The internet did not win because anyone built a faster network. It won because a handful of deliberately boring conventions were agreed to once and then never had to be thought about again. IP gave every machine an address. TCP made delivery reliable across a network that guaranteed nothing. HTTP supplied a shared grammar for requesting and returning documents. DNS translated names a person could remember into the numbers a machine could route to, and certificates layered on top of DNS made it safe to enter a credit card number, because a browser could verify that the site answering to a name was actually run by the company that owned it. None of these were the interesting part of the internet. They were the part that allowed the interesting parts to be built by millions of people who never had to coordinate with one another, and the agentic web is now discovering that it is missing an equivalent.
The agent stack has moved quickly on the upper layers and barely at all on the lower ones. Anthropic's Model Context Protocol standardized how a model reaches a tool or data source. Google's Agent2Agent protocol, donated to the Linux Foundation in mid-2025, standardized how two agents exchange structured messages across vendors. Both are grammars. Neither answers the question DNS answered forty years ago: given a name, who is this, who operates it, what is it permitted to do, and where do I reach it. Agents today find each other through hardcoded URLs and proprietary vendor registries, which is functionally the pre-DNS hosts file, distributed by hand and trusted on faith. That works at the scale of a pilot and fails at the scale of production, which is precisely where enterprises are now arriving.
The Agent Name Service is the attempt to close that gap. It originated in the OWASP GenAI Security Project, where Vineeth Sai Narajala and Ken Huang published a DNS-inspired architecture for verifiable agent identity in May 2025, and on June 23, 2026, the Linux Foundation announced its intent to launch ANS as an open standard with backing from Cloudflare, GoDaddy, Infoblox, Salesforce, and Cisco. What matters is what ANS declines to build. There is no new lookup network and no central registry. It anchors agent identity directly into DNS, the infrastructure that already handles more than a hundred million queries per second and that no single company owns, so an enterprise publishes its agents under domains it already controls. Public key infrastructure provides verification, allowing a counterparty to confirm which organization an agent represents, what permissions it holds, and whether its code and operational history remain unaltered. A companion Linux Foundation project, DNS-AID, launched a month earlier, built on work by Infoblox, and handles discovery for both agents and MCP servers. The two compose in roughly the way DNS and certificates compose today.
The strategic question underneath this is the same one that governs the openness spectrum in models themselves (https://alten.capital/blog/models-closed-open-weight-open-source). Every platform shift offers a choice between walled gardens and shared infrastructure, and it is usually decided before most participants realize a choice is being made. If agent identity ends up owned by three hyperscalers, the agentic web becomes a set of federated silos with toll booths at the boundaries. If it ends up in DNS, it becomes an open substrate anyone can build on, which is the outcome that makes AI genuinely operate as an operating system rather than as a set of vendor products (https://alten.capital/blog/ai-as-an-operating-system).
For technology services firms, the practical consequence is that agent deployment work becomes portable. Today, identity, permissioning, and discovery plumbing are rebuilt for every engagement because each client's stack resolves those questions differently. A shared naming and trust layer converts that plumbing from bespoke work into assumed infrastructure, which lets a delivery organization's accumulated assets compound across clients rather than being rewritten for each one. That is the mechanism by which outcome-based delivery actually produces software-like economics (https://alten.capital/blog/from-time-to-outcomes) rather than merely being priced that way. ANS is still an intent to launch rather than a standard in place, and many of these efforts never achieve ubiquity. But the ones that do tend to look exactly like this: unglamorous, built on something that already works, and invisible the moment they succeed.
Alten Capital invests in exceptional management teams to accelerate high-growth technology services businesses. Reach out to explore partnership opportunities.