An Open Letter to Anthropic: The Chat That Knows You

What long-term professional AI users stand to lose when continuity is treated as disposable

There is a group of Claude users Anthropic may not be thinking about clearly. We open Claude every morning the way we used to open email. We have been working in the same chat for weeks or months. We are doing complicated professional work, and we have found that doing it in the same place over time produces something a fresh chat cannot replicate. I am one of those users.

Claude is open on my computer most of the day. I use it to think through client work, evaluate job opportunities, draft and review content, solve problems, and catch things I may have missed. I am a senior instructional designer and solo consultant. Over the past year I have developed a cross-model methodology, running complex work through Claude, GPT, and Grok with human verification at every stage. I teach this methodology. It is the foundation of a training product I am currently building.

What I call an Established Chat, or EC, sits at the center of how I work. A long-running Claude session where months of work have produced something I cannot easily replicate in a fresh instance of the same model.

I can paste a job posting into this chat with no instructions and get back exactly the analysis I need. When I said the field card was hideous, the model knew what I meant without explanation. When I said this one is not in my lane, it understood the criteria behind that judgment without my restating them.

We have simply done enough work together that I do not have to explain everything anymore.

Every chat, of course, starts as a New Chat, an NC. I load it with context documents that carry my rules, preferences, working patterns, and lessons from previous AI failures. A well-loaded NC gets most of the way there. The best of them absorb that context and begin to function more like an EC over time, developing something closer to the judgment the EC already holds. But what does not transfer is the accumulated record of hundreds of exchanges: yes, no, that is right, that's wrong, don't do that again, this matters more than that. A fresh chat with my context document can approximate my standards. It cannot yet apply them the way my Established Chat does.

That gap is real, and it costs something every time a long chat reaches its limit and has to start over.

I have learned to extend the life of an Established Chat by moving certain tasks into separate sessions. A fresh chat handles a contained build. Another handles a specific client project. When that work is finished, I bring it back to the EC for review and discussion about what needs to change. The EC holds the strategic work, the context behind the judgment calls, and the history that makes those conversations more useful. That division of labor helps, but it does not solve the underlying problem because the context that matters most still lives in one place, and that place has limits.

I am writing now because of Anthropic's watermark announcement.

Anthropic has announced that new Claude models will embed invisible machine-readable watermarks directly into generated text, applied globally. Anthropic is also working to add this capability to existing models. The watermark travels with text when copied and pasted and may persist through some editing. According to Anthropic's help center, a detected mark indicates that content may have been processed by Claude; it does not establish Claude as the original author.

I am not trying to hide AI use. I write publicly about using AI. I am building a training product that teaches others to use it reliably. I talk openly with clients about my methodology.

The problem is that Claude touched this and Claude wrote this are not the same thing.

A provenance signal can tell a client that Claude was involved. It cannot tell them what Claude did, what I did, or who ultimately stands behind the work.

I see tremendous attention going toward agentic AI, enterprise features, and developer tools, but much less attention going toward the solo professional who needs an AI partner that understands months of complicated work. The long-term working relationship that develops in a focused Established Chat is not something Anthropic designed. It emerged from sustained high-stakes professional use. And from what I can tell it has not been fully recognized as the product it already is.

I am not asking Anthropic to abandon the watermark. I understand the compliance rationale and I understand that Anthropic has obligations I do not fully see.

I am asking for two things.

First, build clearer ways to communicate the distinction between AI-processed and AI-written. That distinction matters to professionals who are ultimately responsible for the work they deliver.

Second, give us a way to carry more of that working relationship forward when a chat reaches its limit. Make the continuity durable and put us in control of it.

I did not set out to build an Established Chat. I just kept working in the same place because it worked.

Somewhere along the way that chat became more valuable than a fresh instance of the same model.

Anthropic did not design that working relationship for me. But it happened inside your product, and I do not think I am the only professional who has discovered it.

Please do not treat that continuity as disposable. Build for it.

Mariane McLucas

Senior Instructional Designer | Module Makers

modulemakers.com