Models barely change. Did I over-engineer my AI stack?
4 months ago I tried to future-proof my AI setup.
Instead of picking models directly, I built everything around 3 classes:
- flagship
- standard
- fast
The idea: when new models drop, the system remaps automatically.
It worked.
In 4 months, it updated itself once. No intervention needed.
Which is also the problem.
Now I’m wondering if I built abstraction for something that barely happens.
Are people actually seeing meaningful model churn?
Or is everyone just hardcoding and moving on?
Trending on Indie Hackers
Hey — this is a thoughtful setup. The abstraction makes sense in theory, but yeah… model churn hasn’t been fast enough to fully justify it for most people (yet).
From what I’ve seen, most builders still hardcode for now and only switch when there’s a clear jump (cost/performance), not constantly.
Curious — are you seeing any benefit on the cost/latency side from this setup, or has it mostly just stayed idle so far?
Also, I’m running a small experiment with builders working on infra/AI workflows like this.
$19 entry, winner gets a Tokyo trip (flights + hotel). Round 01 is live (100 cap).