So basically, it's like saying "you can't sue me if my software has bugs"

> "The only layer immune to "efficiency" is the one that approves it."

No responsible parent will allow a child on a balance bike (or one with training wheels) in any scenario where they'll 'endanger' anything.

Balance bikes teach you to ride a bike (maintain balance while steering). Training wheels teach you to pedal a bike.

Learning balancing and steering first, and then learning to actually cycle the pedals is the right order of events I think.

You solution is to hope that a better technology will be invented that won't be subject to what every technology before it has been subject to. You rally against "the algorithm" and beg for another algorithm to save you. You want to live in a filter bubble and not be challenged on you believes. I can't think of anything worse, but I wish you the best.

Also I don't know what comment you read but it's clearly not mine...

This is a key point, and one that is rarely articulated.

The problem -- for me, anyway -- with the Current Era is specifically the workload, but the relentlessness.

I'm old enough to remember the Old Days of big release dates. You'd grind for a while, sometimes in a deathmarch, but after the release the organization would relax and there would be _some_ slack.

Now it's an endless river of prompts, push to prod, fix, repeat ad nauseam. Because implementation is now much cheaper but genuinely good ideas are as scarce as ever, companies are turning into a random walk optimizations weirdly mirroring training regimes. The idea apparently being that you make enough features/changes you'll eventually find the key to unlocking hyper-growth.

But unlike a training system, all these intermediate steps compound. Customers find their software mutating under them at an alarming rate. Maintainability becomes barely possible, even with Claude doing a ton of work. We're just creating technical debt we expect the future AI to fix, and getting no closer to the "revenue unlock."

Many are chained with fear to this dehumanizing and demoralizing system, and it sucks.

OpenAI and Anthropic have completely different motives, they're in a zero-sum game with each other and with cheap Chinese models. Regulatory capture that results in artificial barriers of entry is their best bet at healthy profit margins even if it comes at the cost of less overall AI buildout.

It's not like "the AI industry" is a single entity with a single purpose, these are different companies with different objectives.

there is alot of beta involved, i.e. put 10M into payroll startup xyz, expect some rate of return.

They might not always live up to that, but at least the aspiration is there.

No such pretense from American VCs.

[0] https://en.wikipedia.org/wiki/Serve_the_People

When starting a brand new project it’s better to have a lot of human involvement to make sure the AI is creating a foundation that actually matches requirements, both spoken and unspoken. But once the project reaches maturity it’s more realistic to imagine an AI working on top of that concrete layer.

So goes management. Bootstrapping a new company? Needs human involvement. A company just ticking along? I’m less convinced.