Google’s promised leap forward in AI coding has hit a wall. At Google I/O in May, CEO Sundar Pichai told the world that Gemini 3.5 Pro — the company’s next flagship model — would ship in June. It’s now the end of July, and the model is nowhere to be found. According to a Bloomberg report cited by Reuters on July 16, the release has been pushed back months while Google scrambles to improve its coding performance. For Windows developers and IT teams who had planned to evaluate the model against Microsoft’s deeply integrated Copilot ecosystem, the delay is more than a calendar slip — it’s a signal that Google’s AI machine may be grinding gears just when it needs to accelerate.

The absence of Gemini 3.5 Pro is not simply a missed deadline. It exposes a pattern that has dogged Google for years: brilliant research, enormous resources, but a bureaucratic tangle that slows the leap from lab to product. Under the hood, engineers are reportedly frustrated, and overlapping AI projects across Google Cloud, DeepMind, Android, and consumer teams have created what one former employee described as a “boil the ocean” problem. The result: a flagship model that was supposed to anchor Google’s enterprise AI push is stuck in testing, while rivals ship and gain ground.

What Actually Happened

Google I/O 2025 opened with a clear message: the next-gen Gemini 3.5 family would lead the company’s AI charge. The lower-tier Gemini 3.5 Flash did arrive in June, along with a wave of Gemini-powered features across Search, Workspace, and Android. But the higher-end Pro model — the one aimed at developers and enterprise workloads — never materialized. I/O keynote materials and a company blog post explicitly pointed to a June launch for 3.5 Pro.

As weeks slipped by without an update, Bloomberg’s reporting — summarized by Reuters — filled in the blanks. At least ten current and former employees described internal frustration over the delay. The core issue: Gemini 3.5 Pro’s coding abilities were falling short of internal goals. That’s a critical vulnerability. Coding assistants have become the proving ground for large language models. Enterprise buyers judge a model not by how it writes poems, but by how reliably it navigates a codebase, suggests fixes, and integrates with tools. Google’s own Sundar Pichai had acknowledged after I/O that the company was behind in coding. The admission, coupled with this delay, suggests the gap is wider than originally thought.

Compounding the technical hurdles, the report paints a picture of internal fragmentation. Multiple factions — Google Cloud, DeepMind, the Android team, and even consumer product groups — are all building overlapping AI coding tools. Decision-making involves layers of stakeholders, making it hard to align on a single release strategy. “Getting everyone to move in the same direction is like trying to boil an ocean,” one former employee said.

Google has not publicly confirmed the Bloomberg details or offered a revised timeline. But the gap between the May commitment and the current silence is deafening for a company that declared a “code red” over AI in 2022 and restructured its research arms specifically to accelerate.

What It Means for You

If you’re a Windows user simply dabbling with Google’s AI through the Gemini web app or Workspace, the delay has little day-to-day impact. Gemini 3.5 Flash continues to serve conversational queries and document summaries just fine. The current Pro model (version 2.5) remains available for complex tasks.

But for Windows developers and IT administrators, the stakes are higher. Many organizations have been running head-to-head evaluations of AI coding assistants, comparing GitHub Copilot, Amazon CodeWhisperer, and Google’s Gemini tools. The promise of Gemini 3.5 Pro was a model that could finally match — or beat — OpenAI’s GPT-4 class on code generation and context handling, all while tying into Google Cloud’s infrastructure. That evaluation now stalls.

Practical consequences:
- Windows development workflows: Teams that planned to integrate Gemini 3.5 Pro into Visual Studio Code or JetBrains IDEs via Google’s Cloud Code plugins will need to stick with older, less capable models or switch horses. Microsoft’s GitHub Copilot, which now runs on GPT-4o and integrates natively with Windows 11 development tools, becomes a more compelling default.
- Enterprise AI strategies: Companies that spread bets between Microsoft Azure and Google Cloud may re-weight toward Azure if Google’s flagship model continues to lag. The delay also complicates any plan to use Gemini 3.5 Pro for internal code generation or automated testing, forcing teams to either wait indefinitely or accelerate adoption of alternatives.
- IT budgeting: For shops that have already provisioned Google Cloud AI services in anticipation of Gemini 3.5 Pro’s arrival, the lack of a timeline introduces uncertainty. Cloud commitments often come with reserved capacity; without a firm release date, that capacity may sit underused.

