Alan Pinstein

@andrewchen I’ve tried both on projects. If I yolo it, the AI spins a lot more and it takes me tons of prompts to the desired result. I’d love to know if the yolo teams have found a real solution to this or if they just have problems they aren’t aware of.

^ this is about 40/60 with the majority doing yolo- about what I expected! Obv X (and my readership) are a tech forward group

One question I've been asking founders is: do you try to review all the code that the LLMs write or do you just accept it? I think it's about 50-50 right now but the momentum is towards just accepting the AI-generated code and I think that number will eventually go to 100% This is one of the most telling indications of how AI-native a team is. It's hard to get super high throughput if you are reviewing every line Poll: what do you do?

@thdxr I’m stuck bw doing real PR’s and yolo’ing (never look at the code). Greenfield apps. So far, reviewed apps are more manageable. Using Reveiwable to make PR easier, helps a ton. Made AI skill for threaded PR cleanup, has been very effective.

sent this to the team today everything great comes from being able to delay gratification for as long as possible and it feels like we're collectively losing our ability to do that

@atmoio I was able to help a biotech startup a ton w no industry other than a 30-year old bio degree. You do have to get comfortable w different work. But, trying growing all your food for a season and see if you’re ok going back to grocery stores :)

@atmoio On point. Been going thru it since late ‘22, as a manager, executive and coder, in reverse order. I am a builder at heart. All knowledge jobs are forever changed. I find value in focusing on societal needs. Fixing them (even w AI code) is fulfilling. Easy now to switch industry.

I was a 10x engineer. Now I'm useless.

@SeeBQ The random rewriting of unrelated code was common. I specially address this in my agent guidance but I bet the models have gotten better as well. I have also taught mine to read and address line-threaded pr comments which is very helpful.

@apinstein None of that yet…. I had an old Claude.md but it rewrote the whole thing once and I gave up. And not sure if GPT-codex uses that one. I am also hearing I need to give my agents a soul.md and a memory.md

Yes, the LLM coding agents are getting better. But still having to really coax them to get a quality PR that a good dev team would accept. "still got a scrollbar on the right, can you just remove it?" "why does it still have a scrollbar? start over on design and don't use it." Is it just me?

@arvidkahl hah yeah I've been joking that what used to take bad developers 6-12 months to turn into an un-touchable hairball, AI can now do that in about 7-10 days.

Just like regular 0.8x developers, but this time much faster, unresolved technical debt will kill first any real improvements to the product, then morale, then the business.

@ggraham @CompletedStreet Interesting, I don’t think I’ve heard you explain it like that before. I like it. Mis-priced or mis-captured externalities are a huge problem. So you’d prefer cars pay a mileage based tax over tavt for instance?

Hard for me to answer that. My concerns about auto-dependency and infrastructure mainly stem from costs being socialized. I think that as long as we perpetuate a dynamic where the farther a vehicle travels the greater subsidy it receives, things will keep getting worse, regardless of whether vehicles are autonomous or not. My view is the culprit is the externalization of cost, not autonomousness or even the car itself.

Self driving vehicles will replace transit just like Uber replaced transit.

@ggraham @CompletedStreet Do you think broad AV (cars and vans and buses, it will all happen) hurts or help wrt your concerns on roads, infrastructure, etc?

I don’t entirely agree with myself from ten years ago, but I still mostly do. One of the areas of agreement between Now Me and Past Me is that self-driving vehicles will radically transform mass transit. I think the robotaxi is a step in the direction of the rickshaws described herein, and when the form/mass of autonomous urban transportation gets much smaller/lighter (ie golf cart rather than car), I think mass transit is *really* gonna change. geoffgraham.substack.com/p/driverless-c…

@KyleSamani @rubenharris Everybody is going to switch to usage based pricing for agents. However, there are very few companies that have returned to growth from usage based pricing. I expected it sooner. Hoping it was lack of agents. CO’s have this year to prove it out.

