How we work

A person approves what a client would see.

Agents draft, research, and classify. You approve outreach, spend, and anything compliance-sensitive. Those steps are hard to undo, so they are not left to an agent.

How that is built—git, CI, preview, and the gates—is on How we build.

Say what the work is
We write what a pilot will do, what it will not do, what it costs, and which risks it carries. A deck that cannot name those is not a plan.
Human in the loop
Agents speed up drafts, research, and classification. You approve outreach, spend, and compliance-sensitive steps.
Portable by design
Prefer open formats, standard interfaces, and an exit path. A system you cannot leave is a system you do not control.
Narrow access
Least privilege. Secrets stay out of code. Reviews are written down. Broad access fails in more places when it fails.
Stewardship of client trust
Federal contracting, personal data, and financial data get extra care. Decisions are written down. Gaps are named when they are known.
A scope that ends
A pilot has a fixed scope. Continuing is a new decision. An open-ended retainer hides whether the work is still worth doing.

The engagement model

Start with a bounded pilot. Continue only if the next scope is worth doing. The human gate stays in place either way.

1. Discover

A short conversation: what could be delegated, and what must stay with a person.

2. Pilot

A fixed scope: document lookup, an agent workflow, or a pipeline. The end state is written down before it starts.

3. Continue or stop

Expand only when the pilot shows the next scope is worth it. Stopping is a valid result.