Sapiom

Where agents run

[Sapiom @sapiom]

1/ We've acquired @fewsats, the payment infrastructure for AI agents.

The industry over-invested in reasoning and ignored access. Models can plan, but the second an agent has to act, there is nothing beneath it.

Intelligence without access is a demo.

[Sapiom @sapiom]

@jeffinvenice @venice_inc @appliedcompute we want to chat! sent ya'll a dm :)

[Sapiom @sapiom]

1.37M gateway calls today alone with a 99% success rate and latency of 1.16s

for our customers, Sapiom actively sits in the critical path for agent execution providing routing, payments, observability and governance that can handle real volume reliability

[Sapiom @sapiom]

1.37M gateway calls today alone with a 99% success rate and latency of 1.16s

for our customers, Sapiom actively sits in the critical path for agent execution providing routing, payments, observability and governance that can handle real volume reliability

[Sapiom @sapiom]

agents running on @sapiom executed the same volume of work in the last week (1.2 to 1.9 million calls per day) as they did in late May but at 70% lower cost per call

the underlying provider costs on our highest-volume service dropping 96%

same workloads at a fraction of what they cost

[Ilan Zerbib @i_zerbib]

Most people hear 'routing' and think model-to-model failover. That’s useful, but it’s the small version of the problem.

The real problem is helping agents complete work. That means routing every capability they need (intelligence, search, browser, voice, compute) to the best path at the right moment, based on cost, latency, reliability, and success rate.

Reliability is one outcome. Staying cheap is another. Actually finishing the task is the point.

Agents won’t succeed by having access to one model or one tool. They’ll need access to the whole machine economy, intelligently routed.

That’s Sapiom.

[Ilan Zerbib @i_zerbib]

x402: $100M in transactions since launch.

Agentic economy in 2030, per McKinsey: $5 trillion.

That is 8.7x a year for five straight years. A 50,000x climb.

We are just getting started.

[Sapiom @sapiom]

@AnthropicAI had an outage yesterday and it's a good reminder that provider incidents are an inevitability of building on third-party infrastructure at scale.

when one provider degrades, routing shifts automatically to another without any manual intervention required.

for @sapiom yesterday looked like 1.37M gateway calls processed at a 99.12% success rate.

one provider's bad day doesn't have to be yours.

[Sapiom @sapiom]

one of the most important things we've learned about agents in productions is stickiness comes from breadth of capabilities, not depth

if you only give access to 1-3 capabilities they churn inside 3 days...
at 4 - 29% retained past a week...
at 5 - 61%
at 6 - 88%
at 7+ - 96-100%

if you make builders wire in each capability manually with separate credentials, separate billing, separate integration you've just made breadth expensive

this is a big why our platform has to shipped the full capability stack on day one

[Ilan Zerbib @i_zerbib]

SF agent builders: what are you working on? Drop a demo or a link.

No pitch, genuinely want to see what people are building in production right now. Always down to grab coffee with people pushing on this.

[Sapiom @sapiom]

there's a big difference between using AI to optimize your internal ops and building a product where agents are what you're actually selling to customers.

there are many options for the first thing.

Sapiom was built for the second.

[Sapiom @sapiom]

Agents can think. Agents can plan.

But when they try to act they hit infrastructure built for humans.

No wallet. No identity. No access. No way to move without a human in the loop.

That’s the gap we’re closing.

[Sapiom @sapiom]

Huge week for the team @sapiom

1. We partnered with @Mastercard on their new Agent Pay for Machines. Now they close the gap at the settlement layer and we handle runtime, execution, and access. Together, agents can act, transact, and get real work done in production.

2. We acquired @fewsats. Founder, Jordi Montes (@positiveblue2) invented the L402 protocol, the first payment rails built for machines, not humans. He spent years building toward the same belief we hold.

This is just the beginning with a lot more is coming over the next few weeks.

Stay tuned!

[positiveblue ⚡️🍠 @positiveblue2]

Excited to share that I’ve joined @sapiom as a founding engineer.

We’re building the execution platform for agents in production.

If you’re running agents and hitting the limits of what they can do, get in touch.

[Sapiom @sapiom]

6/ Read the full story here https://t.co/yw7ofSdHhM

[Sapiom @sapiom]

1/ We've acquired @fewsats, the payment infrastructure for AI agents.

The industry over-invested in reasoning and ignored access. Models can plan, but the second an agent has to act, there is nothing beneath it.

Intelligence without access is a demo.

[Sapiom @sapiom]

5/ We weren't looking for an acquisition but for builders who understood that payments cannot be an integration bolted on after the fact. They must be a primitive at the execution level.

@positiveblue2 joins @sapiom as a Founding Engineer to lead Payments.

Welcome aboard, Jordi and Fewsats!

[Sapiom @sapiom]

@siddharths @i_zerbib thank you! it's just the beginning.