This specific paragraph really stands out to me Agents will do an increasing amount of non-deterministic work, which will increase demand for deterministic tools that already exist The traditional SaaS that benefit will be those that have usage based pricing. The losers will be companies that have per-seat pricing

@addyosmani This is a problem but I think education supported by AI addresses it. The junior pipeline will move to education and self-started projects. My son is doing robotics and mostly hand codes c/c++ because he enjoys it. AI helps him learn faster. The best programmers did this already

AI amplifies senior expertise but may starve the junior pipeline that creates it. That is, if we only optimize for today's productivity vs. the long-term. The muscle memory of fields like software engineering develops through repetition and guided practice. Knowing when to trust your instincts. Recognizing antipatterns. Understanding second-order consequences of technical decisions. AI agents can generate code, but can't transfer the tacit knowledge that separates someone who can review AI output from someone who can architect systems. What worries me is if we skip the "10,000 hours of practice" phase and jump straight to "overseer of AI output," are we actually training architects? Or are we training people who don't know what they don't know? The industry keeps saying "juniors will do different work now." but there isn't yet alignment about what that work actually is, how it builds toward senior capabilities, or whether it creates the judgment AI assistance assumes you already have. Maybe the resolution isn't either/or. Maybe it's hybrid pair-programming where juniors work alongside AI but with better senior oversight and deliberate skill-building, not just task completion. You could call it trio-programming. Seniors who see mentorship as force multiplication but not a tax on their productivity. If companies optimize purely for cost-per-line-of-code today, we'll pay for it in a leadership vacuum in the coming years. I hope we're more mindful of the future than that.

Ranked choice voting will push Congress to work together for all voters. Ask your members of Congress to support the RCV Act today! #RankedChoiceVoting fairvoteaction.quorum.us/campaign/RCVAc…

@DevinOlsenn Looking at the screen, FSD saw the spinning car, was already path-planning left in lane and slowing down when you took over (which was just before it was obvious the spinning car was going to end up in your lane). Pretty sure it'd have slammed the brakes momentarily....

FSD V14.1.7 | Near Accident This is without a doubt the closest I have come to getting into an accident in a long time. I think had I not taken over and slammed on the brakes we would not have slowed down in time. You can see that FSD does an incredible job reacting to the car (even before I noticed it) and was slowing down, but it was not slowing down enough to stop in time. As soon as I saw the car swerving to the right I worried she was going to bounce back into my lane so that is when I decided to go full on the brakes to stop the car. You can see prior to me taking over we were full regen braking (which is FSD reacting well) but it wasn't using any real brakes at all, and ABS was not active until I took over. Let me know what you all think in the comments. I still think FSD is incredible - the reality is if you watch the video the car saw and reacted to the situation before I did, it just didn't brake as agressively as it needed to in this situation. FSD + an attentive driver is without a doubt in my mind still safer than a normal human driver who is not using FSD.

@ggraham Haha you missed out, I got Intergenerational Mortgage™️!

I just trademarked both Century Mortgage™ and Millennium Mortgage™.

@blader I love hot sauce and startups, this is some interesting marketing but I’m in for it.

while supplies last at burnrate.cash

so we raised a new round, but haven't told anyone yet our investors are like, hey put our money to use how come burn isn't going up i gotchu fam introducing burn rate by runway an award winning, blind taste-tested (by me!) hot sauce, wrapped in a genuine $100 bill for $13.99

@ggraham My strategy is just to spread an entire packet of mixed seeds into the bed and let nature go :)

@apinstein Pic?

CNC machines and 3d printers are tools for conviviality.

@ggraham This was from march 2024. Don’t have a better pic.

@ggraham Lexan (double wall) was surprisingly expensive but worked so well. I had happy lettuce under snow-covered roof last year!

@ggraham Wow this is a cool project. I bought a 4’x8’ lexan last year as a small test to build a roof on my raised bed. Worked amazingly well. My bed is 3’ wide so I just folded it along a seam to make a roof :) This should be epic!

@ggraham Also just got back from Europe and seeing the history in museums had an AHA moment. I used to think the Europeans were “behind” the US on markets and capitalism; but was reminded they are actually probably 100’s of years ahead of us.

The One Ring is Central Banking. I am pretty sure Tolkien didn’t intend this, but here me out: - Central banking enabled the perpetuation and amplification of the Great War’s inhumanity (which inspired him to write LotR); - It enabled us to finance the creation of the atomic bomb and other grand abominations that either threaten humanity’s existence or atomize society; - One of those examples is the interstate highway system which enabled the “Scouring of the Shire” (the destruction of rural communities); - It enables rulers to drain the life force from the future by encumbering the unborn with debt (which finances wars, bombs, highways, etc); - Via “Zero Interest Rate Policy”, it reduced the cost of capital for ambitious institutional investors so that they could create AI; and - When within the grasp of a ruler, virtually none can resist the temptation. They believe they can do good with it, and think that when it has served its purpose, they will put it away rather than become its servant.