The Messy Middle
Most of what we’re building in AI right now is throw away in the long term.
I want to start off with a bold statement. (Brace yourself!)
Most of what we’re building in AI right now is throw away within the next 5-7 years.
It’s all scaffolding and throwaway frameworks.
Everyone is chasing speed by creating faster wrappers, quicker prompts and “agents” that look autonomous but collapse hard under the weight of complexity.
None of this is the end goal → It’s the messy middle.
Think back to the teacher we all had at some point in our lives who said:
“I am not teaching you how to memorize. I’m teaching you how to teach yourself how to learn.”
In research labs, medical diagnostics and robotics systems already tie into high-stakes execution loops where precision matters.
Everywhere else?
We’re still teaching AI how to learn.
Right now we’re modeling our own behaviors by breaking down tasks, orchestrating steps and showing feedback loops to LLMs and early agents. These brittle systems aren’t intended to last. They’re training wheels to teach us how to augment with AI by teaching it how to learn how to learn.
Here’s the truth → most of the tools we’re building today still come from yesterday’s playbook!
Static sites, 3-tier apps, dashboards, landing pages, logins → these are artifacts of a UI-first world that assume humans must navigate fixed structures.
Productivity and PM tools → most bolt AI on top of old workflows, forcing models into rigid templates instead of rethinking the workflow itself.
Dev frameworks → still anchored in building persistent layers,when the future will be ephemeral with UI generated on-demand and destroyed when the task is done.
In other words, we’re training AI using today’s POV and modus operandi while trying to make AI fit old patterns instead of reinventing new ones.
The goal isn’t faster apps or shinier dashboards.