Brian Reed: Hey — I want to run something by you before we even get started. I've been trying to describe this week to people and I keep having to stop and say, let me see, which version of 'agent' are we talking about?
Eliza Ward: Oh, that's — yeah, that maps exactly to what I've been stuck on.
Brian Reed: Because August 3rd, Microsoft ships Copilot Studio with the GitHub Copilot harness into general availability — that's enterprise agents, workflow orchestration, credentialed inside your company's directory. And then there's the Cloudflare story about agents holding Solana wallets, executing the x402 protocol, paying other machines without any human approval at each step. Same week.
Eliza Ward: They don't even sound like the same category of software.
Brian Reed: No — one is a product inside your IT infrastructure. The other is, if the framing holds, an independent economic actor. And Solana is cited as holding something like 49% of x402 agent-to-agent transaction market share as of early February 2026.
Eliza Ward: Where did that number come from?
Brian Reed: That's the part I don't get — there's no methodology attached. It just keeps circulating. Which means one side of this story has a dated GA announcement and paying customers, and the other side has infrastructure and a market share figure we can't verify.
Eliza Ward: Right — and that asymmetry is probably the honest frame for everything we're about to get into. What does an agent actually need to be able to do, and who's decided that in production already versus who's still proposing it?
Brian Reed: And here's where it actually clicks for me — think of it like a contractor buying supplies. Option A: your company hands them a corporate card with a spending limit, approval workflows, someone in accounting watching. Option B: the contractor shows up with their own wallet and settles directly with the lumber yard's AI purchasing system. Same outcome on paper. Completely different chain of authority underneath.
Eliza Ward: That's the whole thing, yeah. One card has your org's name on it.
Brian Reed: Right — so a Copilot Studio agent or a Databricks agent triggering a workflow at 2 a.m., rerouting supply orders for a hotel chain, that agent is operating inside your access controls. It can't spend money it wasn't authorized to spend because, I mean, it doesn't hold funds at all. It moves information through governed systems. Whereas the Cloudflare durable runtime with an on-chain wallet — that agent can autonomously negotiate, settle, hand stablecoins to another agent, and no human ever sees it happen.
Eliza Ward: Wait — and that's not just a preference. Those are actually incompatible trust models.
Brian Reed: Completely. One uses cryptographic identity proofs and on-chain settlement. The other uses your company's directory. You can't just swap the payment layer and keep everything else.
Eliza Ward: Nathan McCauley from Anchorage Digital actually framed this — an agent without payment capability is an advisor. An agent with programmable funds is an independent economic actor. That distinction matters because it's not just philosophical, it's the actual architectural decision you're making when you deploy.
Brian Reed: So the part I don't get — can you even migrate between them? Like if a company starts with Copilot Studio and then wants on-chain settlement later, is that... actually, no, I think the answer is the authorization model has to be rebuilt from scratch.
Eliza Ward: That's the right read. They're not competing on features — the question 'who authorized this agent to act' gets a fundamentally different answer depending on which stack you're on. One answer is your IT department. The other is a cryptographic proof on Solana.
Brian Reed: And that architectural gap is — hang on, because the McCauley framing does a lot of work here. If the threshold is 'does the agent hold funds,' then every enterprise agent this week is still an advisor. Copilot Studio agents trigger workflows. Databricks agents integrate governed data. Neither one settles externally. Neither one holds a wallet. By that definition they're all very sophisticated advisors.
Eliza Ward: Which is useful — but McCauley runs Anchorage Digital. Anchorage is a crypto custodian. He has an obvious interest in the wallet being the thing that matters.
Brian Reed: Right — so we treat it as a lens, not gospel.
Eliza Ward: Exactly. But even with that caveat — the distinction still cuts. Because, wait, the Solana side of this is the Send AI toolchain, the Solana Agent Kit, sub-second finality, fees under a tenth of a cent per transaction. That's machine-speed micropayments — technically viable in a way that wasn't true two years ago. The question is whether 'technically viable' and 'actually deployed at scale' are the same sentence right now.
Brian Reed: And they're not the same sentence.
Eliza Ward: No. And — okay, Microsoft Research dropped Orchard the same week. Early August 2026, open-source framework for scalable agentic AI. Same week as the Copilot Studio GA.
Brian Reed: So that's not just shipping a product — that's seeding an ecosystem around it.
Eliza Ward: Right. Solana has infrastructure. Microsoft is building distribution scaffolding — open-source, third-party developers, the whole thing. That's a different kind of gravity. Doesn't move them past the McCauley threshold, but it means the advisor category gets very, very populated very fast.
Brian Reed: Which makes the evidence question even sharper — and the part that actually changes my read on all of this is what we haven't verified yet, the numbers behind the on-chain side. It gets complicated there.
Eliza Ward: And that's the thing that actually breaks the clean version — the 49% figure is doing enormous work and we can't trace it back to anyone.
Brian Reed: Right — so let me just say this plainly. August 3rd, Copilot Studio GA, you can look that up. August 5th, Google's Agentic Data Cloud, database agents, named date, named product. The blockchain side? SvpChain, the $AGENTS token on Solana — they appear in reporting by name, but I mean, the substantive confirmed detail is... basically nothing. A name and a ticker.
Eliza Ward: That's — yeah, not nothing, but not much.
Brian Reed: And the x402 market share number — 49%, early February 2026, Solana — it keeps circulating because it sounds like a measurement. But there's no named source, no methodology, no third party ran that audit. It's probably promotional. Which means we're building an architectural argument on top of a stat that might be a press release.
Eliza Ward: Wait — and the x/social sentiment on this actually leans the same direction. The skepticism isn't just ours.
Brian Reed: Though social skepticism isn't a verdict either.
Eliza Ward: No, definitely not — but here's what actually tilts the evidence floor for me. There's a January 2026 academic paper on agentic BPM — agentic business process management — and it frames the enterprise trajectory as a disciplined shift from task automation to autonomous, data-driven process execution. Peer-reviewed. The blockchain side has builder Twitter and unverified market share.
Brian Reed: So the enterprise case has theoretical scaffolding, dated GA milestones, actual customers. The on-chain case has — I mean, a protocol and a claim.
Eliza Ward: Which puts us here: neither side can point to production data showing agents have actually crossed into true economic actorhood. That's the honest floor. One side is much closer to it.
Brian Reed: And that's — I mean, that's not a caveat you can footnote. That's the whole game. We spent an hour mapping two ecosystems — Copilot Studio, Cloudflare's durable runtime, x402, Solana's rails — and neither one can point to a production agent that has actually earned or spent money on its own, at scale, without a human somewhere in the approval chain.
Eliza Ward: Which means — wait, the architecture choices being made right now are locking in the answer for a capability that hasn't been demonstrated yet. Cloudflare builds the wallet layer, Microsoft ships Orchard as an open-source scaffold, Coinbase and Solana hold the x402 standard, and by the time an agent actually does initiate a spend autonomously at scale, the rails are already picked.
Brian Reed: Two job fairs, no confirmed hires.
Eliza Ward: That's — yeah, that's exactly it. If agents stay inside enterprise walls, the whole on-chain settlement story is a sideshow. If they ever escape into permissionless commerce — booking, trading, hiring, no human sign-off — then Cloudflare's runtime and Solana's sub-cent fees become table stakes overnight. We just don't know which one gets tried first.
Brian Reed: And I think that's actually an honest place to sit. Not uncomfortable — just unresolved. By 2027 we'll probably know. Right now it's genuinely unclear.
Eliza Ward: Good conversation. Appreciate you thinking through it.
