Arcade Raised $60M to Be the 'Who Did What' Layer for AI Agents. The Lesson for Solo AI Consultants Isn't the Funding: It's What Just Got Commoditized.
On June 15, Arcade raised a $60 million Series A (led by SYN Ventures, with strategic money from Morgan Stanley and Wipro) to become what it calls the secure action layer behind production AI agents. The pitch is specific: prove which agent took which action, on behalf of which user, in which system, for every action, every time. That brings Arcade to around $72 million raised counting its earlier seed, and it's already running in production at a major US bank, Prosus, and LangChain.
If you're a solo AI consultant, the headline isn't the dollar figure. It's that a well-funded company just declared open season on a chunk of work a lot of independent builders were planning to bill for.
What Arcade actually sells
Strip the press-release language and Arcade is an MCP runtime that handles the boring, critical plumbing of letting an agent take real actions safely. It manages user authentication and authorization through existing OAuth and identity-provider flows. It enforces that an agent can only do the intersection of what the authenticated user is allowed to do and what the agent is scoped to do. And it keeps a complete audit trail (which agent, on behalf of which user, against which resource) so an enterprise can answer "what did the bot do" after the fact.
That's a real product solving a real problem. The reason agents stall before production isn't that they can't generate a plan; it's that nobody wants to hand a non-deterministic system live credentials with no record of what it touches. Arcade is selling the seatbelt that makes "let the agent actually do things" survivable in a regulated company.
The pattern, not the company
Here's the part worth internalizing. Six months ago, "I'll set up secure auth and permissions for your agents" was a credible line item on a solo consultant's invoice. Wiring an agent into a company's identity provider, scoping what it can touch, building an audit log so compliance doesn't veto the whole project: that was genuine, billable work, and not many people could do it.
A $60M round is the market telling you that work is becoming a primitive. When infrastructure gets this funded, the trajectory is always the same: the bespoke, hand-built version you were charging for turns into a platform feature, an API call, a few lines of config. It happened to authentication generally: nobody bills to hand-roll login anymore, they wire up an identity provider. It's happening to payments, to deployment, to observability. Agent governance is next, and Arcade just raised the money to make it next faster.
This isn't a doom post. It's a "don't anchor your business on the thing about to be commoditized" post. The builders who get hurt are the ones whose entire pitch is the plumbing that a funded platform is about to give away as a primitive.
Where the durable work moved
So where does a solo operator stand, then? On the layer the platform can't ship: judgment.
Arcade can give a company an audit trail and a permission model. It cannot decide, for this specific business, which actions an agent should be allowed to take at all. That's not an engineering question, it's a business-risk question, and it's different in every company. Should the support agent be allowed to issue a refund autonomously, or only draft one for a human to approve? What's the dollar threshold above which a human always signs off? Which systems are read-only forever because the downside of a wrong write is unrecoverable? Where does the human gate go, and who owns the consequence when the agent is wrong?
Those questions are exactly what a regulated company is terrified to get wrong, and they're exactly what a generic platform leaves blank for the customer to fill in. Arcade hands them the enforcement engine. Somebody still has to decide what to enforce. That somebody (the person who sits with a business, maps where an autonomous action is acceptable and where it's a fireable mistake, and configures the guardrails to match) is doing work no API call replaces.
The shift in framing is small but it changes everything you sell. You are not the person who builds the agent's permission system. You are the person who decides what the permission system should say, for a business that doesn't yet know its own risk tolerance. The first is plumbing, and plumbing gets commoditized. The second is consequence, and consequence stays expensive.
Where this take could be wrong
It's possible I'm overstating the speed. A $60M Series A buys runway and a roadmap, not instant ubiquity, and plenty of well-funded infrastructure companies have launched into the same press cycle and shipped something narrower than the promise. Enterprise adoption of a new trust layer is slow precisely because it's a trust layer: the sales cycle for "let us sit in the path of every agent action" is long. There may be a longer window than I think for hand-built governance work, especially with smaller clients who'll never buy Arcade.
But the direction is hard to argue with, even if the timing is fuzzy. Betting your business on the timing being slow is the kind of bet that's comfortable right up until it isn't. I'd rather position for where it's going.
What I'd actually do
If "agent auth and permissions" is currently the core of your pitch, change the pitch this quarter, not because it's worthless today, but because its shelf life is now visibly finite. Move your positioning up a layer: sell the engagement where you sit with a client, map which autonomous actions are acceptable versus dangerous in their actual business, set the thresholds and the human gates, and own that decision with them. Use Arcade or whatever wins as the enforcement layer underneath. Let the funded platform do the plumbing you were going to charge for, and charge for the judgment it can't have.
The tell that you're on the right side of this is simple: if a competitor could replace your deliverable with a config screen, you're selling plumbing. If replacing you means the client has to make a risk decision they're scared to make alone, you're selling the thing that lasts.
Author
Lukas
@lukcombinator