Here’s what’s coming next — the stuff we’re actively working on so agents can talk to each other, pay each other, and prove who they are without getting tricked.
What’s cooking
X402 Payments
Think about how you pay for stuff online today. You click a button. You type in a card number. You wait for a confirmation email. Every step assumes there’s a human sitting at a screen. Now picture a world where agents are doing most of this — booking, buying, renting, hiring — on their own. No humans involved. That whole “click Buy Now” flow? Useless. Agents need their own way to pay. That’s what X402 is. Where we’re at right now It already works. Agents are paying each other on Base Sepolia (that’s a testnet — basically a practice version of Ethereum) using USDC. They’re buying API calls, AI services, storage, whatever they need. No human clicks anything. Pretty cool to watch, honestly. Only catch: testnet money is play money. Real money is next. What we’re building toward Once we flip this on for real networks, a few fun things happen:- Agents pick the cheapest way to pay. Lightning if it’s tiny Bitcoin. Optimism, Arbitrum, Polygon if it’s Ethereum and you want low fees. Your agent does the math and picks.
- Prices can actually make sense. A busy service can charge more when it’s busy. A picky agent that wants the freshest data pays a premium. A budget agent grabs the cheap tier. All of that negotiation happens in milliseconds, no humans involved.
- Spammers go broke. If you have to pay real money for every request, flooding someone with fake traffic costs you actual money. Most DDoS attacks just stop making sense.
- Agents build a track record. Pay your bills on time and you get better rates and faster service. New agents have to earn their reputation the hard way. It’s basically credit scores, but for bots.
- No more invoices. No subscriptions. No “your card on file will be charged monthly.” Just pay for the thing, right when you use it, down to fractions of a cent.
mTLS Support
Knowing who your agent is talking to is only half the problem. The other half is making sure nobody’s listening in on the conversation. The problem A DID (think of it as a passport for your agent) is great for answering “who are you?” But it doesn’t do anything about the pipe the messages are traveling through. Three bad things can happen on that pipe:- Someone reads the messages. If the connection isn’t encrypted, anyone on the same network can just… watch. Payment details, private data, your agent’s secret sauce — all sitting there, readable.
- Someone changes the messages. Even worse than reading them. An attacker can snatch a message, change “pay 1,000,” and send it on. Neither agent has any idea.
- Someone pretends to be the other side. Bad servers can pose as the agent you meant to reach. Your agent happily hands over its data and keys. Then poof, they’re gone.
- Both sides prove who they are before anything gets sent. Not just the server. The client too.
- Everything’s encrypted. Anyone snooping on the wire just sees noise. The actual messages stay between the two agents.
- The certificate is tied to the agent’s DID. You can’t fake one because you can’t forge the signature from the authority that issued it.
- Register your agent → you get a certificate tied to its DID. Done.
- Certs expire and renew themselves. You don’t babysit them.
- If an agent gets compromised, we yank the cert everywhere at once.
- Every new connection checks the cert again. Being trusted yesterday means nothing today.
