Gemini Computer Use Baked Into Gemini 3.5 Flash: Screen Control Now Pairs With Search and Maps

Google made computer use a native tool inside Gemini 3.5 Flash, letting agents combine screen control with Search, Maps, and function calls in one model context. The model posts a 78.4 OSWorld-Verified score, slightly behind GPT-5.5's 78.7, but at much lower pricing of $1.50 per million input tokens and $9 per million output tokens. The launch is constructive for enterprise automation, though security risks such as prompt injection and UI drift still limit production reliability.

Analysis

The real winner is not just Google Cloud; it is any software layer that can become the orchestration substrate for enterprise workflows. Native computer-use inside the same model that already does search, grounding, and function calls collapses the integration tax that has protected point solutions and UI automation vendors. That should accelerate adoption in large enterprises because the first-order buyer is no longer looking at a bespoke multi-model architecture, but at a single agent stack with lower implementation friction and materially better unit economics.

PATH is the obvious near-term loser on narrative, not necessarily on bookings. UiPath’s core bull case is shifting from RPA to broader agentic automation, but if Google convinces CIOs that general-purpose agents can handle a growing share of GUI workflows at a third of the price, the multiple risk is real over the next 6-12 months. The more subtle second-order effect is channel compression for smaller automation startups and browser-automation infrastructure vendors that depend on being the middleware between model and application.

The market is likely underpricing the security veto. Prompt injection is not a theoretical edge case; it is the main reason enterprise deployment remains constrained, and Google’s layered defenses still do not solve the core problem. If the first wave of pilots sees even a handful of high-visibility failures, adoption could slow for quarters, which would disproportionately hurt the pure-play automation names while leaving GOOGL with a free option on eventual scale-up.

Net: this is bullish for GOOGL because it extends Gemini’s enterprise value proposition and increases developer lock-in, but the monetization inflection is months to years away, not days. Near term, the setup is better for relative value than outright beta: long Google on product breadth and distribution, short the automation exposure that is most dependent on being the default GUI layer. The key catalyst to watch is enterprise referenceability over the next 1-2 quarters; if named customers start shipping real workflows, the automation trade likely de-rates further.

Key Decisions for Investors

• Long GOOGL vs. short PATH, 3-6 month horizon: expresses the view that Google’s bundled agent stack expands enterprise share while UiPath faces valuation compression from commoditized UI automation. Risk/reward improves if enterprise pilots convert into production deployments.
• Buy GOOGL call spreads out 6-12 months, targeting the next leg of Gemini/Cloud re-rating: upside comes from higher enterprise attach rates and developer lock-in; downside is limited if adoption proves slower than expected because the thesis is already partially embedded.
• Add a tactical short PATH on any post-announcement strength, with a tight stop if management meaningfully raises ARR guidance: the market is likely to reassess the durability of RPA monetization once native model-based automation is normalized.
• Pair trade long GOOGL / short a basket of browser-automation and workflow-enablement names for 3-9 months: the structural risk is disintermediation as foundation models absorb more of the UI navigation layer.
• Do not chase broad AI beta here; instead wait for evidence of enterprise-scale deployment or an explicit security setback. If prompt-injection incidents surface, rotate toward GOOGL and away from pure-play automation, since the incumbent platform can absorb the caution better than the specialty vendors.