Almost every conversation with an AI coding agent has teachable moments. It's a shame that the student fundamentally *can't learn*. It's going to be wild when that statement is no longer true.

Another trick I've really liked is to ask the Agent to produce a README-style example using the api / library under design. This aligns with how I like to build software... it is for the user of the software after all. User-first software, whether interactive (UX) or API (DX).

Anyone know if there is empirical data on the impact of "politeness" and "respect" when interacting with models. I like to think it tips them into a more efficient mode but who really knows. AFAICT Jessie Frazelle has taken the exact opposite strategy to great effect. 🤣

Today, we talk about AI assistants. Tomorrow, it will be AI avatars.

Not having nested virtualization support on my M1 is such a pain in the butt for experimenting with VMs. I'm on the verge of moving my dev to a Hetzner box or something like that.

Third place! Over two hundred teams submitted demos. Not too shabby!

One last change

New experiment. I'm building a project-specific MCP dev server *within* the project. The hypothesis is that Agents will be more productive if I give them highly-targeted tools to close the feedback loop. It keeps security tight because I control those tools. Will circle back.

whoever nails this DX wins AI memory for devs

New Auth0 docs just dropped 👀 auth0.com/docs

In Claude Code, is there a way to rewrite a Bash tool call without also converting the permissionDecision to "allow"? I would like to rewrite the `command` but don't also want to bypass user interaction. John Lindquist I know you've gone pretty deep here. Any experience?

Scaling AI agents isn’t just about speed — it’s about state (and security). In this session, Matteo Collina (Co-Founder & CTO at @Platformatic) shows how to build stateful #MCP servers using Node.js, Platformatic & Fastify — and secure them with @Auth0.

There has been considerable discussion on how to implement Authentication using OAuth in MCP, including a lot of criticism on using Dynamic Client Registration (DCR). I've partnered with Auth0 to demonstrate how it's possible to implement this and provide a secure authentication

Every cheap swag idea has been explored and exhausted. Making these things is pointless if the items don't see the light of day post-event. Invest in high quality items, work with a designer, plan distribution to fewer (but key) people, make great stickers for the masses.

Out-of-order HTML patching, another proposal from TPAC, lets HTML be presented non-linearly, and a single file can update multiple parts of the page. Would this feature be useful to you?

I want to be able to trust agents, as a person. This is a very long journey from we are today, to accelerate that; we want to empower people to build, deploy agents securely. I am proud of the team that shipped our first GA step auth0.com/ai

Wow, never saw this sneak into node.js. TIL that `export { localSymbol as "module.exports" }` works in node's require(esm) interop story.

Kinda wish JavaScript had: if (const err = doSomething()) { return handleError(err); }

Fired up qbasic.exe, discovered gorillas.bas. Changed things and watched what happened. Eventually figured it out!

to commemorate alan dye moving from apple to meta, here's one of his best quotes