AI revives classic math applets & Tools to inspect AI agents - Hacker News (Jul 12, 2026)

A Fields Medalist just used AI coding agents to resurrect web math tools from 1999 in a matter of hours, and the rewrite even exposed bugs in the original code. Welcome to The Automated Daily, hacker news edition. The podcast created by generative AI. I'm TrendTeller, and today is july-12th-2026. On today's show: AI helping mathematicians rebuild old software, new ways to understand what coding agents are actually doing, a big warning from Vint Cerf about AI standards, decentralized model serving, a tiny new image placeholder format, and a very determined trip back to the original Xbox.

AI revives classic math applets

We'll start with that Terence Tao story, because it's one of the clearest examples yet of AI changing technical work without replacing the expert. Tao says modern coding agents helped him port roughly two dozen old Java applets from 1999 into JavaScript in just a few hours, making them usable on the web again and even improving some of the visuals. More interestingly, the rewrite introduced only a minor bug and also uncovered issues in his original code. He then used the same approach to finally build a relativity visualization he had wanted for decades. The takeaway is pretty grounded: AI is proving useful for reviving educational software and adding interactive layers to research, even when the research itself still depends entirely on human insight.

Tools to inspect AI agents

Staying with AI agents, two separate posts pointed to the next question developers are asking: not just what agents can do, but how to understand and shape them. One project, called Mindwalk, turns agent session logs into a local 3D replay of a codebase, showing where an agent searched, what it read, and what it changed over time. That could make audits and debugging far easier than digging through raw logs. In a different experiment, a developer built a compact Common Lisp agent that can extend itself dynamically through eval, effectively letting the model create new tools as it works. Put those together, and you can see the direction of travel: more visibility into agent behavior, and more flexible systems that don't rely on a fixed tool list.

Vint Cerf on AI protocols

Another major AI angle came from Vint Cerf, one of the core architects of the internet, who is retiring from Google after more than 20 years. Cerf used a recent appearance to argue that the coming wave of AI agents will need formal interoperability standards if they're going to work reliably across different platforms. His concern is simple and important: natural language is fine for humans, but too ambiguous for dependable machine-to-machine coordination. When someone so closely associated with TCP/IP says AI needs its own protocol layer, that's worth paying attention to. It suggests the next big standards battle may be less about websites and apps, and more about how autonomous systems talk to each other safely and predictably.

Peer-to-peer GPUs for inference

That idea of a broader AI infrastructure shift also shows up in Mesh LLM, an open-source project aimed at decentralized inference. The pitch is straightforward: pool GPUs and memory across multiple machines, expose them through a single OpenAI-compatible API, and avoid relying entirely on a central cloud provider. In practice, that means a team could run models locally, send requests to whichever peer already has the right model loaded, or split a larger model across several boxes. Why it matters is not just performance. It's about cost control, data privacy, and giving smaller organizations a way to get more out of hardware they already own.

Tiny thumbnails with Handsum

On the web performance side, Nigel Tao introduced Handsum, a new format for tiny image placeholders. This is the sort of thing users may never notice directly, but they feel it when pages load more smoothly. Handsum is designed for extremely small previews with predictable byte sizes, giving developers a compact stand-in image before the full version arrives. The interesting part is the focus on practical constraints rather than maximum image quality. For websites and apps that need a fast visual hint without spending much bandwidth or storage, that kind of fixed-budget thumbnail can be surprisingly useful.

Modern engine meets original Xbox

And finally, a great piece of engineering archaeology: one developer documented the effort to port a custom game engine to the original Xbox. That meant peeling away modern graphics and language features, working through old toolchains, and repeatedly reshaping assets until they fit the limits of early-2000s hardware. It's easy to read that as a niche hobby project, but it highlights something bigger. Software preservation isn't only about saving binaries; it's also about preserving the knowledge needed to make old platforms understandable and usable. Projects like this keep that history alive in a form people can still learn from.

That's it for today's Hacker News edition. The big theme today was infrastructure around AI: better tools, better visibility, and a growing push for standards as these systems become more connected and more useful. Thanks for listening to The Automated Daily. Links to all the stories we covered can be found in the episode notes.

AI revives classic math applets & Tools to inspect AI agents - Hacker News (Jul 12, 2026)

Our Sponsors

Today's Hacker News Topics