We scan new podcasts and send you the top 5 insights daily.
Mockapetris recalls writing the first DNS RFCs and getting numbers from his boss, who simply published them on a server for people to download. Today the IETF runs a heavier process with informational and standards-track RFCs, politics, and paperwork—even claiming DNS though it predates the IETF.
Mockapetris says DNS’s magic had two parts: it let millions of organizations own separate chunks of the namespace indefinitely, yet users experience one contiguous tree. When looking up a name, the first DNS query asks which name server to consult, not for an IP address.
Mockapetris copied the SRI host table into DNS every morning, so DNS held all existing names plus new DNS-only entries. Early adopters got both namespaces, while people who stayed on the old host table missed places where the “cool kids” hung out.
Mockapetris rejected Xerox PARC’s fixed three-level naming because users want freedom to organize information. He designed DNS so names could nest however many levels people needed, noting hidden applications often go many levels deep. If you allow nesting, he argues, support any reasonable amount.
Historically, the "law of the sea" governed ships in international waters. Today, the internet is the new global commons where data packets travel between "ports" (computers). The rules governing this flow are increasingly defined by code and protocol, creating a new digital legal framework.
Mockapetris explains www began as a convention for pointing to a domain’s web server, alongside names like ftp.stanford.edu for other services. As the web came to dominate, separate service machines faded. The convention stuck even though the original multi-service model mostly died.
Mockapetris calls DNS a possible privacy risk but says web advertising violates privacy far more effectively. He notes that when an ad loads, 150 or more organizations may be asked whether to serve it, using up to 144 characteristics about the person.
To ensure the AT Protocol remains open, Bluesky is handing over critical components to independent bodies. The core identity directory is moving to a Swiss nonprofit, and the protocol specification is being submitted to the IETF. This preemptively removes Bluesky as a single point of control or failure.
The AI space moves too quickly for slow, consensus-driven standards bodies like the IETF. MCP opted for a traditional open-source model with a small core maintainer group that makes final decisions. This hybrid of consensus and dictatorship enables the rapid iteration necessary to keep pace with AI advancements.
Mockapetris admits he does not quite understand the economic incentive for running root servers. Early roots were more distributed; now Cloudflare, Google, and other hyperscalers absorb more DNS traffic, partly for speed and prestige, making the system less distributed.
Instead of waiting for formal bodies, Google DeepMind is developing and open-sourcing its own technical standards for AI agents. This strategy aims to solve immediate interoperability problems and establish a market-wide de facto standard through rapid, widespread adoption, bypassing slower, formal channels.