The questions buyers actually ask.

Keeping Copilot is the right call — for personal productivity it's excellent, and rolling it out was a good decision. What it structurally can't do is own a workflow: it doesn't know your estimating logic, can't carry a quote from request to approval, and doesn't answer from a knowledge system you govern. We build that layer. Most clients run both, side by side: Copilot for individuals, AI employees for the workflows.

No. That's the point. We build systems your existing IT team can run, we train them during the engagement, and if they're not ready at handover, managed operations covers the gap until they are.

Fixed fees, quoted before we start. Workflow Discovery, the 90-Day Foundation (scoped to one measurable workflow result), and managed operations. Go/no-go gates mean you can stop at weeks 2, 4, or 10 and keep everything produced.

A supervised AI employee in one channel typically inside six weeks. A full foundation (knowledge systems, harness, first agent in production) in ninety days. We publish the week-by-week anatomy on the How We Engage page.

The model is the most replaceable part of the system. Your knowledge graph, your skills library, and your harness outlive any model release; when a better model ships, we swap it in and your system gets smarter overnight. That's why we engineer the harness instead of betting on a vendor.

No. Never ours, never a provider's, under the configurations we deploy. The full detail, including where data lives and what we access, is on the Security page.

Then we say so, in writing, at a gate, and you keep the workflow map, the architecture sketch, and everything built to that point. We'd rather lose an engagement than publish a case study we have to hedge.

Neither, on principle. The cost-and-privacy math decides: high-volume, sensitive workloads tend to earn local hardware; spiky, capability-hungry tasks tend to earn frontier APIs. Most clients end up hybrid, with routing rules instead of religion.

The founder, with specialist engineers from a small trusted bench. Two Foundation engagements at a time, maximum — a quality ceiling we set on purpose, so nobody gets handed to a junior team after the kickoff call. If both slots are full, we tell you the next start date and run Workflow Discovery in the meantime, so you arrive at your Foundation with the map already drawn.

A workflow owner for about two hours a week, a decision-maker at three milestone gates, and scoped read access to the systems the workflow touches. No war rooms, no steering committees.

Yes, by design: open formats, documented skills, playbooks, and training in weeks 11 and 12. Handover is a milestone, not an afterthought.

Everything lives in your repositories and your tenant from day one: code, pipelines, prompts, skills, vector stores, graphs. Open models where they clear the quality bar. Firing us is a permissions change. We put it in the contract.

You're never stranded, by design. Every artifact — code, pipelines, skills, docs, and the weekly written status history — lives in your repositories from day one, current and complete. The bench engineers on your engagement work under the same four commitments. We won't pretend a bench engineer replaces his judgment overnight — but the system that de-risks your key people is the same system that de-risks us. That symmetry is deliberate.

Yes — send it early. The diligence pack collects what most reviews ask for: our controls mapped to standard questionnaire fields, subprocessors with retention terms, and every exit path documented, each with an honest status. Whatever your questionnaire asks that the pack doesn't yet answer, you get back in writing.