We produce a lot of knowledge work. Client analyses, competitive studies, brand reports, and the notes and handoffs behind them. Over the last few months that record grew past three hundred session notes and a hundred handoffs. We had also built tools to make the work better: a shared format for our notes, a way to turn documents into typed concepts, and a nightly process that looks for connections we missed.
Then I noticed something. The nightly process skipped the folder where all our session notes live. The record of the work was not reaching the system built to learn from it.
The part most teams skip
Douglas Engelbart spent his career on one question: how a group gets better at thinking together. His answer separates three kinds of activity. The A activity is the work you exist to do, which for us is the client work. The B activity is any effort to improve how that work gets done, like a new tool or a better process. The C activity is the effort to improve the B activity itself, so you get better at getting better.
Most teams put almost everything into A, a little into B, and nothing into C. Engelbart's point is that C is where the gains compound.
Improving the work once is a one-time gain. Improving how you improve makes every later improvement faster and sharper.
We had been doing B for other people. The tools I mentioned were built for client engagements. We had not run them on ourselves. Engelbart had a word for running them on yourself: bootstrapping, applying your own improvements to your own work first. The test of a method is whether you use it on yourself, and we were failing it.
What we changed
Three small changes.
First, we folded our two overlapping dashboards into one, because neither covered everything. A single view now lists every artifact the work produces: session notes, handoffs, lessons, transcripts, and digests. It counts them straight from the files, so it never drifts from what is actually there.
Second, we treated each work record as a typed concept instead of a loose file. Every artifact already carried a small block of metadata at the top: what it is, which project, a one-line summary, and its tags. Once that block is consistent, one view can organize hundreds of files, and a program can read them.
Third, we added a view we call legibility gaps. It lists the records that carry no tags or no summary. Those are the notes the nightly process cannot use, because it has nothing to match them on. It is the same test that process runs. Putting it on the dashboard turned a vague worry into a specific list to fix.
Why this counts as C, not just a better tool
Fixing a dashboard sounds like B work. What makes it C is what it changes about the loop.
Engelbart called the shared, always-updated record of a group's work a Dynamic Knowledge Repository. The point of one is that a record of past work becomes an input to the next piece of work, not a filing cabinet you never open.
Our nightly process is a small version of that. A mechanical scan reads what changed and proposes connections from shared tags. A second step, with a person in it, picks the one connection worth making now. When the work record is legible, that loop runs on our own history, and each pass makes the next connection easier to find. That is the compounding Engelbart described, running in one team's notes.
The concept bundle, and why it is attached
Here is the part that involves you. Engelbart's 1962 work defined augmentation as raising a person's ability to handle a complex problem, and he was firm that it comes from two things changing together: the tools, and the habits and language around them. A new tool with unchanged habits does not augment much.
A blog post is a weak way to change either. You read it, you close the tab, the ideas fade. So this post comes with the ideas as files. Attached is a downloadable collection in Open Knowledge Format: fourteen short concepts, each a plain Markdown file, covering the Engelbart ideas above and the specific changes we made.
The collection includes a file called agent.md. Hand that file to your own assistant and it will place the concepts in your note system, link them to notes you already have, and explain each one as you read, without overwriting anything you own. If you would rather do it by hand, there is a readme. Either way, when you finish reading you have added the supporting concepts to your own knowledge work, in a form your agent can help you use.
Download the concept bundle. Give agent.md to your assistant, or open the readme and do it by hand.
We call that concept bundling, and this post is the working example of it. The bundle is the thing the post describes, shipped inside the post.
Where this is going
We will write up the full method later, with the schema and the tooling. This is the first pass, run on our own operation on 2026-07-19. The most useful thing it turned up is a gap: the nightly process still skips the folder where session notes live, so those notes reach the graph through an explicit field for now, not the scan. Fixing that is next.
That is the loop working as intended. The reason to point your methods at your own work is that your own work tells you where they fall short.
Sources
Doug Engelbart Institute, ABC Model for Continuous Improvement: https://www.dougengelbart.org/content/view/230/158/
Douglas C. Engelbart, Augmenting Human Intellect: A Conceptual Framework, 1962: https://www.dougengelbart.org/pubs/augment-3906.html
Douglas C. Engelbart, Toward Augmenting the Human Intellect and Boosting our Collective IQ, 1995: https://dougengelbart.org/pubs/books/augment-133150.pdf