> 90% of the code is 90% of the work. The last 10% of the code is the other 90% of the work.

Don't think I've heard that one but certainly rings true to my experience.

Reminds me of "ninety percent of the game is half mental"

Fully agree. Shipping a complete product with a functioning user acquisition funnel is much harder. It's like; you have to build the whole product first with lots of features and then you have to try to create a highly condensed overview of all those features to expose them all on the landing page.

If you can't make the visitor understand your entire complex product in 10 seconds, then you've lost them.

Your product has to be complex because that's where the software market is at. All of the low-hanging fruits have been taken by the time you identify them. Sure, someone will find a way to make money using new low-hanging fruits that arise due to technological changes but it's not going to be you. You probably don't have the business connections to make that work.

> There's so much work in delivering products that will carry your brand, and then must be supported.

People think otherwise with AI partly because Anthropic kept telling us that they didn't have to write code or review code any more for most of their work. Their agent swarms just comb through their github, slack and wikis to figure out what to do next, and another swarm of agents just review, test, merge, deploy, A/B test, and revert the code. Boris alone merged nearly 300 PRs in the past week (or two?). So the top research labs seem have broken the productivity seal.

And then they talk about this recursively self-improving AI that is so powerful, so autonomous that they advocate that every company should be prepared to "pause" the effort. And their Fable/Mythos has this specific restriction as mentioned in their model card[1] that they are going to reject requests to tune and train models because, well you guess it, the models are too powerful to be used by mere mortals.

[1] We’ve implemented new interventions that limit Claude’s effectiveness for requests targeting frontier LLM development (for example, on building pretraining pipelines, distributed training infrastructure, or ML accelerator design). Using Claude to develop competing models already violates our Terms of Service, but enforcing this restriction through our safeguards avoids accelerating the actors most willing to violate these terms. Unlike our interventions for cybersecurity, biology and chemistry, and distillation attempts, these safeguards will not be visible to the user. Fable 5 will not fall back to a different model. Instead, the safeguards will limit effectiveness through methods such as prompt modification, steering vectors, or parameter-efficient fine-tuning (PEFT).

I like this analogy; raising children well like delivering products well pays dividends. They’re less likely to cause problems and if they do, they tend to be smaller in scope.

I skimmed the article, guilty, but I think what I got from it is that CEOs will CEO? No disrespect meant, I’ve seen your name here often and thoroughly enjoy the folklore that you share, but I don’t understand what context you reacted to. Cheers.

> Conceiving them is fun. Delivering them is painful. Raising them, is a lifetime of work.

Then there's the technical debt!

Shipping is frankly the easy part. It's the operating overhead that often breaks you.

I liken it to free puppies.

now it's closer to 95% of work can be done by AI and requires 5% mental effort, but 5% of the work requires 95% of the mental effort to finish because of all the unoptimial decisions AI has taken. I find that AI works best in small micro-service type architecture where each component has a clear goal and doesn't have interconnected parts within the same application that can break. But you do run into an issue where changes in microservice a need changes in microservice b and updating it is not ideal since it usually cascades thru the entire system or requires stacks of legacy support.

[flagged]