@vgr

Trying to think through the correct mental model update to the “below the API” idea for the AI era. I think there are two possibilities: a) “It’s all API and everyone’s entangled in it, there’s no above/below” … OR b) a more radical counter-intuitive possibility: The human is the API People often mistakenly credit me with the “below the API” idea even though I cited the original 2015 source (Peter Reinhardt) in the 2017 post where I popularized it. But let’s understand the phrase clearly before trying to update it. The phrase was accurate for the late SaaS era where there was a clear pair of architectural boundaries separating architects from orchestration technology and orchestration technology from individual workers. The “API” was merely an overloaded metaphor for all the limited-flexibility orchestration intelligence in the world (eg Uber dispatching) between those 2 boundaries. Above/below made sense since there was a clear power gradient. It was floor-limited by embodiment constraints, and ceiling-limited by architectural creativity/imagination constraints. Both boundaries have been blown away now. AIs can participate in the creative architecture “above the API” (getting AI to build systems from scratch with vibe-architecting) and the embodied execution “below the API” (starting with Claude rewriting your filesystem, and then controlling your physical stuff as much as you let it, to the limit of current real-world sensorium signals). So the interaction of humans and machines is shifting and all three zones are bleeding into each other. We could end up with a roughly uniform slurry of human and machine intelligence across the power gradient (option a), but I actually think an inversion is underway where human intelligence gravitates to the middle, forming the new “API” (option b) and creating new above/below API roles for zones now dominated by AI! This is because doing orchestration in an AI-eaten stack is actually the hardest part since it’s much richer and fluid than in the pre-AI “Uber dispatcher” archetype era. Now not only is it easier to delegate both creative architecture and embodied execution upwards and downwards (modulo rate of robotics progress downwards), but the middle is much harder than “uber dispatcher” in ways that aren’t about intelligence per se. This is why using AI tools feels like endless exhausting middle-management work. This is where atemporal first principles architecture growing downwards and phenomenology and empirical realities growing upwards meet in the middle. Humans bring the narrative middleware that integrates the two, along with what I’d call a sentimental sense of history, that’s not just context data (AIs have more of that at every level and temporality by orders of magnitude). It’s “small data” + patterns of care. This is architecturally valuable because it bottlenecks vast architectural possibility spaces to meet specific realities in an opinionated way. https://www.ribbonfarm.com/2017/08/17/the-premium-mediocre-life-of-maya-millennial/

1 reply

0 recast

21 reactions