最近 X 上 AI 版 GTA 太火了！ 奶奶版、护士版、辣妹版都看过了。 那我肯定要来点不一样的 —「东方武侠游戏」全新体验 用 @itsPolloAI 玩转 Seedance 2.5， 纵马、拔剑、踏入江湖。 武侠开放世界，正式开服，兄弟们，上号！🔥

For the last year, whenever I met other founders, I was asked "How did coding agents change things for you". Obviously, it changed a lot internally. Coding, reviewing, fixing... But for me, the more interesting aspects is how the workloads on our system have changed. And what this means for distributed systems architecture. I've been chatting (and sometimes info-dumping) on this topic for months now and I'm excited to share all these ideas with QCon audience! This talk will be 60% stories of what the agents did to various systems, and 40% ideas on how to build better systems for agents.

This vision of (fake)Agile👇 drives me nuts. Just No. There are no waterfall tracks. I collect just enough information to start, not finish, the work, and I do that just before starting. I write my tests first and run them every few minutes (TDD). I deploy often—minutes, not hours or days (CI/CD). I get feedback (and additional requirements) as I work and adjust what I'm building based on that feedback. The whole team works on each story and does not split the work into parallel tasks (scatter/gather). The relationships between stories are a (mostly undirected) graph, not a hierarchy (epics). There is no central orchestrator—decisions are made by the team as a whole. There is nothing at all Agile about that diagram, though many people work that way. Small waterfalls are still waterfalls. This is a waterfall shop holding onto dysfunctional ways of working with a death grip.

Thank you! I am working on a deterministic algorithm to create different track types, I am refining it and adding a "difficulty" level from smooth to hell-no! 😂

Agency is rarely given. It’s taken, then recognized.

AI has collapsed the cost of building products, but not the cost of knowing what to build. - The old loop: spec, scope, build - The new loop: build, play, design, ship Prototypes are cheap. Great products still require taste, judgment, and a clear understanding of what users actually need. a16z's @joshelman on the artifact a product manager actually produces: a16z.news/p/product-manageme…

I realize now that some vibe coders are people who always wanted to dabble in writing code but jobs and life took their lives in different directions. But now, with AI, they can scratch that creative itch.

Having a singular focus on one goal is really powerful. I don't think I had the luxury of being able to even have one all defining goal since I was 28 or so when I was trying to become a software engineer. I had just quit my job as a PM at a hardware company, put my entire life on hold and just spent all my waking hour trying to learn this thing that had consumed my life. Things progress really fast when you just have one focused goal. No need to make plans, no distractions, no decision fatigue. What would have taken you years can really be condensed down to months. Just think about college and the core eng classes. There is probably like 8-10 that is required info you actually need for work and the rest is just nice to know or if you want to specialize. For one course ~3 hours of lecture per week ~15 weeks per semester ~45 hours of lecture total for the whole course ~360 hours of lecture time for the most important courses. Its not a lot. Yeah I know just listening to lectures is not the same as understanding but the point is that if you are focused you can get to your goal fast. Extremely fast.

if a coordinated slow down does not happen, will ai labs still voluntarily slow themselves down? that's pretty much the only question worth answering right now

If vibe coding is so great, where is the vibe coded Instagram competitor? Not a demo, a real app with real users. Months of this pitch and still nobody has shipped one.

serious question, at this point why not simply cancel the ipo’s & nationalize the labs? theoretically nationalization would solve some of the incentive problems immediately. the us treasury becomes the only shareholder. the govt can coordinate compute, model releases, security standards, & pacing without asking rival firms to collude. you could theoretically put openai, anthropic, & eventually other designated frontier labs under something analogous to a federally owned strategic technology corporation, with the president appointing the board rather than literally having sam & dario file weekly tps reports to the oval office.

Nobody said “vibe” coders are replacing senior engineers on a 12 year old enterprise platform. They’re building the other 90% of software:
a tracker for the shop
a family recipe app
a WordPress workflow that used to take a whole Saturday
a local agent that actually does one job You don’t need to know every framework convention to ship that.
You do need taste, persistence, and the ability to tell the model “that smells wrong, try again.” That’s the new class.
Not “fake developers.”
Just people who can now finish things they used to only describe.

new sf run club called “pacing the frontier”

how do you make workflows not suck as a team gets bigger it's just easier when people have access to things. but then there's risk with people having access to things i hate when people get blocked because they have to ask someone to do something, this is 10x worse now that we have agents

Please, stop pasting AI generated essays into PR descriptions 😫 Just write two simple bullet points so we can review the code and move on.

OMG why did nobody tell me about cmux? The ability to open a browser "within the terminal" is genius and it allows to automate small everyday's tasks