You May Already Own the Thing You’re About to Lose
Microsoft pulls its best free research tool on August 18 — and the replacement may already be sitting in a Microsoft account you pay for and forgot about. This week’s real divide isn’t who has AI. It’s who knows what they already own.
HEADLINE STORY
On Tuesday, August 18, Microsoft shuts off Deep Research in the consumer Copilot app. It was the best thing on that free tier. You gave it a complicated question, waited a few minutes, and got back a real report with the sources listed.
Microsoft points people to a replacement called Researcher. There are two ways to get it, and for a business they are not the same thing.
One is Microsoft 365 Premium at $19.99 a month. Read the fine print on that one. It’s a household plan. The AI belongs to whoever owns the subscription, it doesn’t reach your company’s files or email, and your IT people have no control over it. Fine for personal use. It is not a business tool, whatever the price makes you think.
The other is a Microsoft 365 Copilot license, which runs roughly $18 to $30 per user per month on top of what you already pay for Microsoft 365. That’s the one that connects to your company’s actual work.
Now here’s the part that matters for a dealer.
If you run Microsoft 365, there’s a fair chance somebody already bought a few of those licenses. During a renewal, or bundled into a package, or because a vendor made the case a year ago and you signed off and haven’t thought about it since. If that’s your situation, Researcher is in your account Tuesday morning and you don’t have to buy a thing.
And Researcher isn’t the only thing sitting in there. Since this summer, a Copilot license also lets you pick which company’s AI does the work. Microsoft’s own default is OpenAI, the ChatGPT people. Anthropic’s Claude is on the menu too. Same license, no extra cost, and you choose it from a dropdown inside Word or Excel the way you’d pick a font.
There’s a fair chance nobody at your company knows any of that.
This isn’t about deciding whether to buy something. It’s about finding out what you already bought.
So I’d spend twenty minutes on this. Pull up what your Microsoft agreement covers, have somebody confirm the model options are switched on, then ask the two or three people who’d actually use it whether they’ve ever opened it.
Most companies I talk to are paying for more AI than they use. Not because anybody made a bad call. Because the seats got bought during a renewal, the announcement went out once, and nobody followed up.
WHAT ELSE MATTERED
The output gap between firms nearly tripled
OpenAI published research on its own enterprise customers this week, and one number stuck with me. The company ranks those customers by how much finished work each active user produces. Back in January, the top tier was running about 2.6 times the typical firm. By June it was 8.3 times. The spread nearly tripled in seven months, and the top firms aren’t logging in more often. They’re finishing more per person who shows up.
Two caveats. These are OpenAI’s own customers, and they skew large and tech-heavy, so don’t take the exact multiples to the bank. But the same research found something I’d bet holds at any size: once people have had the tools six months, employees early in their careers are using them noticeably more than executives are. Most owners assume the opposite.
AI labeling became law — for images and audio, not text
California’s AI Transparency Act took effect August 2. It’s the first live state requirement that big AI providers mark what their tools produce, and it covers images, video, and audio, not text. The obligation sits with the AI companies, not with you. Anthropic separately said it will watermark AI-written text to satisfy the EU’s version. Nothing here means a dealer has to label anything today. It’s just the first sign that this is settled by law now instead of by preference.
THE PATTERN BEHIND THE WEEK
The gap this week isn’t between companies that have AI and companies that don’t. Almost everybody has it. The gap is between the ones who know what’s in the building and the ones who don’t. A capable feature gets pulled and most of the people who had it never found it. Firms pull eight times more work out of the same tools. Somebody’s already paying for Researcher and won’t use it.
None of that is a technology problem. It’s an inventory problem — and inventory is something you already know how to run.
WHAT TO DO NOW
- Find out what you’re paying for. Microsoft first, since Tuesday’s change makes it timely, but the same question applies to every subscription with AI attached. Somebody should be able to answer it by Friday.
- Change the question at your next check-in. Not who has access. Who finished something real with it this month — a quote package, a drafted policy, a vendor comparison you’d act on.
- Give your two or three best users twenty minutes in front of everyone else. People pick this up by watching somebody good do it, not from an email.
WHAT I’M WATCHING
- Whether OpenAI’s gap number holds once they break it out by industry. Right now it’s one blended average across a customer base that leans big.
- xAI’s Grok 4.6. Released August 7, and independent testing this week landed it near the top of the field at roughly a third of what the leading models charge. Worth watching whether that drags prices down over the next few months.
- Whether AI keeps moving into sales work as fast as this week’s data suggests. Legal and recruiting are growing fastest right now. Sales is the one you’d feel.
ONE QUESTION WORTH ASKING
If somebody handed you a list tomorrow of every AI tool your company pays for and who actually uses it, would anything on that list surprise you?
