⋅ ⋅
Right now, with current AI models and the form factor they’re provided in, we essentially have human intelligence and judgement on tap.
Using the big labs’ APIs, you can turn the faucet and have freely flowing intelligence, opinions, judgements, images, music, software, and more.
We're still figuring out how best to harness this newfound resource, but it’s already clear this isn’t AI’s final form. Right now, we’re in what I would consider the technological equivalent of petroleum’s kerosene era: when oil was first discovered bubbling from the ground, the most obvious use was simply replacing whale oil in lamps. At that time, it wasn’t immediately obvious that this crude, sticky substance could power far more than lanterns - it had within it the potential to completely reshape society.
Petroleum’s influence ultimately extended far beyond lighting, and it became the fuel powering internal combustion engines, enabling automobiles, airplanes, and industries that were previously unimaginable. Similarly, I think our current use of AI is still narrowly focused, mostly enhancing existing workflows, and we haven’t yet grasped its full transformative potential. The real "combustion engine" moment, the Killer App capable of unlocking entirely new ways of living and working, has not yet arrived imo, and products like ChatGPT only offer a small glimpse of what’s possible.
This also echoes the early days of electricity. Initially, electrical power was adopted only superficially, layered onto factories and workshops built for older sources of power. It took decades before architects and engineers redesigned factories and entire cities around electricity, fundamentally reshaping industry and daily life. Only then did electricity drive a Cambrian explosion of new appliances, products, and societal change.
Today, I believe we find ourselves at exactly that point with AI. We're currently at the stage of longer-lasting lamp lights and early electrical fixtures, applying AI incrementally to familiar tasks. Companies are refactoring traditional products like IDEs, automating customer support, or streamlining routine research tasks - pouring raw intelligence into yesterday’s structures and workflows. These improvements, though useful, are modest compared to the transformative potential awaiting discovery. The real power of AI lies beyond merely boosting efficiency; it will manifest in applications we cannot yet even imagine.
History shows that the most powerful applications and businesses are often those uniquely unlocked by new technologies. During the internet revolution, entirely new business models emerged that simply weren’t feasible before. An older article from @packyM titled Crypto Bezos highlighted Amazon as a prime (heh) example: it leveraged the internet’s instant global connectivity to aggregate demand, offer an unprecedented selection of goods, and deliver them directly to customers’ doorsteps all at a scale, efficiency, and speed previously unimaginable. This logistical and technological paradigm enabled enormous value creation by fundamentally reshaping how commerce worked and, prior to the internet, couldn’t have existed.
Truly AI-native workflows will similarly explore what’s possible when the engine is built or the factory floor is designed from the ground up to be electrified, so to speak. We haven’t reached this point yet, but some early experiments hint at the possibilities ahead.
For example, at @AlmanaxAI we’re attempting to build an early-stage implementation of this engine idea. Rather than simply applying intelligence to existing workflows, we’ve structured our product from the ground up around AI-first concepts, allowing specialized security agents to interact, consult, and collaboratively uncover vulnerabilities in a codebase. We’re still early, and our understanding of the best possible architecture continues to evolve, but we’ve seen firsthand how transformative even initial experiments can be. Tasks that once required individual security engineers hours, days, or weeks are now increasingly feasible within minutes.
Now, going back to the Amazon example, it didn’t reinvent commerce entirely; it simply leveraged new technology to achieve huge efficiency gains and vastly improve customer experiences. Similarly, the concept of security scanning isn’t novel - but rethinking it from the ground up with flowing intelligence reveals entirely new approaches. Almanax isn’t the combustion engine itself, but it’s an early attempt at steering this raw new force in a useful direction, and each new frontier model release is an even more highly-distilled fuel. To borrow @DarioAmodei ’s framing from Machines of Loving Grace, if a data center could soon house the equivalent of a country full of geniuses, what kind of security infrastructure becomes possible when that collective intelligence is focused on understanding as well as fortifying your codebase and critical infrastructure? That’s the question we’re exploring and we’re just getting started.
However, more generally beyond security, the really exciting opportunities lie ahead in creating novel applications and experiences that simply couldn’t exist before we had freely flowing intelligence and judgement on tap. Rather than refining longer-lasting lamp lights, I can’t wait until we get our first glimpses of the engine itself.