The "plumbers trying to fix toilets" analogy is spot on. There is often a lot of gatekeeping in tech around the "sanctity of the struggle" when learning, but if the goal is just to ship a fix and move on, much of that manual research is just yak-shaving.
I also really resonate with the idea of writing as a way to "actually think." Using GenAI for design docs feels like offloading the most important part of the engineering process, the mental model construction. It’s the difference between using AI to handle the boilerplate and using it to handle the logic.
That 80/20 local minimum where prompting fails is still the biggest hurdle in 2026. At what point do you find it's faster to just drop the agent and finish the last 20% manually?
The "plumbers trying to fix toilets" analogy is spot on. There is often a lot of gatekeeping in tech around the "sanctity of the struggle" when learning, but if the goal is just to ship a fix and move on, much of that manual research is just yak-shaving.
I also really resonate with the idea of writing as a way to "actually think." Using GenAI for design docs feels like offloading the most important part of the engineering process, the mental model construction. It’s the difference between using AI to handle the boilerplate and using it to handle the logic.
That 80/20 local minimum where prompting fails is still the biggest hurdle in 2026. At what point do you find it's faster to just drop the agent and finish the last 20% manually?
These thoughts? https://lobste.rs/s/ypes5v/thoughts_on_generative_i