Google is betting that its AI story is bigger than any single model delay, but the market isn’t buying it yet. On Wednesday’s earnings call, Sundar Pichai defended the postponement of Gemini 3.5 Pro (still in partner testing) by pointing to Gemini 3.6 Flash, which improved more than 10 points on a coding benchmark while consuming fewer tokens. Cloud revenue grew 82 percent, well above the 64 percent analyst estimate. Yet shares fell more than 3 percent after hours, and the stock is down roughly 9 percent since late April. Pichai’s Gemini 4 roadmap<\/a> promises monthly model releases and a “very ambitious” frontier model, but the timeline remains unspecified.<\/p>\n If your engineering org has standardized on Gemini for coding assistance or agentic workflows, the delay isn’t academic. The companies that moved fastest on AI coding tools in 2024 picked a frontier model and built workflows around it. Google is now openly conceding ground in that specific category while asking you to wait for Gemini 4. That’s a real gap window, and your developers are already aware of what Anthropic’s Claude and OpenAI’s models can do right now.<\/p>\n The Flash vs. Pro distinction matters more than Google’s messaging suggests. Gemini Flash is a smaller, faster, cheaper model optimized for high-volume, lower-complexity tasks, think customer service routing, document summarization, or cybersecurity alert triage. It is not the same tool as a frontier reasoning model. Google’s pivot to highlighting Flash on the earnings call is a legitimate product story for operations and infrastructure use cases, but it doesn’t answer the question analysts were actually asking: can Gemini compete with GPT-4o or Claude Opus at the hard reasoning and code-generation tasks that CTOs are trying to automate? Pichai said yes, eventually. That’s a different answer than yes, now.<\/p>\n The $195 to $205 billion capex commitment is the number that should recalibrate your vendor confidence model. Google is spending at a scale that makes retreat from AI infrastructure essentially impossible, which means Gemini 4 will ship. The falsification condition for Google’s ecosystem argument is simple: if Gemini 4 launches without recovering benchmark leadership in coding and agentic tasks by Q1 2026, the Flash-focused “portfolio” framing stops being a strategy and starts being a consolation prize. Until then, a hedged two-vendor posture, Google for volume inference, a competitor for frontier reasoning, is the position that carries least regret.<\/p>\n Agentic coding refers to AI systems that don’t just suggest lines of code but autonomously plan, write, test, and iterate across an entire software task, like a junior developer who can hold a whole feature in mind rather than autocompleting one line at a time. It’s the difference between autocorrect and a self-directed contractor. For CTOs, it’s the capability that determines whether AI cuts developer headcount requirements or merely speeds up individual keystrokes, which is why Google’s acknowledged gap here carries strategic weight.<\/p>\nWhat this means for your business<\/h2>\n
Concept deep-dive: Agentic coding<\/h2>\n