Google made a specific promise at Google I/O on May 19: Gemini 3.5 Pro would be generally available by June. Sundar Pichai, when pressed on the timeline, said "give us until next month." The audience groaned audibly. They were right to.
As of June 29, Gemini 3.5 Pro is still in limited Vertex AI enterprise preview. The public launch has been pushed to July. And in the same week the June deadline slipped, four senior Gemini researchers announced they were leaving for Anthropic. Google's AI coding teams have lost six researchers in five months.
That combination is not damning by itself. But it is a signal, and I think it's the more interesting story here.
What a Month's Slip Actually Costs
A one-month delay sounds minor. It rarely is when you're managing a product roadmap around it.
If your team planned feature launches, customer commitments, or integration timelines around Gemini 3.5 Pro hitting GA in June, you just ate a planning hit. The areas Google says it's still refining: coding output quality, token efficiency, and sustained performance on long tasks. Those are exactly the three things that matter most in agentic workflows. You either ship with what you have or you wait on a deadline that already moved once.
The "cleared for July launch" status is reassuring. But I've learned to treat that as another target, not a commitment. Google hasn't explained what specifically wasn't ready, which makes it hard to calibrate confidence in the July date.
Reading the Researcher Exit Pattern
Four senior researchers leaving for Anthropic in one week is not normal attrition. Senior ML researchers at Google don't leave for fun. They leave when they think they'll work on more interesting problems, or ship faster, or have more autonomy over what they build. I don't know the specific reasons. Nobody outside those conversations does.
But six researchers gone in five months from a team whose flagship model just missed a public deadline is a pattern I take seriously. Research-level departures take six to twelve months to show up in model capabilities. The people who left in February and March are already gone. Their work stops compounding into whatever comes after 3.5 Pro.
This doesn't mean Gemini is collapsing. Google has deep benches, TPU infrastructure advantages, and multimodal capabilities no one else can match at scale. But when you're doing model vendor selection for a multi-year project, departure rates on the research team matter as much as last month's benchmark scores.
What Gemini 3.5 Pro Was Supposed to Fix
Google has been positioning this model as the point where they get competitive again on complex coding tasks. For builders running agentic pipelines, the gap has been real: Gemini Flash variants are good at simple extraction and classification at low cost, but they degrade faster than Claude Sonnet or GPT-4.1 on multi-file reasoning, long-context debugging, and tasks that require sustained coherence across many tool calls.
If you've felt that gap in production, 3.5 Pro was the promised fix. July will tell us if it closes, or just narrows.
The Vendor Concentration Question
This is a useful forcing function for a check I think a lot of teams skip. How much of your product's capability is genuinely locked to a specific Gemini model?
Not just model calls in general, but the specific context window assumptions, the Vertex AI-native features, the multimodal pipeline that only works because you're deep in the Google ecosystem. If 3.5 Pro ships in July and underperforms, or slips again, how long does it take you to route around it?
The practical move right now: keep model routing abstracted in your stack. Know which calls are truly Gemini-specific and which could move to another provider without major rework. If more than 30% of your critical path is locked to one provider's specific model generation, that's a concentration risk worth addressing before something forces the issue.
I've been running Gemini 1.5 and 2.x Flash variants in production pipelines for a while. They're genuinely good for cost-sensitive pipelines where task complexity is bounded. They're not where I'd put a reasoning-heavy agentic chain without a fallback.
What I'm Actually Watching
The July launch is not the end of this story. The researcher departure rate matters more to me over the next six months than any 3.5 Pro benchmark. If departures continue and the model ships mediocre results on the three areas Google flagged, that's a real trajectory problem. If the model ships strong and the departures were situational, this was a blip.
Google still has meaningful advantages: Gemini Flash for cost-sensitive workloads, genuinely differentiated multimodal capabilities, and infrastructure scale nobody else can match. None of that disappears because 3.5 Pro slipped. But the combination of missed timelines and sustained talent departure is worth tracking, not explaining away.
Watch July. Trust the shipping date when you can actually test the model.
Comments
Post a Comment