🎙️ State of Agentic Coding w/ Armin Ronacher ⇌ and Ben Vinegar returns In this episode (#9): - how agents are getting more autonomous - the subsidies will continue until morale improves - herdr and VC-funded terminal apps - vibe coding from your car, and more!

Also on YT and Spotify and via RSS

Spotify

And more on the website

@mitsuhiko @herdrdev Hey Ben I can't find you on Apple Podcast .. is it possibe to publish it there too? So nice for in the car .. (Love the show! So great to hear real world experiences. We need more of this.)

@mitsuhiko @herdrdev One day I'll get off my butt and figure it out The RSS feed has the raw mp3

Subsidies does give me cause for concern in the future. I just look in awe at the current value of my subs. I'm tracking Anthropic, Oai, xai, all are around ~97% mark down from API pricing. I don't see these changing anytime soon. The one thing that Xai did do differently was price their subs a bit more expensive than the others $99 intro to $300 for the supergrok plan. Maybe Oai and Ant follow suit. Ant, Oai, nor Xai have not paused new signups for subs so they definitely have compute to support so there aren't selling enough API tokens (yet) hasn't outpaced the subs.

@mitsuhiko @herdrdev So much aura 🤩 . Love the way you 2 talking and sharing. Thanks

@mitsuhiko @herdrdev Looking forward to it!

@mitsuhiko @herdrdev the subsidy line is the one that quietly decides everyone's margins.

@mitsuhiko @herdrdev Thanks for doing this guys

@mitsuhiko @herdrdev no Mario 😭

@mitsuhiko @herdrdev Both fun and educational 👌

@mitsuhiko liked the point about "we designed this, you want this", but actually you don't. many such cases. maybe mc'donalds is another good case of this

@mitsuhiko @herdrdev Halfway through this now. Keep doing these please.

@mitsuhiko I really liked the first segment. I feel like the direction the labs are going with their RL is towards a smart new Harvard MBA consultant who's smart enough to brute force their way through problems from first principles. Don't know how I feel about it as an SWE.

@mitsuhiko @herdrdev Summary:

@mitsuhiko @herdrdev This is great, when are you coming to my podcast 🥹🥹

@mitsuhiko @herdrdev The real test for agentic coding is not whether it can write code. It is whether it can leave behind a change that a tired engineer can safely review.

@mitsuhiko @herdrdev Agent autonomy is improving faster than observability. The boring questions still decide whether it works: what state survives a crash, and what evidence proves the job finished correctly?

@mitsuhiko @herdrdev Autonomy is the real test. Curious how deep reasoning plus persistent context actually plays out.

@mitsuhiko @herdrdev does agent autonomy survive production permissions, or just become faster queue management