We covered the first three floors of AI adoption last time. The next three are where AI stops being a tool your team uses and starts changing how the firm actually operates.
Floor Four: Projects
A project is a dedicated space where everything about a specific matter lives together. Our team uses these heavily for client work and trial prep.
The value is in the ongoing context. Every prompt, document, and conversation inside that project builds on what came before, so the AI is not starting fresh each session.
Our attorneys have found this especially useful for trial prep. They upload the case file and work through question sets, cross-examination angles, and different perspectives to consider. The output is more than what gets used, but that surplus is what makes it valuable. You end up better prepared than you would have been working manually.
The other benefit is that context stops living in one person’s head. It becomes team property, which matters as a firm grows.
Floor Five: Simple Automation
Floor five is where processes start running on a schedule without anyone triggering them. One of the people on our sales team, who does not consider herself technical, built an automation that runs every morning before the team arrives.
It pulls the transcripts from the previous day’s consultation sets, cross-references the notes in our system, and generates a briefing for each upcoming consult. When the team member sits down for that appointment, they already have a clear read on what the client cares about and where the conversation needs to go.
That work used to be an admin’s job. It suffered from all the usual problems — quality varied depending on who did it, things fell apart when someone was out and information got lost in the handoff. That version of the workflow is gone.
Floor Six: What Our Team Calls the Factory
Floor six is a full system where multiple AI agents work together on a single request, passing work back and forth. Our team calls it the factory.
The pattern mirrors how you would want a good team to operate. A frontline agent does initial research on the request. A more capable agent reviews that research critically and pushes back. Once they agree, a plan gets built and reviewed at a higher level again. Only after all of that does the plan come back for human approval.
You are essentially building the process a great team would follow if it had unlimited time and resources.
Deep research. Deliberate planning. Multiple layers of review before anything executes.
Most of us skip those steps in the real world because we do not have time. A factory makes them feasible because it runs at machine speed.
Once a plan is approved, the same process runs for each task inside it. A single prompt can produce work that would otherwise take a team days or weeks.
The Cost Reality
Yes, this uses real money. But the output of a functioning factory replaces the work of several people on any given project.
The important discipline is matching AI capability to the actual work. You would not put your most senior partner on a routine intake task. The same logic applies to which model handles which step. Get that match right and the economics work.
How to Actually Get There
Climb the floors in order. Prompting first. Then skills. Then tool connections. Then projects. Only after those feel comfortable does automation make sense. The factory comes last.
The signal you are ready for the next floor is usually frustration. When you find yourself repeating the same instructions over and over, you are ready for skills. When you find yourself doing the same task manually every day, you are ready for automation.
None of this happens in a week. What matters is starting and moving deliberately from one floor to the next. The firms that begin that climb now will look meaningfully different two years from now than the firms still deciding whether to start.
