Notes From the Practice
The Practitioner's Ledger
The practitioners who do well with digital staff all share one habit, and it isn't better prompting. They keep a ledger — a private record of decisions, standards, voice, and observations. Most don't realise the ledger is the differentiator until they watch someone without one struggle.
Most attempts to make an AI agent useful start with the agent. The tool is installed. Prompts are configured. Output appears. When the output disappoints, the response is to try a different model, or a longer prompt, or a more elaborate system. The fix is sought in the same direction as the problem.
The fix is almost always elsewhere. It is behind the agent, in what the agent knows about you and your work. And the thing that holds that knowledge — the practical artefact — is something old: a ledger.
What a practitioner's ledger contains
Not numbers. Not the accountant's kind, though those exist too. A practitioner's ledger holds:
Decisions — what you chose, what you considered, what you'd revisit. Pricing changes, scope choices, tool migrations, hiring calls. Each decision worth two sentences, dated, tagged by the area of work it touches.
Standards — how you do things. Email tone for clients. Document structure. Calendar policies. The five-sentence rule. The "I don't take calls before 10" rule. Anything you've quietly agreed with yourself but never written down.
Voice samples — three to five examples of writing that sound unmistakably like you. Stripped of identifying detail. Stored where an agent can read them.
Observations — what you noticed about a client's preferences, a vendor's quirks, a workflow's failure mode. The institutional knowledge that lives in your head and disappears when you forget.
Scopes and exceptions — what an agent is and isn't authorised to do. What requires escalation. What categories of work can run unsupervised. What categories must always pass through you.
None of this is exotic. All of it already exists, somewhere, in your head, in your inbox, in scattered notes. The ledger is the act of collecting it in a form an agent can read.
The discipline of keeping it
The hardest part isn't the format. The format barely matters — Markdown, Notion, Obsidian, a folder of text files all work. The hardest part is the habit.
Most operators are not in the habit of writing things down for anyone but themselves. They keep decisions in their head, communicate standards by example, hold institutional knowledge as personality. This works at a certain scale. It stops working when the operation grows past one person.
When you delegate to digital staff, you cross that threshold without leaving the desk. The agent is, in effect, a new colleague who needs to be told how things work. Either you tell them — by writing — or they make it up. Most operations that fail with agents fail because the practitioner declined to write.
The discipline that fixes this is small and unromantic. Five minutes at the end of the day, capturing one decision. A weekly review, half an hour, refining standards as you notice them. A monthly read-through to catch what's drifted.
It is not exciting. It compounds.
Months three, six, twelve
The first month of ledger-keeping is the worst. The notes feel obvious. The discipline feels like overhead. Nothing seems to come back.
By month three, something shifts. You ask an agent to draft something and the draft sounds right. You're not sure why until you remember you wrote a voice sample two months ago that the agent is now drawing on. The work happens in the background. The link between input and output is no longer a direct line.
By month six, the ledger is doing real work for you every day. You add notes without thinking. You revise standards because something didn't fit. The agent's output is increasingly indistinguishable from your own first drafts. You have stopped wondering whether agents work.
By month twelve, the ledger is the operation. Were you to walk away for six weeks, the staff would continue running not because the model is good, but because what the model knows about you and your work has accumulated to a point where continuity is achievable. The ledger is the asset; the agent is the interface.
How this differs from context engineering
Context engineering is the setup: the technical configuration of how an agent reads your notes, what tools it can use, where the vault lives, how queries route. Most setups can be done in a weekend.
Ledger-keeping is the daily practice. The setup is meaningless without it. A vault with no notes is just a folder.
This is the part most people miss. They configure their tools elaborately, point an agent at an empty Obsidian, and conclude after two weeks that the technology isn't there. The technology is fine. The vault was empty.
What we do
In each engagement, the first phase of work is composing the foundational ledger for the principal. We interview, we observe, we read past correspondence, we surface the standards held in their head, the decisions they've made and never written, the voice they write in. We assemble all of this into the form an agent can read.
Then we hand it back. The principal maintains it from there, refining as the operation evolves. Once a month, we review the ledger together and update what has drifted. The agents themselves are the easy part. The ledger is the work.
Where to start, if you're starting
pen a folder. Call it whatever you like. Inside, make five files:
decisions.md— write one decision you made this week, in two sentences, with what you considered.standards.md— write one thing you've quietly agreed with yourself but never put in words.voice.md— paste three emails you've recently sent that feel like you.people.md— write one note about the most important person you work with: their preferences, their communication style, what matters to them.observations.md— write one thing you noticed this week about a workflow or client.
That is your ledger.
Add to it tomorrow. Add to it the day after. In ninety days, you will have a ledger that runs an operation. That is what a practitioner's ledger is. It is not a technology. It is a habit.
