Hands on with Gemini 4 Argon

Summary
This video is an official launch showcase for Google DeepMind’s Gemini 4 Argon, hosted by Paige Bailey (Google DeepMind). She speaks with industry partners Curtis Liu (CTO & Co-founder of Amplitude), Gopal Srinivasan (Global Google AI Practice Lead at Deloitte), and Yusuke Kaji (GM of AI for Business at Rakuten) about their benchmarks, multi-agent architectures, and real-world deployment results using the model.

What is shown

• [00:00] Curtis Liu discusses Gemini 4 Argon’s persistence in retrieving and inferring answers from data.
• [00:08] A brief look at Amplitude’s internal interface running prompt evaluations on a laptop.
• [00:16] Title sequence: “Hands on with Gemini 4 Argon”.
• [00:19] Gopal Srinivasan describes using the model for domain-specific knowledge work and complex reasoning tasks.
• [01:03] Yusuke Kaji discusses Rakuten’s internal benchmark testing of Gemini 4.
• [01:20] Amplitude’s data visualization workflow shown on a laptop, including an automated Onboarding Funnel Summary report and charts.
• [01:31] Amplitude’s Langley Admin Dashboard showing Gemini 4 making sequential tool calls via MCP (amplitude_mcp_get_properties, amplitude_mcp_validate_chart_definition, amplitude_mcp_query_dataset).
• [01:44] An analytics breakdown identifying UI drop-off causes and proposed technical fixes.
• [02:10] Deloitte’s Finance 360 dashboard, showing the CFO Executive View, Workflow Monitor, Automated Reconciliation Orchestrator, multi-agent process traces, and a financial ERP knowledge graph.
• [02:53] Closing title graphic for Gemini 4 Argon.

Claims & numbers

• Gopal Srinivasan claims that in Deloitte’s evaluations, long-running agent tasks achieved a 100% completion rate without failing [00:58, 02:22].
• Yusuke Kaji claims Gemini 4 performed significantly better than previous frontier models on Rakuten’s private benchmark without increasing costs or latency [01:05].
• Curtis Liu states that Gemini 4 scores over 90% on Amplitude’s internal benchmarks for accurately interpreting customer questions and generating data charts [01:28].
• Gopal Srinivasan claims Deloitte condensed a complex workflow that previously required about a dozen sub-agents into just two agents due to Gemini 4’s native multi-system tool-calling and reasoning capabilities [02:02, 02:11].
• Gopal Srinivasan claims the Deloitte financial agent fleet processes over 7,000 general ledger accounts overnight without errors [02:26, 02:40].
• Curtis Liu states that Gemini 4 delivers “frontier-tier level quality for half the cost” [02:45].

Notable quotes

• [00:00] “It works hard. Like, it’s very persistent in trying to get things.” — Curtis Liu
• [00:57] “Long-running tasks, 100% of the time in our evals have completed successfully.” — Gopal Srinivasan
• [02:45] “You’re getting a frontier-tier level quality for half the cost.” — Curtis Liu

Assessment
This is an official launch case-study video combining interview testimonials with recorded enterprise UI demonstrations. The screens show authentic tool-calling traces and dashboards, but the presentations are curated and edited summaries of partner deployments rather than raw, live hands-on testing.

Described by gemini-3.8-flash on 2026-10-13 from the video’s audio and frames.