There has been a lot of recent data points on how AI agents will automate away most of the knowledge work and cause mass unemployment. <\/p>\n\n\n\n
But a fundamental paradox is that AI, at least today, is still indeterministic. This is why Gen AI is so valuable – exactly because of their ability to handle indeterministic input. But this indeterministic nature is not always desired, not at every level of a serious workflow. <\/p>\n\n\n\n
Just think about humans – you employ a human instead of machine because of their agency and the indeterminism. But to manage a team of humans, you need deterministic workflows, e.g. expense reports, IT security measures. It’s not about how intelligent the humans are, but the deterministic measures are a necessary mechanism to prevent errors, whether intentional or not.<\/p>\n\n\n\n
This will hold true for AI. In no way humans will let agents run without a harness in serious workflows, no matter how intelligent they are.<\/p>\n\n\n\n
That means even when agents are very cheap to run, for serious workflows, the cost of running agents will still be non-zero. Just like TSMC need to plan carefully when building new factories\/plants even though once built, these factories are run automatically.<\/p>\n\n\n\n
The cost of AI is the cost of designing and building the factory.<\/p>\n","protected":false},"excerpt":{"rendered":"
There has been a lot of recent data points on how AI agents will automate away most of the knowledge work and cause mass unemployment. But a fundamental paradox is that AI, at least today, is still indeterministic. This is why Gen AI is so valuable – exactly because of their ability to handle indeterministic … Continue reading The factory analogy<\/span>