AI Can Now Hold the Expert Conversation, Live

Google Research reported this week that AMIE, its medical conversational AI, matched board-certified physicians in live video consultations. Read that carefully, because the important word is live. We have had AI that reasons well about a problem once somebody writes it down. This is AI handling the interaction itself: hearing the concern, asking the follow-up, adjusting as it goes. For anyone whose business is essentially expert conversation, and that is most professional services, this is a more consequential threshold than any benchmark score.

The moat was never the analysis

Professionals watching AI improve have generally consoled themselves with a reasonable observation: clients do not hand you a tidy written brief. They arrive with something half-formed and slightly wrong, and your value is in asking the right questions, noticing what they did not say, and steering toward the real issue while the conversation is happening. That interactive skill, not the analysis, has been the durable moat. Which is exactly why a system that can hold the conversation competently is a different kind of news from one that writes a good report, a step beyond the expert-level analysis milestones we covered earlier.

Keeping it in proportion

Now the caveats, which are real. Matching experts in a controlled research setting is not the same as replacing a licensed professional who carries accountability, knows local context, and can be held responsible when something goes wrong. Regulated professions have liability and licensing structures that move far slower than capability, deliberately. This is a research result, not something you can buy on Monday. The honest read is not replacement but a shift in where your time is worth spending.

Notice that the left column is where a lot of billable time quietly goes, and the right column is what clients actually say they value you for.

The uncomfortable question worth asking

What share of your billable hours goes to conversations that follow a predictable shape? For most firms the honest answer is larger than they would like: the same intake questions, the same standard follow-up, the same quarterly status call. That portion is becoming automatable, which means it will eventually be priced by whoever automates it first. That is not a reason for gloom. It is a reason to be the firm that automates the routine layer itself and redeploys the time, rather than the one defending a margin on work a machine can now do.

What to do about it

Sort your client interactions into predictable and genuinely-judgment-requiring. For the predictable ones, look at where AI could handle intake, preparation, or first-pass questions, always with a human reviewing before anything reaches the client, and always telling clients when AI is involved. Then reinvest the recovered hours in the complex work only you can do. The firms that move deliberately here will look faster and more available to clients. The ones that assume the talking part of the job is permanently safe are making a bet that just got noticeably weaker.

Frequently Asked Questions