The Gemini 4 Argon paywall tells you where Google's margins are
Issue #058 · 4 min read
The Gemini 4 Argon paywall tells you where Google's margins are
Plus: Anthropic rebuilds Git, GitHub audits Copilot's judgment calls, and a bug found wrapped in toilet paper
The big story
An OpenAI engineer has been charged with planting a hidden recording device in a conference room.
The FBI says the device was plugged into a TV in the room. A second one turned up in the employee's bag, wrapped in toilet paper — that's The Register.
This is not a hacking story. It's an insider story, and those are the ones companies can't patch their way out of.
You can rotate API keys and scan for secrets in your codebase all day — GitHub shipped exactly that kind of tooling this week, blocking leaked Anthropic keys by default (GitHub changelog).
None of it stops someone with a badge and a reason from walking a bug into the room where the real decisions get made.
If you run a product org that handles anything sensitive — pricing, acquisition talks, safety reviews — the lesson isn't "buy a bug sweeper." It's that your threat model probably still assumes the danger is outside the building.
What shipped
Gemini 4 Argon is out, and Google wants you to notice the price tag. It's live in the Gemini app and Antigravity — but only for paying users, with developers getting a 50% API discount through November. It's also now generally available in GitHub Copilot for "complex coding, agentic, and reasoning tasks," and on Vercel's AI Gateway. Three separate distribution channels shipped the same model in one day. That's not an accident — that's Google deciding Argon is good enough to gate behind a paywall everywhere at once, which tells you more about their confidence in it than any benchmark would.
Anthropic is rebuilding Git itself, not just the workflow on top of it. Claude is generating code fast enough that normal source control is buckling, so Anthropic built something new to handle it — detailed by Simon Willison and covered by The Register. GitLab is making the same bet from the other direction, building orchestration that removes the wait between dev stages and tools to cap agent spend before it runs away from you. If your engineering team is shipping AI-generated PRs faster than reviewers can read them, this is the infrastructure problem arriving on schedule — not a hypothetical one.
GitHub now audits who's dismissing Code Quality findings and why. Admins can review manual dismissals and reopenings in the org audit log (GitHub changelog). Small feature, but it's a tell: as more review gets delegated to agents, the thing worth auditing shifts from "what did the code do" to "who overrode the warning."
What I'd actually do this week
Check who in your org can physically access rooms where unreleased roadmap or pricing gets discussed, and whether that list has grown faster than your security review of it.
If your team is testing Gemini 4 Argon, run it through Vercel's gateway or Copilot rather than the consumer app — the discounted API rate expires in November and you'll want real cost data before it does.
If engineers on your team are merging AI-written PRs faster than anyone reviews them, GitHub's new audit log for dismissed findings is worth turning on now, before the backlog of overrides becomes its own investigation.
Reply if you've had to explain to a security team why "the AI wrote it" isn't an answer to "do you understand this code" — want to hear how that conversation went.