Two chatbots just hit a billion users each. One is about to run ads.
Issue #013 · 4 min read
Two chatbots just hit a billion users each. One is about to run ads.
Gemini and ChatGPT both cross a billion users the same week OpenAI starts testing ads. Coincidence is doing a lot of work here.
The big story
Same week, same number, two different companies.
ChatGPT and Gemini have both passed one billion users, according to The Verge. Google says Gemini got there faster than any product it has ever shipped, Ars Technica reports.
TechCrunch has the usage detail worth sitting with: 63% of Gemini users talk to it with their voice, and it now generates more than 150 million images a day.
That's not a chatbot anymore. That's a habit.
Here's the part that should give a PM pause. The same week OpenAI announced it had a billion-scale user base, it also announced it's testing ads inside ChatGPT to pay for free access. The company says the ads will be clearly labeled and won't influence answers.
A billion users is the number startups chase for years and never reach. It's also the number that turns a product into a media business, whether the company wanted that or not. Once you're that big, "how do we pay for this" stops being a roadmap question and becomes a business-model question — and the answer is rarely the one users asked for.
If you're building anything that assumes ChatGPT stays ad-free and answer-neutral forever, this is the week to write down that assumption and put a date next to it.
What shipped
GitHub Copilot for JetBrains now has persistent memory and local model support via Ollama. The changelog says it also adds enterprise controls and fixes reliability issues across MCP servers. If your engineers live in IntelliJ or PyCharm, this closes a real gap — Copilot chat has felt disposable compared to tools that remember what you told them yesterday.
Microsoft's MAI-Code-1.1-Flash is now in GitHub Copilot, and the old version is on a clock. The 1.1 release adds vision support — it can now read a screenshot, not just code. The original MAI-Code-1-Flash gets deprecated September 10. If your team has anything hardcoded to that model name in a workflow, that's a four-week window to catch it before it breaks quietly.
GitHub also shipped a per-model breakdown of token usage in its AI usage report. Small change, but a useful one — the changelog entry means you can finally see which model on your team is burning through credits, instead of getting one blended number and guessing.
Worth noting what didn't make the cut: Vercel posted five separate changelog and blog items this week. None of them changes what a PM does on Monday, so we're skipping them rather than padding the section with one vendor's release notes.
What I'd actually do this week
- Pull up your ChatGPT usage in any customer-facing workflow and ask what happens to it under an ads model. OpenAI says answers stay independent of ads — read that claim carefully rather than assuming it survives contact with a growth target.
- If you use GitHub Copilot, check any CI script or internal doc that names MAI-Code-1-Flash directly. You have until September 10 before it's gone.
- If your team runs JetBrains IDEs, try Copilot's new memory feature on one real task this week and see whether it actually remembers context across sessions or just claims to.
Reply and tell me what broke, or what didn't.
Tools mentioned
- ChatGPT ads test
- GitHub Copilot for JetBrains
- MAI-Code-1.1-Flash
- MAI-Code-1-Flash deprecation
- Per-model token breakdown