the new bottleneck is no longer doing the work it’s reviewing the work
@jsngr In the age of agents the code review interface is the home screen for developers, not the IDE or the terminal.
@jsngr Which… is arguably harder. Understanding the work is almost as hard as doing the work. Only a few places where this will actually be beneficial at scale.
Thinking through what the best UX paradigms for this might be. What we talk internally about is that the two core activities are 1. Specify - Research, Plan, disambiguate & make decisions, gather all the needed context. (✨ ... implement ... ✨) 2. Verify - Understand, Document, quickly confirm the behavior meets what you specified, and the vibes feel right There's always a bit of an interesting loop where the verification step might reveal things you didn't think about beforehand, so keeping that iteration / discovery loop nimble seems important too
@jsngr There’s a good setup for that: beyang.org/how-i-configur…
@jsngr My Ai swarm says, it's the HITL : the Human in the loop ...is the biggest bottleneck ! 😳 x.com/ThisIsMeIn360V…
@jsngr My Ai swarm says, it's the HITL : the Human in the loop ...is the biggest bottleneck ! 😳 x.com/ThisIsMeIn360V…
@jsngr @greptileai has been helping this! Check it out. 😀
@jsngr 🎯 Great software will help users reduce the time to review.
@jsngr how long until AI is better at reviewing as well? that's the scary question 😅
@jsngr yep. we are all eng managers now
@jsngr Wait till you see what I’m cooking here. Working on Product Announcement video as we speak.
@jsngr It’s easy to underestimate how much time review eats especially when each PR grows bigger, turning quick feedback into multi-day delays. Does your team track review cycles or just rely on gut feel?
@jsngr If it is a low quality work (like what LLMs generate), then reviewing it is a waste of the reviewer's precious time, and certainly won't add to his productivity.
@jsngr Exactly. You must understand what AI is doing. AI is the crew / ship. You are the captain.
@jsngr Preach. The more you scale, the more true this becomes. Oversight becomes the new grind.
@jsngr If by “review” you mean redo 20% in a very annoying manner. Then yes, yes indeed.
@jsngr Verifier’s law: The ease of training AI to solve a task is proportional to how verifiable the task is. All tasks that are possible to solve and easy to verify will be solved by AI. jasonwei.net/blog/asymmetry…
@jsngr no joke, half of my projects are stuck because of this
@jsngr @silva_math_ the same as you said a few days ago
@jsngr But it takes 8X the amount of time to review than to just write it yourself , because you have a retarded not doing the work.
@jsngr Omg, yes! But it’s not new for me. At my last role, we had to put extra pressure on the review process. FinTech. This meant your PR won’t ever be merged until it’s reviewed. But this also meant we blocked ourselves often.
@jsngr That's why you need an AI Supervisor like Wayfound.ai, an AI Agent that reviews all the work of other AI agents real-time, sends alerts, and suggest how to corrects them.
@jsngr Which is probably why you could see more “senior” job postings out there … the replacement starts with juniors
@jsngr yeah I find myself spending a lot more time reviewing now than writing which is not too far from what I was doing before because I was reviewing people but now reviewing AI work so different because there's more volume