AI-Wiki

Google Research

4 min read

Google Research

Confidence 0.80 · 5 sources · last confirmed 2026-06-12

The research arm of Google, focused on advancing fundamental and applied research in AI, computer science, and adjacent fields. Producer of the Gemini model family (cited extensively in the wiki — the 280×-cost-reduction marker comes from Gemini-1.5-Flash-8B), and a recurring data partner of the AI Index (the 2025 and 2026 AI Index editions disclose Google Research as an analytics/funding partner).

This entity page covers the research organisation; for product-line discussion (Gemini family, foundation models broadly) see foundation-models.

Appears in this wiki via

  • 2022-10-06-yao-et-al-react-synergizing-reasoning-acting — publisher (Brain team, with Princeton) of the foundational ReAct paper (ICLR 2023). The wiki’s earliest primary research source and the genealogical root of its agent/harness cluster. Authors Shunyu Yao (Princeton intern) and Yuan Cao (Brain team) were promoted to entities on this ingest.
  • 2022-11-08-yao-cao-react-google-research-blog — the Google Research Blog popularisation of the same ReAct work (8 Nov 2022). The accessible register paired with the rigorous paper.
  • 2026-05-07-globerson-et-al-scalable-measurement-durable-skills — primary publisher of Towards Scalable Measurement of Durable Skills (preprint, 12 April 2026). Globerson et al. (~40 co-authors) introduce the Vantage assessment platform, Executive LLM steering protocol, and Gemini-based AI Evaluator — anchoring the wiki’s durable-skills concept.
  • 2026-04-21-forsgren-macvean-build-core-skills-thrive-ai-era-developer — Google I/O 2026 talk by Nicole Forsgren and Andrew Macvean, both leads on Google’s Developer Intelligence team. The talk anchors the DORA programme (Forsgren-led) as a Google-Research-lineage empirical backbone for two load-bearing claims: (a) AI’s productivity paradox at the team-level (the engineering-team correlate of micro-productivity-trap); (b) AI as amplifier-and-mirror — magnifying existing organisational strengths and weaknesses. Sister-vantage to the Globerson et al. paper: both Google-Research-lineage outputs map durable / non-routine engineering skills, one from psychometrics, the other from inside-Google measurement.
  • Mentioned across multiple wiki sources as a major model contributor (Gemini family is referenced in AI Index 2025 §1 — Gemini-1.5-Flash-8B as the cost-reduction marker; AI Index 2026 continues to track Google’s models alongside OpenAI, Anthropic, Meta, and others).

Notable people (in this wiki)

  • Yossi Matias (dangling — co-author on the durable skills paper; promote on 2nd-source coverage) — VP and Head of Google Research.
  • Nicole Forsgren (dangling — first-mention; high probability of recurrence given DORA stature and Accelerate authorship) — lead, Developer Intelligence team; DORA founder.
  • Andrew Macvean (dangling — first-mention) — lead, Developer Intelligence team.
  • ~40 other co-authors of the Globerson et al. paper, currently listed in the source-page frontmatter but not promoted to individual entity pages (single-source coverage).

Research programmes referenced in this wiki

  • Developer Intelligence team — Google’s in-house team studying how Google engineers work with AI. Anchor for Forsgren & Macvean 2026; named on stage as “measuring the engineering system and all the pieces within that … the end-to-end time to go from an initial seed of an idea all the way through to getting that idea out into a real user’s hands.”
  • DORA (DevOps Research and Assessment) — long-running Forsgren-led programme running annual State of DevOps reports since 2013; provides the empirical backbone for productivity-paradox and amplifier-and-mirror claims cited in the I/O 2026 talk. First wiki-reference here; promote to its own entity page on second-source coverage.

Funders / partnerships visible in this wiki

  • AI Index — Google is listed as a supporting funder of the AI Index 2025 and 2026 editions (2026-04-30-ai-index-report-2026).
  • OpenMic, NYU, UT Austin — academic and startup collaborators on the durable skills work.

Notable products / models

The full model roster is tracked in foundation-models. Selected highlights:

  • Gemini family (1.5, 2.5 Pro, 3.0) — Google Research’s primary multimodal foundation model line. Cited as the substrate for the Vantage Executive LLM and AI Evaluator.
  • Veo 2 (video generation) — referenced via AI Index 2025.

Open questions

  • The relationship between Google Research and Google DeepMind — separately structured units that often co-author. The Globerson et al. paper is published under “Google Research” affiliation; whether DeepMind shares authorship would clarify the institutional boundary.
  • Google’s broader AI Index funding/data-partner role — which surveys, which methodological choices.
  • The institutional boundary between Google Research (proper) and the Developer Intelligence team that Forsgren and Macvean lead — the latter’s outputs map cleanly to engineering-research even if formally attached to Developer Relations rather than the Research arm. Worth clarifying on a third source.
  • DORA’s published methodology for the AI-era productivity-paradox finding. Forsgren cited the result live; a published State of DevOps or DORA AI-specific report substantiating the individual-vs-team divergence would meaningfully strengthen the wiki’s micro-productivity-trap page.

Graph View

Backlinks