Local Models, Friction and Struggle (garden.azl.au)

In a recent exploration of the evolving landscape of AI tools, the author shares insights from their experience transitioning to local models as a complement to larger language models (LLMs) like Claude. While Claude can generate extensive drafts quickly, its eagerness often results in assumptions about the user's intent and experience that don't align with their actual journey. The author’s shift to local models has underscored the need for more deliberate engineering thinking, as these tools require greater manual input and oversight. This hands-on engagement fosters a deeper understanding of the engineering process, where the user must grapple with key decisions rather than relying on an AI's output. This discussion highlights a vital divergence in AI tool utility: while advanced models can reduce external friction, they can also obscure the internal struggle that leads to thoughtful engineering solutions. Local models, in contrast, encourage introspection and critical questioning about the development process, asserting that the struggle itself is essential for achieving engineering excellence. As AI becomes increasingly capable, the author argues that the most successful engineers will be those who embrace the complexities of their work, continually wrestling with the challenges of innovation and decision-making rather than surrendering these responsibilities to AI.