Apple’s M5 Pro 64GB suits local agents

The real question is not whether to buy a new MacBook Pro, but whether local LLMs are mature enough to justify a premium workstation purchase. If local agents are part of the daily workflow, 64GB is the practical floor because 24GB gets tight once the model, IDE, browser, and context all pile up. Qwen 32B can work locally, but it is still a compromise versus top API models: slower, more finicky, and more dependent on prompt and tooling discipline. For coding workloads, memory headroom matters more than the chip bump because the bottleneck is RAM pressure and swapping, not raw Python speed. If local models are only occasional, paying MacBook Pro premium pricing is harder to justify, and API tokens are often the cheaper way to keep moving. Supply and price anxiety are understandable, but they should not drive the decision unless local agents are already central to the next few years of work.