A Context can be very well documented and still be incomplete.

It can hold the right standards, the right articles, the right documentation pages, the right customer interview transcripts. It can have access to a clean knowledge base, to well-sorted user feedback, to properly tagged support tickets.

And yet it can be missing something essential: the fine-grained understanding the PM carries in their head.

Not some hidden grand theory. Rather everything that builds up over the course of the work: a phrase said in passing, a nuance grasped after three exchanges, a customer fear that doesn't show up in a transcript, a political tension, an interpretation no one has written down yet.

That is precisely why session sources are needed.

What a Context is

A Context is a workspace bounded by a mission.

It brings together an objective, a focus, directives, sources, notes, drafts, and sometimes deliverables. Its role is not merely to store information. It serves to give an AI the right working scope, the right references, the right constraints, and a usable memory.

Put differently: a Context avoids starting over from scratch in every conversation.

But to be truly useful, it can't just receive documents. It also has to capture what the team learns progressively during the work.

Three families of material

A Context feeds on several types of material. Confusing them quickly creates bad workflows.

The first family is documentary curation.

Here you find books, articles, standards, product documentation, reference frameworks, best practices. These sources give the Context a by-the-book mastery. They don't fall into the right context automatically: you have to choose them, add them, keep them, then ingest them.

The second family is industrialized user research.

Here you find full interview transcripts, in-app feedback, support tickets, structured feedback, verbatims collected in a CRM or a product tool. Here, part of the collection can be automated. These traces already exist as documents or as usable objects.

The third family is unwritten knowledge.

It's what the PM knows, understands, senses, or connects, but which does not yet exist as a stable source. That is where the session source becomes useful.

What a session source captures

A session source is not one more document.

It is a channel for pouring still-informal knowledge into the Context.

Sometimes this knowledge arrives as a long discussion with the AI. The PM explains the customer, reworks a phrasing, corrects a hypothesis, adds details. The session then becomes genuine raw material for the work.

But it isn't always that substantial.

A session source can also receive a small idea that came up after the fact. A phrase overheard in a hallway. A customer remark at the end of a call. A tension the PM is only beginning to name. A nuance that doesn't warrant a full document, but that deserves not to disappear.

The session source captures useful thinking before it becomes a document.

Why accumulation changes everything

A single isolated phrase can seem weak.

It doesn't always justify a ticket. It doesn't necessarily deserve a synthesis note. It doesn't always appear in a transcript. In a classic information system, it would often be lost, or stored somewhere without producing much.

But dozens of small points accumulated over time can change the quality of a Context.

They surface patterns. They reinforce intuitions. They reveal nuances. They make it possible to understand that a customer isn't just asking for a feature, but is trying to reduce a risk, preserve control, avoid an internal tension, or work around an organizational constraint.

That is where a Context brings something that mere storage does not.

It can reread, connect, compare, stabilize. It can turn an accumulation of micro-signals into notes, hypotheses, angles of analysis, or better-framed decisions.

So the important thing is not the single small phrase.

It's the buildup of small phrases, in a context that knows why it keeps them.

What it is not

A session source is not a full transcript.

If a meeting or an interview has been transcribed, the transcript is an immutable source. It should be added to the context as such.

A session source is not a meeting note either.

A meeting often has its own scope: a project, a feature, a customer, a decision. Its note or transcript can feed the dedicated context. That does not automatically turn the note into a session source.

A session source is not a raw chat log.

The entire history of a conversation is not meant to become a source. What matters is the material deliberately poured into a Context that has a goal and a focus.

Finally, a session source is not a documentary base.

The documentary base preserves references and traces. The session source captures what is being formulated around those references and around the field.

The conditions for success

A session source only has value if it stays framed.

First, it needs a goal and a focus. Without that, it becomes a catch-all. With a good focus, it captures the useful material without deciding its final form too early.

Next, the raw source and the stabilized knowledge must be separated. The session source keeps the primary material. The notes, for their part, formalize what deserves to be reused.

Finally, it needs an ingestion capable of handling repetition.

Saying the same thing three times with three different nuances is not a problem. That's often how an idea matures. The role of ingestion is not to duplicate three notes, but to accrete the nuances into finer knowledge.

The session source therefore calls less for a logic of archiving than for a logic of maturation.

Giving the Context a channel to learn from the human

Organizations are getting better and better at producing sources.

They document, transcribe, centralize, index, and automate the collection of user feedback. This is necessary.

But part of the useful knowledge stays in people's heads.

Not because it would be impossible to write down. Because it is born in interpretation, in conversations, in weak ties, in details too small to spontaneously become documents.

So the PM should not only supply sources to the Context. They should also be able to pour into it, progressively, their understanding of the field.

A well-fed context doesn't only know what the sources say.

It starts to understand what the team has learned.

To go further

The PM as Architect of Context Under the hood of my context engine: how an AI remembers a mission Information systems aren't going away. Their shape is changing. The second brain is a dead end for product management