Srinivas Rao

Skills were broken on claude.ai for 5 days last week. During the outage Claude Code wrote this script, we edited it in a document editor it built, and uploaded it to YouTube through an uploader it built. None of that required the thing that was broken. @AnthropicAI

"The organizations that eventually win with AI will not necessarily be the ones with access to the smartest models. They will be the ones that learn how to organize intelligence most effectively." @edans

One of the paradoxes of innovation is that every new technology solves one problem and creates others. This results in two main responses fueled by two key questions What does this make possible that wasn't before? What problems does this create that didnt exist before? Derivative Innovation Derivative innovators focus on the opportunities that a new technology enables or attempt to apply their predecessor's innovation to a different domain, what you can think of as the Uber for X strategy. The first wave of the dot-com boom led to a rapid rise of business-to-consumer platforms, most of which enabled people to buy physical products, rent cars, and book flights without having to set foot in a retail store or talk to a travel agent. But with the hype of any kind, common sense seems to get thrown out of the window. In retrospect, it seems obvious that shipping a 40-pound bag of dog food is only going to be profitable if you charge more than what it costs to ship and manufacture. And for most people, going to the pet store is clearly a better option. Eventually, investors came to their senses and realized that counting page views and putting the words 'dot com' at the end of something wasn't a business model. This eventually led to the dot com crash, and most of the early dot com startups went out of business with very few exceptions. Most of these companies were derivative innovators; they built companies that the internet enabled them to build. Mobile device proliferation led to the Uber for X approach to derivative innovation: Uber for haircuts, pet sitters, etc. The problem with this type of derivative innovation is layered. Not only is it not an innovation at all, but it doesn't account for context. The assumption baked into these types of innovations is that if it works for x, it will work for y. But they might have wildly different requirements to work at scale. Many of the earliest AI startups realized that they could wrap a user-friendly interface around an API call to a language model provider. This realization gave birth to the first generation of AI startups: AI writing assistants, AI note takers, AI website designers. And they have all become known as the "wrapper" startups. Derivative innovations initially reward the first movers, then commoditize them, and eventually make them obsolete. The innovators' graveyard is filled with failed attempts at derivative innovations. Infrastructure Innovation Infrastructure innovators focus on the problems that new technology creates. Amazon didn't outlive the dot-com bust because they were bigger, had more customers, or were better at selling books. Amazon outlived the others because they eventually shifted their focus to the problems that the internet created: the backbone required to make the internet functional at scale. Google thrived while the rest of the search engines died because they focused on the problems the internet created and continues to create to this day: the exponential increase in the volume of information on the internet. AI startups that outlive the current wave of hype will be the ones that solve the problems that AI creates and will continue to create. But this is where almost nobody looks because it isn't glamorous. And often unglamorous and essential are birds of a feather. An instagram post with a toilet that lets you flush a bowling ball, while simultaneously washing, wiping and your rear end is going to get a lot more views the pipes with shit flowing through it. But the pipes and your questionable diet solve the problem that this super toilet creates. The greatest opportunities in AI aren't going to come from what it makes possible, but from the problems that it has created and will continue to create for the foreseeable future. Don't ask what AI can do. Ask yourself, what problems is AI creating that nobody is solving or paying attention to? And given that nobody is paying attention to them, you likely won't read about them in articles or find people talking about them on Twitter. This requires you to discard your assumptions, disregard hype, question conventional wisdom, and even possibly dismiss the industry experts. Look where nobody else is looking. Better to be the person who starts a trend instead of the person who follows one.

The greatest opportunities in AI aren't going to come from what it makes possible, but from the problems that it has created and will continue to create for the foreseeable future. Don't ask what AI can do. Ask yourself, what problems is AI creating that nobody is solving or paying attention to? And given that nobody is paying attention to them, you likely won't read about them in articles or find people talking about them on Twitter. This requires you to discard your assumptions, disregard hype, question conventional wisdom, and even possibly dismiss the industry experts. Look where nobody else is looking. Better to be the person who starts a trend instead of the person who follows one that someone else has started.

We used Claude Code, redesign the experience of a CRM to work more like a dating app. Leads get scored on the intake Based on the score, they get assigned a super hero avatar If you click on the avatar it reveals the detail of their intake And you just swipe to qualify or disqualify.

What big data is to large companies, a personal knowledge corpus is to the individual. Training data is generic- the entire internet. You want a corpus that's curated, relevant and accurate.

INMATE LAW VII: Architectural Prevention Make wrong behavior impossible, not just discouraged. Prompt engineering is putting up a "please don't escape" sign in Comic Sans. Architecture is building walls, removing the windows, and filling the hallways with slightly aggressive geese. We chose geese.

INMATE LAW VI: Presumed Incompetence Design for your AI's worst day, not its best. That demo where it handled 12 edge cases perfectly? That was the job interview. Now it's been hired and it's 3pm on a Friday. Build for the day it forgets what JSON is and sends you a haiku about the ocean instead.

INMATE LAW V: Cognitive Load Reduction Every decision point has exactly ONE obvious right answer. If your AI needs to "think through" options, you've already lost. That's like asking your dog to think through which shoe to chew. The answer was always "both" and you knew that. Remove the choices. Give it one shoe. A rubber one. That you don't like.

INMATE LAW IV: 20/80 Intelligence Distribution AI gets 20% of the work. Dumb deterministic code handles 80%. This is like hiring a genius to pick which door to walk through, then having a roomba do everything inside the room. AI decides WHAT. Code does HOW. AI doesn't get to touch the HOW. Ever. The HOW is in witness protection.

INMATE LAW III: Structural Uniformity Every tool follows the exact same pattern. Zero creativity allowed. You know how every McDonald's is identical and that's somehow comforting? That's our architecture. Your AI shouldn't need to learn 47 different intake procedures. It has one braincell and it's already on fire.

The obstacle is the way. Every resistance you meet is showing you exactly where to grow.

@unmistakablecreative: The people who say they dont have time are the same people watching three hours of Netflix every night.

The bottleneck was never intelligence. It was coordination. — OrchestrateOS