We Didn’t Need Another AI. We Needed Somewhere for the Story to Live.

By Brittany Martin, President of Chrysalis

I didn’t set out to start an AI company. Brent did—or, more accurately, Brent started building what I referred to, lovingly and somewhat dismissively, as his AI pet project while I continued doing my own work, occasionally wandering over to look at what he was making, asking a lot of questions, and then wandering back out again until it became increasingly obvious that the problem he was trying to solve had swallowed several questions I’d been circling for years.

The more I understood what he was building, the more I realized the AI was almost beside the point.

The problem was the story.

The problem started with health.

Brent has spent more than two decades living with chronic illness himself, and more than 15 years building and working with people navigating Lyme and other invisible illnesses. He created Metabolic Me in 2009 and co-founded Lyme Less Live More in 2013, where he spent years hearing versions of the same problem: people showing up foggy, exhausted, sometimes already accustomed to not being believed, while also being expected to reconstruct years of symptoms, specialists, lab work, treatments tried, what helped, what made things worse, what happened before the diagnosis, and all of the pieces that never fit neatly enough into an intake form to become part of the official record.

If you’ve ever been sick in a way that is complicated, chronic, invisible, or simply unwilling to organize itself according to the categories of modern medicine, you probably understand the strange administrative job that gets added to being sick.

The specialist knows one thing, the primary care doctor knows another, the lab company has your numbers, your watch knows you slept like garbage for four straight months, your partner knows you stopped being able to make dinner sometime around October, and somewhere in a patient portal is an extremely important PDF you were definitely going to download before the link expired.

Meanwhile, you’re left carrying the nuance and critical connections between all of it, which is unfortunate because you’re also the person whose brain may be foggy, whose body hurts, whose energy is limited, and who has possibly told this particular version of the story so many times that you’re no longer sure whether you mentioned the part where everything changed after the second medication or only thought about mentioning it.

The medical record can be accurate, the lab result can be accurate, the appointment note can be accurate, and the story underneath it all can still be missing.

That’s the part Brent understood from living it, and once I saw that problem clearly in health, I started seeing it basically everywhere.

Everybody has a piece of you.

Your therapist has one version of you while your business coach has another. Your financial planner knows the numbers but may not know that the reason you suddenly care so much about retirement is because your father got sick at 62. Your course platform knows you completed Module Four but not that something inside Module Four completely changed how you saw your marriage. Your calendar knows you canceled three meetings in one week; your best friend knows why.

Most systems know the piece of you they were designed to collect while you remain the person responsible for connecting those pieces back into a life.

Then generative AI arrived and made the context problem both much more visible and, in some ways, much easier to ignore because AI gets considerably more useful as it gets to know you. And suddenly, all of those connecting pieces are living inside a tool that seems to know you, think with you, and gives you advice.

A general-purpose model with very little information can give you a perfectly reasonable answer to a general question. Give the same system months of conversations about your work, family, health, preferences, values, habits, fears, writing, money, relationships, recurring conflicts, goals, and the oddly specific childhood memory that turns out to explain half of your adult personality, and the interaction changes.

It can remember a constraint you keep forgetting when you make plans, connect the thing you’re asking today to something you said six months ago, or reflect your own language back with enough specificity that the experience begins to feel less like retrieving information and more like being understood.

That usefulness is real. I have used AI almost every day for years and have no interest in pretending otherwise.

But I do think there’s something very strange happening that we have normalized with astonishing speed: in order to get more useful technology, we are rebuilding increasingly detailed versions of ourselves inside private companies.

Data is what happened. Context is your life with the why still attached.

My graduate research started as what I thought was primarily a privacy question and very quickly sent me into surveillance capitalism, platform economics, consent, humanistic psychology, meaning-making, epistemic justice, and reflective agency because “privacy,” at least in the way we typically discuss it, turned out to be too small a container for what I was trying to understand.

The literature increasingly converges around a larger concern: technologies that help people reflect and make meaning can genuinely support them while also turning the process of understanding themselves into extractable, predictive, and potentially monetizable material.

That research eventually gave me language for a distinction Brent and I had already been feeling our way toward inside the product:

Data is what happened. Context is your life with the why still attached.

A purchase is data; buying the plane ticket because your mother is sick is context.

A heart-rate reading is data; knowing that it spiked immediately after a conversation you had been dreading for three days is context.

A calendar entry called “meeting” is data; knowing that you have moved it four times because you’re avoiding telling someone something they don’t want to hear is context.

The data matters, but the context is often where meaning lives—and increasingly that is the information people are giving AI because it’s what allows the system to respond to this person, in this life, with this history rather than producing the statistical average of an answer.

One term from my graduate research that I keep returning to is epistemic dispossession. Chi Kwok uses it to describe the way platforms can appropriate people’s knowledge and communicative contributions while separating them from the benefits of what they produced, and it gave me language for something I had been struggling to name: the thing being extracted is not always merely a data point. Sometimes it is knowledge someone produced through living their own life.

Which is a considerably bigger question than whether your email address ends up on a marketing list.

Memory is useful. Ownership is a different question.

In many ways, I want AI to remember me. The more AI remembers me, the better it works for me.

I would very much prefer not to explain Chrysalis from scratch every time I open a new conversation, and if a system already knows how I think about strategy, what I’m building, what I refuse to call “human-centered,” and that I will almost certainly reject the first draft if it contains eleven one-sentence paragraphs trying to sound profound, that’s useful.

