Yes, I agree. You should use the best tools out there. And I'm going to add that if the tool is doing the thing that needs to be done, jumping from tool to tool just because something might be better at the same workflow is going to waste your time. So somewhere between staying on the trend and being discerning about what's in your stack is the optimal price risk-reward ratio.

https://www.linkedin.com/embed/feed/update/urn:li:share:7455264284818882560?collapsed=1

Is what I'm about to spend the next 4 hours on buying me time, money, or leverage — and if it's not leverage, who or what should be doing it instead?

The local model landscape has matured to the point where a school district with even modest hardware can run capable LLMs (Llama 3, Phi-3, Mistral, Qwen) on machines they already own or can acquire cheaply. No API costs, no student data leaving the building, no per-seat licensing. A single machine with a decent GPU or even an M-series Mac mini can serve a classroom.

https://www.loom.com/share/c743c41c43e849ddb626c0e528d6b835