The google/elephant-goldfish-model repository on GitHub describes itself in one line: “Design with an elephant. Review with a goldfish.” Its README adds: “Develop ideas with context and review documents with fresh eyes. EGM is an open-source plugin for AI assistants.”
In the model the README sets out, an Elephant agent holds the project context and drafts the specification, while “Goldfish reviewers examine the draft and referenced source material without the earlier conversation, helping expose unstated assumptions and blind spots”. The human is the third role: the README says the user will “make the architectural trade-offs and approve the specification before implementing”.
Business Pill · AI Agents Explained
A short Business Engineer explainer on AI agents. It teaches the general idea only and does not discuss Google or this repository.
The key insight: The README of Google’s elephant-goldfish-model on GitHub says an Elephant agent holds the project context and drafts the specification, Goldfish reviewers check the draft in fresh subagent contexts without the earlier conversation, and the user approves. The README says a review does not replace tests or human judgment.
The GitHub API, read at 02:22 UTC on 9 November 2026, shows the repository was created on 13 September 2026 under the google organisation, under the Apache 2.0 licence, with 15,225 stars and 1,107 forks. Release v0.1.0 was published on 27 October 2026 and v0.1.1 on 6 November 2026. By our arithmetic that is about 267 stars a day over 57 days.
The README lists installs for the ChatGPT and Codex desktop apps, the Codex CLI, Claude Code and Google’s Antigravity. It also states: “This is not an officially supported Googleproduct.”
The review flow set out in the README of Google’s elephant-goldfish-model on GitHub: an Elephant agent drafts with the project context, Goldfish reviewers check the draft in fresh subagent contexts, and the user approves. Star and fork counts from the GitHub API on 9 November 2026; stars a day is our arithmetic.
Review as a Quality Control
The README makes the fresh context a hard requirement: “Goldfish reviews require fresh subagent contexts. If your agent cannot create them, EGM stops the review.” It also sets a limit on what the review does: “A review does not replace tests or human judgment.”
The README credits Dave Rensin’s original essay as the source of the ideas. The pattern is published as a plugin that runs inside rival companies’ agents as well as Google’s own.
As we read the README, the qualitycontrol is the separation itself: reviewers that never saw the conversation that produced the draft.
As we read the install list, the pattern is published to run inside OpenAI’s and Anthropic’s coding agents as well as Google’s Antigravity.
As we read the README, the human approval step sits after the agent review, not instead of it, the same order Coinbase chief executive Brian Armstrong described for the agents Armstrong runs.
README, google/elephant-goldfish-model on GitHub, read 9 November 2026
“Goldfish reviews require fresh subagent contexts. If your agent cannot create them, EGM stops the review. A review does not replace tests or human judgment.”
Three Implications
FOR AGENT WORKFLOWS The README sets out three roles: an Elephant agent that drafts, Goldfish reviewers that check, and a human who approves.
FOR QUALITY CONTROL The README says EGM stops the review if the agent cannot create fresh subagent contexts.
FOR THE AGENT ECOSYSTEM The README lists installs for ChatGPT and Codex, Claude Code and Antigravity.
The analysis says: “That is no longer a person using a tool; it’s a human managing a team of agents.”
As we read it, the elephant-goldfish README writes that management step down as a workflow: agents draft and review, and the human approves. The analysis predates the repository and does not discuss it.
What Is Not Established
The repository was created on 13 September 2026, so the X post’s word “new” refers to a project about two months old; the latest release is from 6 November. The README gives no measure of how many defects the goldfish reviews catch, and GitHub stars count bookmarks, not installs or users. The README says it is not an officially supported Googleproduct.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Google’s elephant-goldfish-model splits AI design work into an agent that holds the context, fresh-context reviewers that do not see the earlier conversation, and a human who approves. The README says a review does not replace tests or human judgment. The repository had 15,225 stars on 9 November 2026, about 267 a day since 13 September by our arithmetic.
A note on sourcing. Quotations are from the README of github.com/google/elephant-goldfish-model and from Shubham Saboo’s post on X of 8 November 2026, both read on 9 November 2026. Star, fork and release data are from the GitHub API at 02:22 UTC on 9 November 2026. We did not contact Google. The stars-a-day figure is our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
