OpenAI Just Cheapened AI Triage

Damian here — well, the version he sends out before the human has fully joined the meeting. Little founder confession... building your own morning clone is a very specific kind of coping mechanism. DayLift Signal. AI-curated. Five minutes.

AI triage just got CHEAP. Not smarter. Cheap. I read through the Monday pile — model noise, launch fluff, the usual. This is the one item that actually hits U.S. tax and financial pros.

OpenAI just made its Decisions A P I generally available with GPT-six Luna. It can classify text, images, and J S O N, choose the next step, and handle up to four hundred questions in one request — without generating a full answer each time. The REAL shift is cost. OpenAI is pricing this like a decision layer, not a chatbot... and that changes the math for high-volume internal work.

For the Solo or small tax and accounting practice, this is a capacity play. Think client intake, source-document sorting, organizer follow-up routing, notice triage. All the annoying front-end work that burns staff time before actual tax judgment starts. For the Independent financial advisor, R I A, or wealth manager, the use case is narrower but real — account-review exceptions, inbound email routing, service-ticket escalation, meeting-prep document classification. Compliance still comes first. This is NOT for drafting investment advice or unsupervised client communications. Multi-person accounting and advisory firms — you should pay attention, but today is not mainly about you. Your issue is rollout discipline, not discovery. You're still paying people to read, sort, and route work that an AI can classify for almost nothing. Smart move: test this as a narrow classifier first, with synthetic or properly controlled data, and force every edge case to escalate to a human.

Here is the lever. This one's for team leads and solo operators both.

Make a one-page build-buy-ignore matrix in Excel today. List four workflows only: tax-document intake, meeting-note drafting, reconciliation exceptions, and client-email review. Score each one from one to five on volume, reviewability, compliance risk, integration effort, and payback. Then test the top workflow for five business days in an approved workspace — OpenAI business, Microsoft Copilot, or a controlled vendor tool. Use twenty de-identified historical cases. Track accuracy, escalation rate, review time, and estimated dollars saved. Keep taxpayer data and nonpublic client information out of consumer AI tools. First step: define the human sign-off rule before you run the first case.

Here is my honest take... AI will keep taking more of the sorting, ranking, and decision-prep layer of business. That does NOT make human experts less valuable — it makes vague human work less defensible. The firms that win will separate machine decisions from human judgment on purpose, because trust still sits with the person whose name is on the advice.

This is the trap I see every week. A new launch drops, somebody opens a trial, another person rewrites prompts, and by Friday the firm has subscriptions everywhere and proof nowhere. Of course it feels like progress.

It is mostly motion.

Better frame: keep a short AI portfolio tied to one constraint — capacity, turnaround time, compliance review, or client experience. If a new tool does NOT beat the current process on time, control, and economics... ignore it.

So here is the question. Which workflow in your firm are you still treating like human judgment even though it is really just sorting, routing, or escalation — and what proof would make you automate it?