Google Gemini 3.5 Pro delayed twice, Kavukcuoglu takes over DeepMind, and what repeated missed ship dates mean for builders evaluating Google for agentic workloads
When Ship Dates Become a Pattern
There is a difference between a delay and a pattern. One delay is engineering. Two delays on the same model, after a public promise at your flagship developer conference, is something else. It is a signal about process, culture, or both. And right now, Google is giving anyone evaluating their platform for serious agentic work a very clear signal worth paying attention to.
The Scoreboard as of August 2026
Anthropic has Mythos in general availability. OpenAI has GPT-5.6 in general availability. Google has a price cut on Gemini 3.7 Flash and a new boss at DeepMind.
That is the actual state of the frontier right now.
Gemini 3.5 Pro was announced at Google I/O in May with a promise to ship “next month.” It missed June. It missed mid-July. Then, according to Fortune’s reporting on the DeepMind leadership shakeup, internal testing showed performance still lagging behind rivals in August, triggering another two-month delay. Per CNBC, Google hasn’t unveiled a frontier model since early 2026. Meanwhile both competitors have had their 2026 flagship models shipping to customers.
The Leadership Change at DeepMind
On August 5, Demis Hassabis transitioned from CEO to chair of DeepMind. Koray Kavukcuoglu, previously the chief scientist, stepped into the top role. Sergey Brin has reportedly been applying pressure to accelerate Gemini development.
I want to be careful here. Kavukcuoglu is a serious researcher with real credentials. DeepMind has some of the best AI talent in the world, full stop. This is not a talent problem.
The problem is the gap between what DeepMind produces in research and what Google ships to customers at competitive pace. Gemini 3 briefly led benchmarks last November. Within weeks, Anthropic and OpenAI responded with updates, and Google was back in catch-up mode. That cycle, lead then slip then delay, has now repeated enough times to stop being surprising and start being structural.
What This Means for Agentic Workloads
If you are architecting agentic systems right now, model reliability and predictable iteration cadence matter as much as raw benchmark performance. Your agents will depend on specific model behaviors. When a provider misses ship dates, it is not just annoying. It means your roadmap is hostage to their internal testing calendar.
Anthropic and OpenAI have both demonstrated they can move fast and ship. Google has demonstrated it can announce fast. Those are different capabilities, and they compound over time in an agentic world where you are chaining calls, building tool integrations, and expecting the model underneath to get incrementally better on a predictable schedule.
The Flash price cut is real and worth noting for high-volume inference workloads. But a cheaper model that was already available is a different value proposition than a new frontier model that keeps slipping.
The Structural Question Nobody Is Asking Loudly Enough
Why does Google keep losing its lead within weeks of establishing it? The benchmark gap between Gemini 3 in November and where Anthropic and OpenAI responded suggests the competitors had work in flight that was close to ready. Google either doesn’t have that same depth of overlapping work in the pipeline, or the internal process for getting it out the door is slower than it needs to be. Possibly both.
Kavukcuoglu’s real job isn’t running DeepMind’s research. DeepMind’s research is fine. His real job is making Google’s AI development function like an organization that can ship at the pace the market currently demands. That is a harder problem than any benchmark.
If the next two months produce Gemini 3.5 Pro and it genuinely competes with Mythos and GPT-5.6, I will update my read on this. But the credibility hole Google has dug with two consecutive missed dates means even a strong launch will face skepticism about what comes next and when.
That is the real cost of a pattern.
Sources
#GoogleDeepMind #GenerativeAI #AIEngineering #AgenticAI #MLOps
Sources & Further Reading
- Behind the exit of DeepMind’s CEO
- Google’s new AI boss inherits a race to catch OpenAI and Anthropic
- EXCLUSIVE: Inside the Google executive moves that led to its big AI reshuffle
- Google Cuts Gemini 3.7 Flash Price in Half as It Claims to Top Claude in Business Workflows
- Google co-founder Brin pressures DeepMind to accelerate Gemini as Alphabet reshuffles AI leadership
