Two years in, most organisations are using AI to do the old work faster. The ones pulling ahead run systems that remember, decisions that trace, and operations that hold when a regulator asks. We get organisations onto the right side of that gap.
Echolocation is how we work: observe first, then orient, decide, act. We sell tighter loops, not answers. Click the field to ping it.
the model layer we work across · routing, evaluation and fallback Marks are the property of their owners, shown to describe our own stack. No endorsement, sponsorship or partnership implied.
Start where it hurts. Each line of work strengthens the others.
We turn scattered experiments into a target architecture, a sequenced plan, and a governance model that survives audit. Including the hardest call: what not to build.
AdvisoryEveryone is an agent operator now, trained or not. We build the judgement that makes that safe: what to delegate, what to verify, what to refuse. Taught on your systems, on real work.
Training and enablementModels drift, data changes, builders move on. We run what has been built to defined service levels, or take the whole function and hand you the outcome.
Managed servicesDurable advantage lives in the layer around the model: memory, provenance, the personal-data boundary, audit. ORCA is that layer, and we run our own business on it.
What ORCA doesPositions we publish and defend. Read them before you hire anyone, including us.
Boyd was right about air combat and he is right about organisations: tempo beats capability. The firm that orients on evidence in seconds beats the one that searches its inbox for an hour.
Read the essayThe frontier labs ship models. Everyone else ships the scaffolding that decides what an agent remembers, what it may touch, and what it must never do. That layer is where your advantage lives.
Read the essayA policy document is a wish. A gate that physically cannot be bypassed is governance. If your controls live in a handbook, you do not have controls.
Read the essayA vibe-driven engagement looks confident and produces deliverables that sound right. Six months later nobody can explain how the recommendation was reached. Vibes do not survive scrutiny.
Read the essayThe EU AI Act is not a compliance cost. Governance built into the architecture is the cheapest durable advantage available in 2026: earlier, cheaper, and harder to copy than any feature you could ship.
Read the essayMost software depreciates. A governed knowledge system compounds: every solved problem, every correction, every decision makes it more valuable than the day before.
Read the essayMost AI engagements stop at a working prototype and a slide pack. The difficult part starts afterwards: the model drifts, the data changes, an auditor asks who approved a decision, and the people who built it have moved on.
We stay for the part that comes after. Somebody remains accountable for the system in production: your team, because we trained them, or ours, because you asked us to run it.
You own the code, the models, the evaluation sets, the runbooks and the documentation. Throughout, and afterwards. Exit is designed in week one, not negotiated at the end.
Due diligence is pre-answered. Residency, subprocessors, data handling and continuity are published, not gated behind a call.
Data handlingThe useful first conversation is usually about the thing that worries you, not about our capabilities. If it turns out we are the wrong firm for it, we will say so.
One reply from a person who has read what you sent.