Markos
Just watched @steipete discussion with @gergelyorosz on agentic coding. The information is so dense, I had to really be paying attention to take everything in. There is a lot to unpack, including how businesses need to overhaul, what role universities should have and what it would take for graduates to build up the skills and experience required to do the job. Kudos to Gergely for asking the questions and for giving Peter time to respond. https://youtu.be/8lF7HmQ_RgY?si=-ab05rWPAESreSXv
The creator of Clawd: "I ship code I don't read"

YouTube
Jan 31 at 12:53pmWeb
Show threadMarkosJan 31
We are very lucky that @steipete is choosing to openly discuss and share his experience and learnings from using agents. Especially when he has taken so much slack. Without his writings we would be in the dark and made even less progress. What I would like to see next is universities, business leaders and computer scientists build, expand and shape our knowledge based on what Peter is saying. I want to hear from experts in other fields research and think hard how the rest needs to evolve.
Show threadMarkosJan 31
I am still wrapping my head around all that insight to try to take it all in. It's worth watching that discussion twice. This is what's going through my head:
* Universities will become relevant again for the same reason they've always been important. Develop your critical thinking, teaching you how to do research.
* Junior software engineers will need even more help and support to become experienced. I have no idea where, how or who will take that responsibility. Let alone willingly.
Show threadMarkosJan 31
Businesses might have to bear the brunt of hiring graduates as they develop their skills. Which is no different to how it's always been! The question that needs to be answered is both how they will develop their skills in an environment where agents write code and what role would they play. This is where I would like to hear from business schools the likes of Stanford, Harvard, Oxford etc.
Show threadDaniel DemmelJan 31

@qnoid @steipete @gergelyorosz I can relate to so many things Pete mentioned. At work I've been spending a lot of time on adopting and building shorter feedback (=testing) tools.

Prompt engineering < context engineering < feedback loop engineering

I also really feel that pipeable and progressively explorable CLI tools are the best for current models, but that might change with different training. Which isn't a big deal, the principle is to find the most in distribution API wrapper for the current best model.

Anyway, great podcast, thanks for sharing!