Charles Pan

@jacksondecampos This needs to be built in

When the agents are driving me up the wall

@nikunj 3 Flash is great, so underrated

It’s criminal how cheap and how good Gemini Flash is.. that too with 1M context windows and structured outputs. Probably, my most used model in production workloads. Separately their new live voice model is mindblowingly good if you haven’t tried it yet!

@housecor It should be what we’re all used to reading - plain English

Imagine a future where we mostly stop reading AI generated code and primarily focus on behavior. What programming language should the AI generate? How do you decide?

@AI__Angelo Agreed. I think we’re at the stage where we’re figuring out where exactly in the loop humans belong

There’s a real point in what you’re saying, but “AI slop” and “YOLOing everything” also flattens what’s actually going on in production systems. Most serious deployments were never truly human-out-of-the-loop in the first place. Even when models generate outputs autonomously, there are usually layers around them: evaluation, monitoring, guardrails, rollback paths, and human accountability somewhere in the system design. The difference now is that teams are figuring out where humans matter most, not whether they matter at all.

I’m glad that people are starting to wake up to the fact that for real production systems, humans need to be in the loop to some extent and not just YOLOing everything with AI slop

I wish I had a model that was post trained exclusively on clean code

@tomhaerter What would you change? I think that git scales with number of contributors pretty well but could be convinced otherwise

everyone: let’s redesign GitHub nobody: let’s reinvent git for agents

@johncrickett A Philosophy of Software Design (Ousterhout) Was lucky to get to know him by taking his class as well!

5 books that shaped how I build software: 1. Expert C Programming (van der Linden) 2. Design Patterns (Gamma et al.) 3. Extreme Programming (Beck) 4. The Pragmatic Programmer (Hunt & Thomas) 5. The Practice of Programming (Kernighan & Pike) Each one changed something specific about how I write code, approach projects, or think about the craft. What are yours?

@sahill_og What does better mean though? Better architecture has always been subjective, dependent on context

@mattpocockuk Request changes on their code, talk to them in person

What do you do if someone on your team is using AI negligently? I.e. not reviewing, not caring, leaning into the slop. This, of course, was a problem pre-AI. But the "code is cheap" mind virus is making it worse IMO.

@ItsAlexhere0 How are we defining better

Software engineers, what’s your plan B if AI writes better code than you in the next 3 years?

@binsquares @ian_dot_so @DeanStratakos It’s dying but we’re trying to bring it back!

@ian_dot_so @ceefryingpan @DeanStratakos what do you guys think?

Is code review dying/dead? Fill me in on what you’re seeing below

@mattpocockuk This is great For the subagents one I’d also add a note about being able to do work in parallel And maybe should add something about MCP?

@icanvardar It’s difficult for not just people with zero software experience, but also people with experience but working in a stack they’re unfamiliar with The models are just so blindly confident that it’s hard to consistently push back on them

people with zero software expertise are starting to rely on whatever these agents produce as if it is always correct. no verification, no second thought, just accepting outputs as truth. that level of blind trust in generated code is going to fail in subtle but painful ways

@Liggi Yes absolutely! Will dm

@ceefryingpan hey man, we're trying out Stage. I've invited a colleague, but they can't seem to accept. It says "this invitation belongs to a different account". We tried cancelling and resending, no luck. Can you help?

@peer_rich code reviews are still how top eng teams stay in sync

we still do code reviews its not used as gatekeeping but rather as mentorship the best teams use reviews to transfer context over simply catching bugs one good comment can save months of future work the worst code reviews are the ones that only ask if something works the best ones ask if the next person who touches the code will understand why it works those are very different questions and the second one is what compounds over time senior engineers reviewing junior work have a choice leave a correction or leave an explanation corrections fix the present explanations change how someone thinks about the next 10 decisions they make without supervision context loss is one of the biggest invisible costs in growing engineering teams every time someone ships something they dont understand the team becomes slightly more fragile code reviews when done well are the antidote they spread the mental model before someone goes on vacation or leaves the company a review culture that feels safe enough to ask why we did it certain ways is worth more than the fanciest ci pipeline tooling catches syntax culture catches thinking you need both but only one of them truly compounds the way you want it to

@DeanStratakos Read my full post here: stagereview.app/blog/introduci…

On a personal note: I started a company! I'm super excited to be building with my longtime best friend @DeanStratakos I wrote a bit more below on why code review is dying and why we don't think it has to be this way 👇

@DeanStratakos @YuzuHealthInc We’re still in the early days but we’re super excited to keep building. If you’re drowning in code review and are also tired using GitHub, give Stage a try and tell us what you think: stagereview.app

@DeanStratakos @YuzuHealthInc We want Stage to be the best code review platform built for developers, by developers. We’re starting off by redesigning the code review process around the way you naturally think, allowing you to review PRs faster and with a better understanding of what’s actually changing.