Google still can't make AI pay like search did
Issue #073 · 4 min read
Google still can't make AI pay like search did
Plus: schools must choose between AI and Search, and GitHub Copilot lands in JetBrains Air
The big story
Google built the most profitable ad machine in history on search.
It has not figured out how to do the same thing with AI. That's the story behind Alphabet's latest earnings, reported by The Verge.
AI Overviews now sit on top of most searches.
They answer the question right there, on the page.
Which is exactly the problem.
The old model worked because you searched, then clicked a link, then an advertiser paid for that click. AI Overviews skip the click. The answer is already in front of you.
Google has not found a version of that transaction that makes as much money as the one it's replacing.
This matters beyond Google. Every company bolting a chatbot onto a product that used to make money on clicks, views, or logins is about to hit the same wall. If your answer box is good enough that nobody needs to click through, you've built a feature that competes with your own revenue line.
Worth sitting with before you ship the AI summary your team has been excited about.
What shipped
GitHub Copilot now runs inside JetBrains Air, using the Agent Client Protocol to delegate coding tasks and review results without leaving the IDE — per the GitHub changelog.
Small entry, but it's another sign Copilot is becoming IDE-agnostic rather than a VS Code feature. If your engineers live in JetBrains tools, this removes one more reason to switch editors just to get agentic coding.
GitHub also split out BYOK usage in the Copilot usage API, so teams can now see bring-your-own-key activity separately from Copilot-hosted model usage — details here.
If your org negotiated its own model contracts outside GitHub's, you can finally see what that's actually costing versus the hosted option. Useful for anyone who owns the Copilot budget line and has been flying blind on this split.
Vercel's Security Review moved into public beta, a new check built into the platform rather than bolted on — announced in the changelog.
Not much detail yet on what it actually scans for. Worth a look if you ship on Vercel, but treat it as early — betas get adjusted before they get stable.
One thing worth flagging rather than covering at length: Google is now letting schools turn off its AI tools, but only if they give up Google Search entirely too — the two are now bundled, per Ars Technica. It's not a PM tool story, but it's the same bundling instinct as the big story above, playing out as policy instead of product design.
What I'd actually do this week
Pull up your own AI feature's effect on downstream clicks or conversions. If an AI summary is answering the question before a user reaches your paywall, checkout, or ad unit, you have Google's problem at a smaller scale. Better to find it in your own data than read about it in someone else's earnings call.
If your team is on JetBrains, try Copilot through Air before your next tooling budget review. It changes the calculus on whether an IDE switch is worth the disruption.
If you have BYOK model contracts running alongside Copilot's hosted models, pull the new usage split and actually look at the cost delta. You may be paying for redundancy nobody asked for.
Reply and tell me what you're seeing — especially if your AI feature is quietly eating your own conversion funnel.