🤖💡 Useful way to think about AI tools:
👉 Initially tune your AI tools to solve specific tasks, with an escalation path to a human
📈 Over time, as these tasks accumulate, they can represent the workload of a full person.
Voice AI is getting so good over the phone that it’s basically becoming a permissionless API to the enterprise.
Enterprises can restrict APIs, but they can’t restrict incoming calls.
OpenTable won’t give you API access to make/modify/cancel a reservation? A voice agent can just
Waymo just dropped the beginning of the end.
This data, and the way this data will keep improving, is the reason why future generations won't be allowed to manually drive cars on public roads.
The future of personal transportation is 100% autonomous
This is an important observation. My preference is to use 3rd party solutions where possible (go faster and have a bigger team build a dedicated solution). The incentive to "cheap out" on the underlying LLM needs to be watched / mitigated.
The problem with using SaaS vendors with their own AI solutions is that their incentives are to use cheap models, as little reasoning as possible & to stick with outdated prompting & RAG strategies than updating them as AI improves
Not all vendors succumb to temptation, many do.