Your Agent Needs a Bank Account
162 projects are racing to build financial rails for software that doesn't have a credit score yet
Capital One just spent $5.15 billion on Brex. Mastercard paid $1.8 billion for BVNK. Stripe bought Bridge for $1.1 billion, then came back for Privy.
That’s over $8 billion in acquisitions in the agent payments space from “legacy” financial infrastructure companies with due diligence teams paid to be skeptical. They looked at agent commerce and decided the trajectory is now worth billions.
But what exactly are they buying?
The map
Jordan Lyall got tired of tracking agent payment projects by hand, so he mapped the whole thing. The result: 162 projects across 6 layers, from settlement infrastructure up to application-level commerce.
$43M spent by agents already (USDC, a currency-backed cryptocoin). So agents use stablecoins at the moment: is that because stablecoins work better for machine-to-machine payments? Or because the only people running these experiments are crypto natives who already hold USDC? Probably a mix.
Stripe is telling you something
Stripe’s double acquisition is an agent/crypto bellwether.
Bridge gives them stablecoin rails.
Privy gives them wallet and key management.
Stripe clearly wants to be the payments layer for agents the same way it became the payments layer for SaaS. It built its empire on making payments easy for developers. Now it’s positioning to make payments easy for developers’ agents. Makes total sense.
Mastercard and Capital One are sweating: if agent commerce routes through crypto rails by default, traditional card networks get aren’t needed.
Beyond payments: the identity stack
But agents need more than money. Shiv Sakhuja’s tweet catalogued the other primitives being built:
AgentMail: so agents can have email accounts
AgentPhone: so agents can have phone numbers
Kapso: so agents can have WhatsApp numbers
Daytona / E2B: so agents can have their own computers
Browserbase / Browser Use: so agents can browse the web
In short, an identity stack for non-human workers.
We’ve covered Aaron Levie’s thesis before: every employee will soon have multiple agents (I use upwards of 12 daily). Those agents become the primary users of all software. The infrastructure for that world is no longer theoretical.
But “fast” doesn’t mean “real.” Most of these primitives are solving problems that barely exist today. How many agents actually need a persistent email address?
Most agent workflows I’ve built use API integrations, not inboxes per se. AgentPhone has a stronger case (voice agents are getting much better), but “agent needs a phone number” feels functionally more like “Twilio with cool kid marketing.”
162 projects, or 12?
Let’s check the $43M number. Tracked across 162 projects = c.$265K / project. Early stages for sure. And most of these projects are pre-revenue, pre-product, or pre-anything really.
The 140m transactions sound better until you notice that the x402 protocol alone accounts for 75 million of them in a 30-day window. One protocol is doing more than half the volume. The “ecosystem” is really a handful of active projects surrounded by a constellation of pitch decks.
So what’s real? The six-layer stack model is useful regardless of how many projects survive. Settlement (payment completion), wallet management, routing, payment protocol, governance, application. That mirrors how traditional payment stacks evolved over decades, compressed into months.
The governance layer (Layer 4, with 32 projects) matters most here. “May this agent spend?” is a question nobody has good answers to yet. Who authorizes an agent’s purchase? What spending limits apply? We probably need a few big meaty disasters to kick in the right guards.
What happens when an agent makes a bad buy? These are unsexy problems. They’re also the ones that guide whether agent commerce scales beyond demos. Ok, when when they do. It will happen for sure.
Follow the consolidation
Last time we covered this theme, we argued that software would need to be fundamentally redesigned for agents. The more interesting development is that finance is being redesigned for agents too. Finance moves faster when incumbents feel threatened.
Stripe is assembling a full-stack agent payments product.
Capital One and Mastercard are acquiring into the space too.
The standalone primitives (AgentMail, AgentPhone, individual wallet providers) are useful today but vulnerable tomorrow.
What’s converging underneath all 162 projects is a single concept:
The agent account.
Compute, identity, payments, communication, all bundled the way AWS bundled compute, storage, and networking into something developers stopped thinking about separately.
Seasoned pros will also be asking “how is this different to a service account?” Great question.
Most of Jordan’s 162 projects won’t survive that consolidation. The problems they’ve identified (authorization, spending limits, accountability) will.
How are you using agents these days?