The week in AI for CEOs: three things that matter, and one workflow to run.

Your first live office hours session is Friday, August 14. Bring a real problem and we'll get it working together, or reply to this email with what you want covered and I'll prep it.

OpenAI just handed most of its customer support to an AI agent

On Tuesday OpenAI introduced Presence, a platform large companies use to build AI agents for jobs like customer support, sales follow-up, and supplier onboarding, over chat and voice. The proof is OpenAI's own business: the agent handling its inbound support matched the response quality of its human help desk within a few weeks of going live, and now resolves about three quarters of requests on its own. Presence is invite-only for large enterprises right now. (SiliconANGLE)

What it means for you: Ignore the product; you can't buy it yet anyway. Look at what OpenAI chose to automate first in its own company: the front line of customer support. The old assumption that an agent talking to your customers is a next-year experiment just expired. This capability reaches the tools you already pay for within months, and there's prep that pays off no matter which vendor wins: have your support lead write down your twenty most common questions and the ideal answer to each. That document is the fuel for any support agent you ever turn on, and writing it will improve your human support next week.

Anthropic opened its data on what people actually use AI for

Yesterday Anthropic released a free connector that lets you ask Claude about its Economic Index, the public dataset measuring how AI is used across the economy. In claude.ai, open the connectors menu, find the Anthropic Economic Index, and turn it on; nothing to install. Then ask in plain English: which occupations use AI the most, what tasks are people automating, how is AI being used in my industry. The answers come grounded in the Index data, and Claude points you back to the source as you go. One caveat Anthropic states itself: the Index reflects how people use Claude, not the labor market as a whole. (Anthropic)

What it means for you: This is worth ten minutes this week. Ask what tasks are being automated in your industry, then in your customers' industries. You're not looking for trivia; you're looking for the two or three tasks other companies already hand to AI that your team still does by hand. Write those down. That short list is your next month of AI projects, picked from evidence instead of vibes.

New study: the companies making money on AI do two boring things

A study released Monday by HCLTech, one of the big global IT firms, surveyed 500 enterprise decision-makers and found a brutal gap: nine in ten say generative AI and agents are transforming how their company works, but fewer than one in five say AI is delivering significant revenue impact. The companies that do make money from it differ in specific, boring ways: they define measurable use cases (about three quarters of them, versus a fifth of everyone else), and they run structured AI training for their people (more than nine in ten, versus one in five). (HCLTech)

What it means for you: A productivity gain you never measure and never reinvest is indistinguishable from zero. If your company is in the nine-in-ten (workflows changed, P&L unchanged), the fix is not a better model. It's the two boring habits above: pick workflows you can measure, and train your people on purpose instead of hoping they figure it out. The first one is exactly what this week's workflow does.