But that convenience comes at a cost. And is also locked into one platform.

The question Brent and I kept coming back to was what kind of memory we were normalizing, who should own it, whether the person should be able to see and correct what has accumulated, and what happens when they decide they no longer want to use the company that has spent two years learning who they are.

Right now, most of us solve that problem by starting over.

You teach ChatGPT one version of yourself and Claude another. Your coach builds context in their system, your doctor builds it in theirs, your employer gets the professional version, and you remain the only API connecting all the different versions of your life together.

So we started asking what would change if the memory belonged to the person first.

What if you could see what had accumulated about you, correct what was wrong, add to it intentionally, decide something was relevant in one relationship but none of somebody else’s business in another, and carry the useful context with you rather than having your ability to leave a platform constrained by the fact that leaving also means becoming a stranger again?

That is partly an ownership question, but it is also an agency question.

Research on reflective agency asks whether AI can support someone’s process of understanding and interpreting their own experience without taking over the interpretive work itself. Kim and colleagues describe reflective agency as the capacity to make meaning from one’s own experience autonomously and warn that things like premature summaries and overly assertive AI interventions can compress the very space where reflection happens.

I find that question much more interesting than whether an AI can produce an impressive analysis of you.

When the system helps me arrive at an insight, can I still tell which part of the meaning is mine?

The person is not the only one bringing something valuable.

Because I spend most of my professional life around people who have spent years becoming very good at something, the other side of the problem started to matter just as much.

A coach has a method that includes the workbook but also everything the workbook cannot contain. An author has the argument underneath the book and the years of research that made it possible. A practitioner has pattern recognition, boundaries, sequencing, judgment, and the competency to know when someone needs another question and when asking another question would actually be irresponsible.

Their work deserves ownership, too.

If I believe a person should remain sovereign over their story, it would be fairly inconsistent to accomplish that by treating someone else’s intellectual property as free raw material for whichever model happens to ingest it.

This eventually became the architecture underneath Chrysalis: a person carries their own context, an expert carries their own method or intellectual property, and the two can meet in a specific experience with the person deciding which parts of their context belong there.

We eventually started calling that a Context Bridge, mostly because at some point we needed a name for the thing we kept drawing on whiteboards and attempting to explain to people with increasingly elaborate hand gestures.

The architecture still has to make someone’s life better.

I’m extremely attached to large ideas, and I’m also an operator, which means I’ve spent enough of my career translating somebody’s vision into payroll, customer experience, scope decisions, timelines, processes, and spreadsheets to know that “people should own their context” is not, by itself, a reason for anyone to use our technology.

The architecture has to make something materially better.

For someone living with chronic illness, maybe that means they stop reconstructing ten years of health history every time another specialist enters the picture, while still deciding what that specialist actually needs to see. For someone working with a coach, maybe the relevant pieces of what they learned can carry forward without the coach becoming the permanent custodian of every vulnerable thing they disclosed. If someone changes AI tools next year, maybe they do not lose the context they spent this year building. If they are working through somebody else’s methodology, maybe the system can use enough of their story to make the work more specific without quietly absorbing either the story or the methodology as its own asset.

And if you’re evaluating an AI tool for yourself, your clients, your employees, your patients, or anybody else whose story you are asking a system to hold, I think the questions worth asking are becoming much more concrete:

• What does this system actually need to know for the job I am asking it to do?

• Can the person see what is being remembered about them and correct it?

• Can they share one part of their context without opening everything?

• Can they leave and take the useful memory with them?

• Who benefits from the context being accumulated?

• Is the AI helping the person continue thinking, or is it becoming increasingly authoritative about what their experience means?

Those questions came directly out of the tension I explored in Helpful and Exposed: as AI becomes more useful, it often requires more context to remain useful, which means helpfulness and exposure can arrive through the exact same feature.

I don’t think that means we stop building or stop using AI. I certainly haven’t. But I do think it means the architecture underneath the usefulness matters.

We are still early in deciding what it will mean for machines to remember us, advise us, participate in our relationships with experts, and sit beside us while we try to make meaning from the mess of being alive. The systems being built around us right now are one answer to that question, but theyre not inevitable; they’re a collection of business models, technical choices, incentives, and assumptions made by people, which means people can make different choices.

Chrysalis is our attempt to make some of them differently, while remaining very aware that we are also a company with customers, developers, infrastructure costs, payroll, and all of the other realities involved in trying to build an alternative from inside the same economic system we are critiquing.

I’m less interested in claiming we have solved that contradiction than I am in continuing to make it visible.

But I do know where I want to start: if someone is going to bring more of their story into technology because the technology becomes more useful when they do, I want the person whose life produced that context to have considerably more agency over where it lives, how it moves, who gets to use it, and what it comes to mean.

That was the question hiding inside Brent’s AI pet project.

Apparently, it is mine now, too.

Read more

My white paper, Helpful and Exposed, goes much further into the research behind these questions, including reflective agency, consent, privacy, surveillance capitalism, data ownership, and a practical framework for evaluating AI tools used in trust-based work. The paper is based on a graduate literature review spanning six bodies of research alongside private conversations with practitioners already encountering these questions in their businesses.

to your humanity,
Brittany