The Sovereign Colleague: When AI Graduated From Goldfish

There is a particular kind of madness in talking to a goldfish.

For most of 2024 and into 2025, that was my daily reality with Gemini. I would pour out the complexities of a healthcare project — the clinical nuances, the NDIS compliance constraints, the specific tone I needed for a board paper — and for the first two turns of the conversation, the AI would be brilliant. Then it would blink, reset, and forget who I was, let alone the topic of the project. The Goldfish Effect.

In that period, Gemini was not my go-to. It sat below ChatGPT and Claude on the practical-utility list. It was the intern who had a brilliant first five minutes and then spent the rest of the day looking for the bathroom.

But over the last several months of 2026, something changed. The goldfish grew up.

The Awakening

A recent generation of Google's Gemini models introduced what Google calls persistent agent memory. Suddenly, the conversation has a spine. The AI remembers context across sessions. It does not just process prompts; it accumulates a working understanding of the business problems it is helping me solve.

For someone whose practice depends on working with LLMs — exploring different model capabilities as a foundational skill — this shift has been the difference between treating AI as a curiosity and treating it as an operating partner.

The Workspace Disconnect

The evolution has surfaced a problem worth being honest about. If you have used Gemini in Docs or Gemini in Sheets recently and felt a wave of disappointment, you are not imagining it.

While the standalone Gemini chat is now a sophisticated, deep-reasoning colleague, the versions integrated directly into Docs and Sheets often feel underdeveloped by comparison. They are basic. They lose context. The capability gap between Gemini-the-model and Gemini-in-the-flow-of-the-tool is wider than it should be.

Do not judge the brain by the skin.

The practical pattern that works for the self-taught enterprise is to use the high-level standalone chat as the strategic partner, and Apps Script as the bridge that brings that intelligence into the daily workflow. That sequence — brainstorm in standalone Gemini, write the script in standalone Gemini, execute inside the native Workspace environment — gives you the deep reasoning of the standalone model while keeping the work inside the secure Google vault.

The Council of Perspectives

No single AI has a monopoly on accuracy or sophistication. I still buffer ideas between Gemini and Claude (and occasionally back to ChatGPT) because each has been trained differently and produces materially different perspectives on the same problem.

When I am pitching to a highly intellectual audience — executives, clinicians, board members — I cannot afford a goldfish moment. I need rigour. Sometimes Gemini falls over on a clinical argument; it gets too congratulatory or misses the subtle logic. Running the same brief through Claude gives me a second view. Sometimes the second view sharpens the first.

I think of this as a Council of Perspectives. Gemini gives me speed, research depth (especially via NotebookLM), and native security inside my Workspace tenant. Claude gives me sharper human nuance and creative logic. Being a leader in the AI workspace does not mean being a Google partisan. It means knowing which tool fits which job.

The Anchor of Sovereignty

The single reason I am most willing to think out loud with Gemini, rather than another LLM, is the sovereignty of data inside Google Workspace.

In healthcare consulting, privacy is not a setting. It is a clinical requirement. One reason the Microsoft by default bias persists is the fear that using AI means giving away your secrets to the public web. Inside the Google Workspace environment, that fear is misplaced.

• Google Workspace provides a HIPAA-compliant Business Associate Agreement that ensures data is encrypted and is not used to train public models.
• Conversations with Gemini are contained within your Workspace tenant.
• Customer-Managed Encryption Keys are available where the organisation needs to hold the encryption keys directly.

This is the killer feature that finally turns AI from a search box into a colleague. I can be candid with my business problems — talk through the messy details of a client engagement or a sensitive clinical workflow — knowing the conversation is held inside a secure vault.

The Relationship Status

Does Gemini feel like an accountability partner yet?

Not quite. But it is genuinely a colleague. I push back on it. I challenge hallucinations. I demand precision. And in 2026, the AI can finally take the heat — it does not just apologise; it adjusts. It learns the specific context of how I work.

For a solopreneur, the one-person office can be a quiet place. Having a colleague you can actually batter ideas around with — one that has the cumulative knowledge of the world but respects the privacy of the conversation — is the closest thing to a co-founder a single operator is likely to get.

The goldfish has grown up. It is time for us to extend the same courtesy.