For short, high-volume tasks — classifying support tickets, tagging leads, extracting structured data from a form — the per-call cost just dropped by roughly three-quarters. A workflow that cost $40/month to run at scale now costs closer to $10.
Tasks that were previously "too expensive to automate at this volume" start penciling out. Running every single inbound email through a classification step, instead of sampling, becomes affordable. The threshold for "just automate it" moved meaningfully lower.
Cheaper inference doesn't fix a badly scoped automation. The task still has to be genuinely narrow (one decision, not a general assistant) for the economics to work — see our earlier piece on why "AI agent" scares small business owners more than it should.