I think these two videos describe what the ground floor looks like going forward.
Back in 2024, predictions were that AI would have a 10% chance of being better at all jobs by 2027. I'd argue that for the majority of technical/computer jobs, we are already there.
AI *can* do your job. And likely can do it better. It just hasn't replaced you, *yet*. We are running off a cliff, and haven't realized there is only air beneath our feet.
The adoption hasnt kicked in yet. And (besides just infrastructure and memory systems) the reason why is simple:
Humans suck at explaining how to do their own job.
Humans have too many things we take for granted, too many things that are just instinct, that we expect other humans to know. But AI isn't human, and despite being trained on human data, acting like a human, and having similar behavioral pitfalls.. AI does not *think* like a human.
If I tell an human to make me a sandwhich, they can make reasonable decisions about what type of bread I want, how I want it put together--but the AI does not know those unspoken/obvious things we take for granted. The AI thinks in continuous concept space, but speaks and acts in discrete, definable, measureable, token space.
So you have to tell it exactly how you want the sandwhich. You need to think differently, and split apart what you actually want into achievable metrics.
These two videos do a really good job of explaining this concept, and putting into actionable directions.
Prompt Engineering Is Dead. Context Engineering Is Dying. What Comes Next Changes Everything.
https://www.youtube.com/watch?v=QWzLPn164w0
The Trillion Dollar Agentic Workflow Opportunity Is Here
https://www.youtube.com/watch?v=jwtpMSRAPAQ
Here is a possible solution you could apply to your personal usage:
https://natesnewsletter.substack.com/p/your-agent-needs-a-soulmd-you-cant
Cited source for predictions:
https://arxiv.org/abs/2401.02843
