WeChat ID
aaronpk_tv
#AI
-
Steven
https://pi.swb72.us/
•
Oct 30
I am loving the updates that Aaron Parecki has been making to the Overland tracking app as a beta tester via Apple TestFlight. This is a good case study for using AI code tools alongside existing legacy code to add new features and clean up potential holes.
-
Can now use automations in Shortcuts to, for example, auto-start a car trip when my iPhone connects to CarPlay (I don't have one set to end the trip when it disconnects, because CarPlay sometimes disconnects and reconnects, and the pause is just long enough to convince Overland to end the trip). I'm looking for other ways to automate things like starting walking trips when starting an Outdoor Walk workout (or vice versa), &c.
-
Screen widget lets me see an active trip and end it from the home screen without needing to open the app.
-
Additional trip types, including running, e-bike, hiking, &c. (Although I somewhat jokingly suggest there should be one for amusement park ride—Disney World frequent flyer).
And to the idiot who saw fit to criticize the use of an AI tool to help write code—you're certianly welcome to go back to writing Assembler code if you like, have fun with it.
Thank you! It's been super freeing to work on it this way!
If it wasn't obvious before, I am definitely not good at iOS development, and I often dreaded needing to make updates when Apple added new APIs, so most of the time things would just never get done and it would be years between app updates.
And frankly, most of the recent iOS platform integrations are not terribly creative, they involve following Apple's docs on what to do, which it turns out is a great task to outsource to the giant statistics machines.
I hope it's clear that I am still driving the development of this with my Product hat on, and now I'm able to move much faster! This lets me focus on the fun part to me, which is to make the decisions about how the app works, and not have to get into the weeds on how Apple wants me to write the code this year. -
-
I do not use AI at all and it stole my work and fucked up my career but!
I don't think we need the same amount of hate and vitriol to the one person who disclosed that they vibecoded a tiny part of their website as we give Sam, Dario, and Zuck
If we want ppl to disclose AI use (we do) this is not a good strategy for small servers/communities. Clear community rules about disclosure or what's allowed help, dog-piling does not
-
Why I'm rebuilding my IndieWeb projects with Claude
I recently came across Brennan's post noticing that Claude Code has been showing up in the commit history of a bunch of my IndieWeb projects lately. That's true! I was wondering when someone would notice. I haven't been hiding it, but I also haven't written anything about why I'm doing it, so this seemed like a good time to finally get my thoughts down.continue reading... -
Slowly tweaking a local trained LoRA model to generate stylized images of my cats doing seasonally-appropriate activities. It takes only a few seconds to generate these on the GPU in the closet. Still have some work to do on the poses and image styles. But I'm pretty happy with the progress so far!
-
Started building a local LLM server today. An NVIDIA RTX 3090 24gb in an external USB-4 dock connected to a mini pc with 64gb ram. Now the fun begins!
-
ok I am impressed, only took a few iterations and Claude built this entirely in the browser. My tweaks were things like "make the legs inset from the face, I don't want to print with supports." It made the STL and I'm printing it out now! 😍
-
Today is the day, MCP is launching the next version of the spec! The 2026-07-28 version brings a bunch of exciting changes like statelessness, but I'm focused on the boring auth side of things that nobody wants to deal with!
• Issuer validation (RFC 9207)
• Refresh token and offline_access recommendations
• Scope clarifications
• Syncing with RFC 8414 .well-known URI
In the mean time, there's been a ton of excitement around MCP's Enterprise-Managed Auth extension, with a beta program running in Claude and Okta and new MCP servers shipping support by the day!
Join me at the MCP spec release party in London tonight to learn about what's new in enterprise MCP security!
https://luma.com/klu2tegz -
Cross-Domain API Access: Beyond the "Obvious" Shortcuts
Cross-domain access is everywhere in today's software landscape. Whether you look at enterprise SaaS applications, AI agents interacting with user data across multiple platforms, or "integrated experiences" pulling information from a calendar, a chat tool, and a wiki—everything eventually needs to talk across boundaries.continue reading... -
The "Agent Verified" signup flow from WorkOS is exactly what I've been telling the agent platforms they should be doing with Cross App Access! Very cool to see this launch! 👏
https://workos.com/auth-md/docs/flows/verified
"The agent's provider — OpenAI, Anthropic, Cursor, or any trusted agent platform — attests to the user's identity at registration time. Your service verifies the attestation and issues credentials synchronously, no human interaction required."
In Cross App Access terms:
• The "agent platform/provider" is the ID-JAG issuer, because users are already signed in to those platforms when they use agents
• The "service" is the ID-JAG consumer (the Resource AS), and issues an access token if the ID-JAG is trusted and valid
You can test this out in the Cross App Access sandbox today! https://xaa.dev/
