Seattle3503

I'm lucky I guess, I feel like the product team is on "my side" with respect to building a resilient system. They understand that bad architecture shows up as a fragmented and friction-full UX.

The demand for new features come from sales and executive people who want to check more boxes on a customer's checklist.

My anecdotal experience has been that is is very reliable on inputs over 500 words. There are a number of third party evaluations that corroborate. Pangram is pretty easy to test for yourself if you are a skeptic.

How do they work after the last person leaves?

2025 called and it wants it's take on AI detection back.

https://www.pangram.com/blog/third-party-pangram-evals

Note that while that page is from Pangram, the studies on it are not. If they are excluding studies that show Pangram performs poorly in realistic conditions, please share those studies!

I was speaking of reinforcement learning with verifiable rewards. It's not clear to me from your comment why re-writing python couldn't be in the set of training tasks used during RLVR.

I wouldn't be surprised if "re-write Python" is in the training set.

The full quote implies more uncertainty

> So based on that, we believe what is most fair is to put current trends in because you cannot predict at what point consumers and demand is going to be -- you're going to begin to see demand destruction based on the pricing environment.

https://www.fool.com/earnings/call-transcripts/2026/04/23/so...

That might be true for some people, but I don't think parents who keep their children's art are doing it for the quality of the art.

I feel like I am coming in to the middle of a discussion. Towards what end?

> One wonders whether a generation that demands instant satisfaction of all its needs and instant solution of the world's problems will produce anything of lasting value. Such a generation, even when equipped with the most modern technology, will be essentially primitive — it will stand in awe of nature, and submit to the tutelage of medicine men.

- Eric Hoffer

What if the real issue is that all important software problems were truly solved decades ago, and we have been reheating the leftovers for years. Having a super intelligent AI reheat it for you doesn't make much difference.

I hope for a "shackled leviathan" as articulated by Daron Acemoglu and James A. Robinson in The Narrow Corridor.

Democracy is the only thing that can save us. But it isn't a given. In resource rich countries democracy struggles to take hold because human labor is minimally necessary and elites can mostly thrive without it. Many democracies survive, because without the people, everyone suffers. AI threatens to make all countries into petro-dicatorships.

I wonder if software will be built inside of companies that have non-tech competencies going forward. The story of 2000-2025 was every traditional company becoming a tech company. The next 25 years might see tech companies without any offering besides "tech" being at a disadvantage because their entire offering is easily replicated.

I think the title should have a "2025" in it.

I gave up on the article and watched the video

Could this be the opening China needs to get their foot in the door? This feels short sighted.

You may not be at a company where people are dropping AI slop into the codebase faster than it can be cleaned up. I've thought about my options for going "single operator", so I don't end up with a mismatch between people who use AI thoughtfully and those those who don't.

I have a hard time believing that won't be overcome or surpassed when you compare it to the cost of producing an adult human capable of skilled trades.

> One wonders whether a generation that demands instant satisfaction of all its needs and instant solution of the world's problems will produce anything of lasting value. Such a generation, even when equipped with the most modern technology, will be essentially primitive — it will stand in awe of nature, and submit to the tutelage of medicine men.

- Eric Hoffer

I still use plan mode, but during plan mode I will spin out sub agents to implement the current draft in a tmp dir and bring back lessons. I feel this keeps me grounded in my original starting point, rather than ending up with a implemention that meanders through the agents own discovery process. It also lets me ask concrete questions about (possible) implementation

Trades could dry up quickly when as we develop effective robots, and there seems to be a lot of energy dedicated in that direction.

"Outside the box" is out.

> It likely is, but at some point it very well might stop being true.

Thats kinda the point right? It depends on how we train them. It is self fulfilling.

Are there ideas coming from Burning Man?

Did you get anything out of the attempt?

> truth be told the more time that passes the smaller the gap will get as the agents get more capable

Yup. A lot of work is going in to reducing the skill required to operate AI agents.

AI is built to adapt to you. ICs aren't.

With the speed AI moves, a lot of technical decisions become reversible. And while engineering often makes decisions that could lie elsewhere in the business, outside of engineering, I could imagine those decisions moving elsewhere in a fully AI world.

Do you have examples of things that would be hard to train for? One that could be compensated for with changes elsewhere in the business process?

> one where software engineers can focus on data structures, software architecture and algorithms.

I see this a lot and I'm not sure why people don't think AI will be able to do this too. The self-play training that got them writing code can be used for this too.