At this year’s Google I/O conference, the Mountain View-based giant made a move that clearly redefines the current stage of the technological arms race. The launch of the Gemini 3.5 model family is more than just another routine performance update. It is first and foremost a strategic shift of emphasis from pure performance in benchmarks to operational usability and agent work. In the face of increasing pressure from OpenAI and Anthropic, Google is betting on personalised task management and automation to attract key business customers.
Inference economics at the heart of the new offer
At the heart of the new strategy is Gemini 3.5 Flash, a lightweight and significantly cheaper model that will become the default Gemini application engine. Google declares that Flash delivers frontier model-level intelligence, running around four times faster than comparable solutions. Crucially for CFOs’ budgets, its implementation costs less than half of that of existing solutions.
In an enterprise world where the cost of inference (inference) is becoming a major burden inhibiting the scaling of artificial intelligence, such price optimisation is a direct hit to the competition. Alongside Flash, the portfolio saw the beta release of Gemini Spark – a general-purpose agent capable of inferring from data from connected applications (including via the personalised Daily Brief assistant) – as well as the multimedia Omni model and the heavyweight flagship 3.5 Pro, which Google is currently using internally and the market will receive next month.
The struggle for an autonomous enterprise
Google’s move is perfectly in line with the market’s turning point, in which companies are moving away from simple digital assistants towards fully autonomous systems. According to recent reports from Gartner, as many as 80% of CEOs believe that AI agents and autonomous coding tools will force fundamental changes in the operational capabilities of businesses.
However, competition in this field is extremely tight. The territory of coding agents has just been entered by xAI with its advanced Grok Build model. Anthropic and OpenAI, on the other hand, with a glimpse of potential IPOs, are aggressively fighting to maintain their leadership positions. Data from the Ramp AI Index shows that although half of US companies already pay for subscriptions to AI tools, the lion’s share of this market is currently controlled by these two rivals.
The ace up your sleeve: Distribution and the ecosystem
To turn this balance of power around, Google is not just relying on model architecture alone. The key to success is supposed to be the Gemini Enterprise Agent Platform and deep integrations with external ecosystems such as Salesforce Agentforce or workflows from Databricks. New features such as the Agentic Data Cloud and a special orchestration layer are expected to reduce the integration friction that has so far blocked many pilot projects from entering the production phase.
With a 14 per cent share of the global cloud market in the first quarter of 2026, Google is admittedly second only to the largest infrastructure players, but has a unique distribution advantage. Its billions of Search and Workspace users give the corporation a reach that start-ups cannot quickly copy. In the final analysis, it is the ease of implementation into existing processes, not fractions of a percent in lab tests, that will determine who wins the mass business market.
