New episode: I sat down with Greg Pstrucha (Sentry, ex-Subroutine) on practical AI. Agent skills that can quietly own your machine, why this moment rhymes with the early web, and where the determinism line actually sits.
➡️ https://t.co/j1FYkYiXof
Nobody gets upset when the computer tells you you're wrong.
The same note in code review stings. As a red squiggle on save, you just fix it. That's the case for shifting left, and what Taskless automates.
Jakob Heuser and the Rise of Taskless https://t.co/4Sn1s3uMjj
Between wrangler deploy inconsistencies, container build issues, and a lot of rough edges, here's what it took to get Taskless' runtime rules sandboxing out:
https://t.co/PZ0KrlbVyA
A bit of a trope, that your agents "know nothing." JP had a good take: treat your agent like a new hire and you will both fix your employee onboarding and give your agent a better shot. https://t.co/3jaeTLTLV5
Take any claim and actively weave it into everything you already know. That's the promise of the Epistemic Graph. Our interview with JP is live: context is not one-off instructions, it's a living thing. https://t.co/5hUVgCFXQD
I went on Unfinished Business to talk Taskless, determinism in AI, and the everyday tools I use to manage the chaos as agents get more capable.
Listen: Jakob Heuser and the Rise of Taskless https://t.co/4Sn1s3uMjj
Let your Twitch chat run the music.
Music to My Ears wires Spotify into Firebot: viewers request songs, you control playback, and it handles search, moderation, and queueing.
Spotify Premium required.
https://t.co/Am9ia4YALG
"Context Rules Everything" - JP. You want one of those AI "Sales Development Reps" that you see on billboards? Well, where is that AI getting its information? Does it match what your engineers know?
JP argues that this context is always the same. https://t.co/gvbtgRBppL
Context is the missing primitive in the AI stack, and most teams are underestimating it. Episode 2 of The Task at Hand: JP, the person behind Wire, on why context is needed in more places than just your coding agent. https://t.co/y8fQD2FmcQ
I grow tired of performative clauding. Why are you so proud of using up all your tokens? Why are you obsessed with your 100-subagent parallel workflow? Tell me what you built, then tell me how Claude made it easier or harder.
Every podcast episode comes with its own blog post plus video, audio, transcripts, and links. If you're more of a reader, go straight to why context matters so much to AI, and how we're getting it wrong. https://t.co/ex0TlDMDqJ
Ah sharing context. The one thing Claude, ChatGPT, and every other frontier model will never do. We had JP on the Task at Hand this month, and he shares why that's a problem worth solving: https://t.co/veQvtC79Ce
@liran_tal@zeeg FWIW, 0.9.x of the Taskless CLI can help you write better eslint rules too. We're just out here trying to give everyone a little bit more predictable output from these non-deterministic systems. <3