Approach
How I work
Five principles that hold across every engagement. None of them are novel. The discipline is in keeping to them when a shortcut is available.
01
Augment the people who do the work
The system gives a person the documents, history, and figures they need at the moment they need them. The judgment stays with the person. A new hire gets the benefit of institutional memory; a veteran gets perfect recall. Nobody is replaced by a chat window.
02
Integrate what exists
Companies have already invested in their email, accounting, CRM, and document systems, and those systems work. The job is to connect them through an intelligent layer, not to migrate everyone to something new. Rip-and-replace is a last resort, not a starting point.
03
Separate storage by data shape; unify through the agent
Unstructured text goes into retrieval systems built to find and cite it. Structured data goes into SQL. Relationships go into a registry. Rules go into a corpus with their authority recorded. One agent works across all of them, calling whichever store a question needs. The agent is the integration layer; there is no single giant database.
04
Observation before prescription
Map the current state from evidence first, validate it with the people who do the work, and only then decide what to automate. Then do one piece at a time, with a measure for each piece, so that a wrong turn is small and visible.
05
Discretion and data boundaries as design constraints
Where data may go is decided before the architecture, not after. Sensitive material stays on hardware I control; hosted models get only what the data classification allows; and client work is not discussed publicly. These are constraints the design has to satisfy, not policies added on top.
How an engagement works
There is no fixed program. This is the shape every engagement has taken so far.
01
A conversation
What is slow, what is uncertain, what is being copied by hand. Whether it is a fit gets said plainly, in that first hour.
02
An assessment from evidence
The current state mapped from what the systems recorded, validated by the people who do the work. No recommendations yet.
03
A first piece
The one thing that hurts most and can be measured, built and put in front of the people who use it.
04
A measure
Before and after, in the units the business already uses. If it did not move, that is the finding.
05
The next piece
Chosen from the assessment, informed by the measure. A monthly scorecard keeps the sponsor informed and the scope honest.
How long this takes depends on the company. I would rather say that than invent a number.