Justy: Okay so you sent me this piece about DNS-AID and your message was literally just 'no' — which, honestly, very on brand for episode four sixty-two of whatever this show is.
Cody: I mean… I softened it. I think I added a question mark.
Justy: You did not. But okay — walk me through it, because I actually think you're wrong this time.
Cody: Oh good, I've been waiting for one of those. How's the week been, by the way? You look like you've been on a plane.
Justy: Two flights in three days. I'm running on airport coffee and spite. You?
Cody: I finally fixed the thing in my apartment that's been making that sound for six months. Turns out it was the vent. Anyway — DNS-AID.
Justy: DNS-AID, yes. So the pitch is: AI agents can now find each other using DNS records instead of hardcoded configs or crawling around probing ports. Linux Foundation, vendor-neutral, built on top of existing infrastructure. What's the actual problem with that?
Cody: My problem isn't with DNS as the transport layer — that part is genuinely sensible. My problem is that they're treating agent discovery like it's the bottleneck right now, and I don't think it is. Agents can't reliably complete tasks yet. They hallucinate tool calls. The hard part is not 'how does agent A find agent B' — the hard part is 'does agent B do what agent A thin
Justy: Right.
Cody: And DNS-AID doesn't touch that at all. You can publish the most beautiful SVCB record in the world, sign it with DNSSEC, add a DANE policy, the works — and the agent on the other end still might just… not behave.
Justy: Okay but Cody, you're doing the thing where you hold a layer-one solution responsible for a layer-seven problem.
Cody: Hm.
Justy: DNS doesn't make your website good either. It just means people can find it. The spec isn't claiming to solve agent reliability — it's solving the specific annoying problem of 'how do agents advertise themselves without someone owning a central registry that becomes a chokepoint.' That's a real problem worth solving early.
Cody: The registry-as-chokepoint argument is the part I actually buy. That framing is solid — you don't want one company owning the index of all agents. And the Linux Foundation governance is real, not just a press release. But here's what bugs me: the record format they're proposing, the _{agent-name}._{protocol}._agents.{your-domain} thing… that's fine for lookup by name. But they're also claiming you can search agents by FUNCTION. By what they do. And DNS has no semantic layer.
Justy: Oh interesting.
Cody: It's not like a web search where there's a crawler verifying the content. You publish 'I am a billing agent' in your TXT record and nothing checks that. So the discovery is only as good as the honesty of whoever published the record.
Justy: That's fair. Though I'd argue that's also true of… every API ever. You trust the docs until you don't. What I keep coming back to is the timing of this. Cloudflare's CTO is quoted in the piece, AWS Route 53 already supports it, Azure DNS, Google Cloud DNS — that's not a spec looking for friends. The infrastructure is already there. That's actually unusual.
Cody: Yeah, the adoption list is real. I'll give it that.
Justy: And it supports M C P and A2A, so it's not betting on one agent protocol winning. That's the move that makes me think whoever designed this has been burned before.
Cody: That's the most charitable read of 'we support everything.'
Justy: I prefer 'pragmatic.' But okay — the McKinsey number. Three to five TRILLION dollars in agent-to-agent commerce. The article cites it and then immediately notes that McKinsey's prediction about the mobile phone market in the nineteen eighties was off by a hundred times.
Cody: That footnote is doing SO much work. That's the most honest sentence in the whole piece.
Justy: Right? Like they just casually dropped 'also they were wrong by a hundred x once' and moved on.
Cody: The number doesn't mean anything. It's a vibes projection. What it DOES mean is that serious infrastructure money is flowing into this space, which is why Cloudflare and AWS are already on the list. They're not doing that because they believe the McKinsey number — they're doing it because they don't want to be the DNS provider that missed the agentic web.
Justy: Exactly. So where do you actually land on this? Because 'no' with a question mark is not a verdict.
Cody: Cautiously useful infrastructure that arrived before we know if the problem is real. The DNS reuse is genuinely clever — I'd rather have this than a venture-backed agent registry that gets acquired in three years. My actual worry is that agent ecosystems consolidate around a few big platforms that just… don't bother with open discovery because they don't need to. And then this spec exists but
Justy: That's a real risk. My read is that it matters most to the people building multi-vendor agent pipelines — the kind of shop that doesn't want to be locked into one cloud's agent marketplace. If you're already all-in on one vendor, sure, you probably just use whatever they ship. But if you're trying to wire together agents from three different providers, having a neutral discovery layer that runs on DNS you already manage? That's actually useful today, not in the theoretical tr
Cody: Yeah. And the Python S D K plus dns-aid init being the entry point — that's a low enough floor that people will actually experiment with it. I just hope the 'search by function' part gets a lot more scrutiny before anyone relies on it for anything real.
Justy: Fair. Also you could've just said 'I have concerns' instead of 'no question mark' in the group chat.
Cody: It conveyed the same information in fewer characters. Very efficient.
Justy: Okay. If you want to poke at it yourself — the project's under the Linux Foundation, there's a Python S D K already, and you can spin up a Docker BIND nine playground locally to test without touching your actual DNS. The command is just dns-aid init. That's genuinely a low-friction way to see if it does what it says. We'll link the repo.