The AI hype-train is about to rediscover the Agile Manifesto

It's interesting to watch the AI coding trends on LinkedIn right now. We're collectively burning through massive amounts of GPU compute just to slowly rediscover every core software principle we abandoned the second LLMs showed up.

We spent two years writing elaborate markdown files to apologize to AI for our messy codebases. We meticulously crafted context files to explain our architecture, completely ignoring the fact that if a frontier model needs a map to navigate your repository, your next human hire is already doomed.

We did the exact same thing with our prompts. We immediately regressed to 1990s waterfall specs, aggressively micromanaging the machine with rigid pseudocode. Then a minor miracle occurred: people started noticing that if you just explain the actual user problem instead of dictating the mechanical solution, the AI does a better job. Give intent instead of requirements. Rely on self-explaining code over scattered documentation. Give the intelligence room to actually think.

We're probably five minutes away from someone publishing the "AI Coding Manifesto." It'll just be the Agile Manifesto repackaged for the prompt engineering era. When it drops, it'll inevitably consist of these seven principles:

1. Self-explaining code over massive system prompts.
2. Core user needs over rigid pseudocode requirements.
3. Clear architecture over infinitely expanding context windows.
4. Human intent over mechanical step-by-step processes.
5. Discoverable modules over companion .md cheat sheets.
6. Tests that read like documentation over hoping the agent guesses correctly.
7. Naming things well over writing a 200-line file to explain what data_final_v2 means.

Good engineering practices were never just for human constraints. They're just how you manage complexity. We didn't discover a new paradigm. We just paid a premium to relearn what we already knew in 2001.