How we work
Measure twice. Cut once.
Every engagement measures the company twice before it changes anything: once to find where the structure holds it back, and once to decide what agentic AI can now do at that point. Then we cut, and we measure again.
A business is an economic engine.
A business turns effort, skill, and money into value. Some of that value is financial. Some is personal: the work people are proud to do. Some is social: the customers it serves and the communities it supports, including the small societies of small businesses.
A company that hands its work to agents and watches them run has built an application. We build the other kind of company: one where agents carry the volume, so its people can carry the quality.
Value moves through three links: supply, development, and distribution.
We borrow the term "value chain" and use it our own way. Every company has three links, and each one has its own throughput.
- Supply. What the company brings in: the people it hires, the materials and knowledge it gathers, and the customers who arrive at its door.
- Development. What the company makes of them: the product, the service, the program, and the work that shapes each one.
- Distribution. How the result reaches the customer: the marketing, the sale, the delivery, and the support that follows it.
In each link a few steps decide the quality the customer receives, and those steps carry the company's differences from its competitors. The rest of the steps move work from one hand to the next. We redesign each link so the work flows through it without waiting, and we build agents to carry the steps that only move the work.
Put people where quality is decided.
The optimum place for a person is the place where a person makes the biggest difference to the quality of the result. That is the whole design principle. The agents take the volume; the people take the judgment, the relationships, and the craft that customers pay for.
What that person finds in front of them is the record: what ran before and how it did, the numbers as they stand, what the customer said last, what competitors are doing, and the rule that applies. The agents put it there before the step opens. The person spends the effort on deciding, and none of it on gathering.
Seven lenses find the ceiling.
A stalled company feels the problem in one place and usually holds its cause in another. We read the company through seven lenses before we accept any diagnosis. Every engagement surfaces some of them. Rarely all. Never just one.
- Offer architecture. What the company sells, to whom, against what alternatives, and how its offers fit together.
- Unit economics. What one unit of work truly costs and earns, across the full loaded cost.
- Organizational design. Who owns which result, where decisions sit, and where judgment lives.
- Workflow architecture. How the core work actually runs, and where agentic AI opens designs the old work could not reach.
- Strategic positioning. How the company really competes, and on what difference its buyers actually value.
- Growth architecture. How new revenue arrives, and whether it compounds or runs in place.
- Ownership and governance. The time horizon, risk appetite, and authority that decide which changes are possible.
Two measurements before every cut.
- The architectural measurement. Where in the structure does the ceiling sit? The lenses answer that, and the answer usually lies a step or two away from the problem the owner first described.
- The agentic measurement. At that point, does the redesign change the work itself, or does it only finish a task faster? We build the first kind.
An engagement that passes only the first measurement delivers insight that never reaches the work. One that passes only the second delivers impressive agents at the wrong point in the structure. We cut when both measurements agree.
Every step goes to one of four places.
- Delete it. A step that exists out of habit, or no longer serves the result, goes first.
- Run it as plain code. A fixed rule with no judgment runs as ordinary software: predictable, cheap, and easy to check.
- Hand it to an agent. A step that needs context and judgment goes to an agent, which asks a person only for exceptions and approvals.
- Keep it with a person. Money, liability, relationships, and the steps that decide quality stay with a named person.
Plain code wherever possible. An agent only where judgment is needed. A person wherever quality is decided.
Three rules hold in every engagement.
- A person approves every step that touches money, customers, or contracts. The agent drafts; a person sends.
- Routine steps run as plain code. A model is called only where judgment is needed.
- Nothing ships until it has been reviewed.
Five steps, the same every time.
- Map the real process. We interview the people who do the work and record the last real run. The documented process is rarely the real one.
- Sort every step. Delete it, run it as code, hand it to an agent, or keep it with a person.
- Build inside your tools. We work in your email, your accounting software, your CRM, and your Microsoft 365 or Google Workspace. Your staff learn no new screens, and nothing moves to new software.
- Gate what is risky, then test the gate. Every send to a customer, every payment, every contract, and every deletion waits for a named person's yes. We test each gate by trying to get around it.
- Measure and hand over. We measure hours, turnaround time, and errors against the starting numbers, then hand over a recorded walkthrough and a written guide.
Your team owns what we build.
The walkthrough, the guide, and the measured numbers stay with you. The people who run the work after handover understand how it runs and why each gate sits where it does.