support from existing investors including Okta Ventures, Menlo Ventures, and Array Ventures, bringing total funding to $50 million. Alongside the round, we’re revealing what we’ve spent those eleven months building: Sapiom Router, Sapiom Agent Studio, and Sapiom Agent Runtime Every tool humans have ever built was designed for a human operator. For the first time, software can be the operator. Agents are deciding which tool to use, which service to call, and which action to take next on their own. Almost none of our software or financial infrastructure was built for that. Today, anyone can build a capable agent demo in a weekend. Getting it to run reliably at scale, at a cost the business can sustain, is much harder. Before moving to Sapiom, one customer running thousands of agents asked what should have been a reasonable question: How many agents ran yesterday, and what did they cost? The answer was $12,000, but nobody knew what it had paid for. They could not see which agents ran, which tasks succeeded, where agents got stuck, or how many times they retried the same failed step. Sapiom is production infrastructure for AI agents. It sits at the exact moment an agent is about to act. For every action, Sapiom selects the best allowed path across models, compute, tools, and services based on cost, performance, permissions, and the boundaries the business has set. Because we run our own inference stack on dedicated GPUs, you get the most affordable path while also keeping your data in the US. Budgets and permissions are enforced before anything runs, and every step is recorded with its cost and outcome. - Sapiom Router lowers what you pay per call by matching every call to the right model instead of defaulting to the most expensive one. It takes two lines of code to try against agents you're already running. - Sapiom Agent Studio is where builders create, test, inspect, and deploy agents as real code inside the coding environment they already use. - Sapiom Agent Runtime is the production infrastructure where those agents run, recover from failures, access outside capabilities, and operate within the boundaries you set. In our first six months, Sapiom has processed more than 270 million transactions across more than 100,000 daily agent runs. One customer reduced its inference costs by 75%. We're starting with cost because that's where the economics break first. Our ambition is to remove everything standing between builders and the next trillion agents. Bring the agents you already have, or start fresh. Your first agent can be running on Sapiom in under 10 minutes.

Got quoted in The Information yesterday, in a piece on Cursor's price hikes. Companies are realizing they're overspending on AI. The default reaction is to find a cheaper model, or, as we've seen countless times in the news and all over social, terminate agents or limit the use of tooling. Over the past few months alone, we've watched companies tell employees to "go all in" on AI, resulting in high frontier model bills, and now those same teams are being told to scale back. Financially, for a business, understandably so. But the only way we unlock the transformation that AI can have on our economy is if we build the infrastructure to support it. Guardrails that show where spend is going, control it before it happens, and route every dollar to what actually works. We expect to cut our own AI spend by 25% doing exactly this — and it's the reason I started building in the first place. On building, this is an exciting time to do so, and IMO, for technological evolution to continue, we have to find a more cost-efficient, realistic, scalable way to do it. We're just getting started. Thank you, Laura Bratton, for having me! I look forward to our next chat. Take a full read in the comments.

You're burning mountains of money on LLMs, but do you know what you're getting for it? Not the total. You get that on an invoice once a month. But the breakdown: what an agent spent, on what task, for which customer, and whether the run that cost the most produced anything at all. Most teams can't answer that, and it isn't because of carelessness. The current tooling being used doesn't tell them. Spend shows up as one aggregate number, weeks after the decisions that caused it. Most people can't remember what they had for lunch yesterday, let alone what an agent they used weeks ago was doing. So then what happens? Someone senior asks whether the agent is worth what it costs; there's no answer. An agent nobody can defend is an agent that gets switched off. status: terminated root cause: cost_ It doesn't have to be this way. I'm freeing up my calendar this week to take more calls. DM me what you're building, and let's see how Sapiom can help.