Home users and small dev shops face a simpler choice: the AI coding assistant market is crowded. If you were curious about Google’s offering, there’s no reason to wait. Microsoft’s Copilot, Anthropic’s Claude, and open-source models like Meta’s Code Llama are all viable today.

How We Got Here

Google’s AI struggles are not new, but the context has shifted. After ChatGPT blindsided the industry in late 2022, executives famously declared a company-wide “code red.” The immediate result was Bard — a rushed chatbot that stumbled out of the gate. In response, Google merged its previously separate AI teams: Google Brain and DeepMind were placed under Demis Hassabis, with a mandate to unify research and ship faster. For a while, it worked. The Gemini 2.0 and 2.5 releases closed the performance gap with OpenAI and Anthropic, and Google started weaving AI into every product.

But coding turned out to be a tougher test. By early 2023, Microsoft had already seized the narrative with GitHub Copilot, built on OpenAI’s Codex model. Google’s own coding tools — such as the Duet AI for Google Cloud (later rebranded to Gemini Code Assist) — lagged in capability and developer mindshare. Pichai’s post-I/O admission that Google was behind in coding signaled that the company knew it had work to do. The reported delays with Gemini 3.5 Pro suggest that work is proving harder than anticipated.

Internal dynamics haven’t helped. Bloomberg’s sources describe a company where multiple AI coding projects compete for resources. Google Cloud wants a coding assistant to sell alongside its platform. DeepMind wants to push the frontier of reasoning. Android sees on-device code suggestions as a differentiator. And consumer teams want coding help for Google’s own developer community. Without a single owner, progress slows.

This isn’t an indictment of Google’s raw talent. The company invented the transformer architecture behind modern LLMs. But as the Llama situation at Meta showed (with the perpetually delayed Llama 4 “behemoth” model), size and research prowess don’t guarantee smooth execution. For now, Google looks like a giant with too many hands on the steering wheel.

What to Do Now

If you’re waiting for Gemini 3.5 Pro to make a move, stop waiting. The original June timeline is dust, and Google has not offered a new one. Here’s how to proceed:

  1. Audit your current AI coding tools. If you’re already on GitHub Copilot, Amazon Q Developer, or another assistant, check whether a delayed upgrade changes your evaluation schedule. Don’t freeze any decisions on the hope that Google will ship soon.
  2. Test the available Gemini models. Gemini 2.5 Pro is still online and can handle complex reasoning tasks. If your use case doesn’t demand bleeding-edge code generation, it may suffice. Compare it against GPT-4o or Claude 3.5 Sonnet in your specific codebase.
  3. Watch Google’s Q2 earnings on July 22. Alphabet reports earnings this week. Investors will press leadership on the Gemini timeline. Pichai or CFO Ruth Porat may provide a status update, though concrete release dates are unlikely on an earnings call.
  4. Evaluate multi-model strategies. Relying on a single AI provider is risky. Consider architectures that let you swap models easily. If you’re on Google Cloud, Vertex AI already supports models from Anthropic and others, so you can pivot without leaving the platform.
  5. For Windows shops, lean into the Microsoft stack if it fits. Microsoft’s end-to-end integration — Copilot in Windows, GitHub, Teams, and Azure — is a practical advantage that Google can’t yet match. If your environment is already Microsoft-centric, this delay only reinforces that path.

There’s no sign that Google has canceled Gemini 3.5 Pro. The model is still in testing, and the company has too much riding on it to scrap it. But the window for Google to establish technical parity with coding frontrunners is narrowing.

Outlook

The next few weeks will be telling. Alphabet’s earnings call on July 22 may offer clues, but the real test will be whether Google can show — not just tell — that it can ship a top-tier coding model before the end of the year. If internal fragmentation really is the bottleneck, we may see another reorganization, perhaps consolidating all coding efforts under a single leader. For Windows developers and IT buyers, the message is clear: rewards go to the fast and the focused. Google has the raw materials, but until it ships, the skepticism is warranted.

In the meantime, the AI coding race isn’t waiting. Anthropic’s Claude, OpenAI’s latest GPTs, and even open-source models are setting a blistering pace. Google’s AI clans may yet unite, but for now, they’re still fighting on too many fronts.