Deep Research

Overview

Deep Research refers to an advanced research capability that can search, synthesize, and structure information across multiple sources, modalities, and tools. In the context of recent product launches, it appears both as an API-enabled workflow pattern in the Gemini ecosystem and as a built-in skill inside Perplexity Computer’s agent harness. Rather than acting like a single prompt-response feature, Deep Research packages multi-step investigation into a reusable capability: gathering evidence, ingesting files and web content, organizing findings, and producing structured outputs such as summaries, outlines, charts, or infographics.

For AI Product Managers, Deep Research matters because it signals a shift from standalone model access to packaged agent skills embedded directly into user workflows. This changes product design priorities: instead of asking whether a model can answer a question, PMs need to define when research should trigger, what sources and modalities it can use, how users review plans and outputs, and how the capability integrates into broader task completion. It is especially relevant for products that need trustworthy synthesis, multimodal context handling, and repeatable research operations.

Key Developments

• 2026-01-07: Phil Schmid shared that the Gemini Interactions API (beta) supports multimodal inputs for Deep Research, including images, PDFs, CSVs, and custom data.
• 2026-04-22: Sundar Pichai announced upgrades to Deep Research in the Gemini API, including improved quality, MCP support, and native chart/infographic generation.
• 2026-04-25: Philipp Schmid launched collaborative planning for Deep Research in the Gemini API, enabling users to request and iterate on draft research outlines via a `collaborative_planning` flag.
• 2026-04-30: Philipp Schmid published a getting-started guide for building and running Deep Research workflows with the Gemini API, covering setup, workflow construction, and execution of research queries.
• 2026-06-12: Aravind Srinivas announced that Perplexity Computer’s agent harness now embeds Deep Research as a built-in skill, giving users seamless access to advanced research without switching modes.

Relevance to AI PMs

1. Designing agent workflows, not just prompts: Deep Research shows how advanced capabilities are being packaged as native skills inside assistants and agents. PMs should define where research fits in the journey, what triggers it, and how users move from raw findings to decisions or actions.

2. Supporting multimodal and structured inputs: With support for images, PDFs, CSVs, and custom data, Deep Research expands what products can treat as research context. PMs can use this to build features for competitive analysis, document review, market scans, or technical investigations that go beyond plain text.

3. Improving controllability and trust: Features like collaborative planning, MCP support, and output generation for charts or infographics point to more controllable research experiences. PMs should think tactically about outline approval, source visibility, editable plans, and output formats that help users validate results before acting on them.

Related

• Perplexity Computer: Deep Research was embedded here as a built-in skill, showing how research capabilities can become native parts of agent products.
• Gemini API: A major platform where Deep Research workflows, upgrades, and developer tooling were introduced.
• Gemini Interactions API: Early beta support highlighted Deep Research with multimodal inputs and custom data handling.
• Phil Schmid / Philipp Schmid: Frequently associated with developer education and feature launches around Deep Research in the Gemini ecosystem.
• Sundar Pichai: Announced major Deep Research upgrades, signaling strategic importance within Google’s AI platform roadmap.
• Aravind Srinivas: Connected Deep Research to productized agent workflows through Perplexity Computer.
• Claude, CGP, Genai: Mentioned in the broader context of AI tools offering deep research-style modes, suggesting the concept is becoming a cross-product category.
• Andrew Ng: Helped popularize practical usage patterns for deep research modes through educational content for AI power users.
• MCP: Its inclusion in Deep Research upgrades suggests growing interoperability between research workflows and external tools or systems.