Arc is a private research-and-engineering house — it holds deep technical options and does not go to market; the ventures do. What Arc takes are selected engagements: one motion that brings already-held capability to a partner's problem, where frontier problems meet real systems.

Field Practice is not Arc's main line — it is the house's own research-and-engineering capability, turned outward. Arc builds its own frontier; engagements are that capability lent to a partner's problem.

01

WHAT IT IS

One motion, at the depth your problem needs.

Field Practice is the public face of Arc's engagements. It is not a services catalogue and not a price list — it is one spectrum, from thinking with you to building with you to handing you the asset. What changes across it is only how deeply Arc embeds, and who ends up holding what.

Arc brings capability it has already built and holds to your problem, and stays accountable for a system that works. Where your problem sits on the spectrum decides how deeply Arc embeds and who ends up holding what — but the engine is the same throughout: research and engineering, deployed where the problem lives.

02

THE SPECTRUM

Think with you · build with you · hand you the asset.

Forward-Deployed EngineeringbuildFeasibilitythinkLicensinghand-offResearch Collaborationpartnerone motion · four depths

The modes below are cut by the shape of the relationship — who holds sovereignty, what Arc keeps, what you keep — not by activity name. Its most committed form is forward-deployed engineering, where Arc embeds and builds while holding its own substrate and method.

03

THE MODES

Which way in is yours.

Think with you

Diagnostic

Find what is real, what is failing, and what must be built before deployment.

Relation
Arc diagnoses an existing system — the low-risk door.
You keep
A failure map, risk register, eval-gap report, and a go / no-go / rebuild call.
Read more →
Think with you

Feasibility Study

Judgement on whether to build — feasibility, risk, architecture, go/no-go.

Relation
Arc brings judgement; you decide. Arc does not build.
You keep
A recommendation — feasibility report, risk map, architecture options, go/no-go.
Read more →
Build with you

Forward-Deployed Engineering

Arc embeds and builds the system with you — making AI production-real.

Relation
Arc builds, holding its substrate and method. The flagship.
You keep
A working, trustworthy system in your environment — not a deck.
Read more →
Supervise the build

Supervised Delivery

Arc owns the architecture and quality gates; your team or a vendor builds to them.

Relation
Arc holds architectural authority and reviews the work — it does not become the build vendor.
You keep
An architecture, technical spec, quality gates, and acceptance criteria — and a system built to Arc's standard.
Read more →
Partner with you

Research Collaboration

Peer exploration of a frontier problem — not only a paper.

Relation
Peer — one-to-one, equal. Not provider→client.
You keep
Shared research output — notes, benchmarks, prototypes, primitives.
Read more →
Hand you the asset

Licensing & Partnership

Licence or co-develop Arc's protected substrate, patents, and methods.

Relation
Arc transfers an asset under terms; it retains the background IP.
You keep
A bounded licence or a co-development relationship.
Read more →
04

WHAT ARC BRINGS

Arc never enters empty-handed.

Arc brings a body of capability it already holds — substrate (retrieval and document structure that stays traceable), evaluation (defensible measurement), and governance (boundaries drawn by construction). Together they are what let Arc make a system trustworthy in production, and they are why every engagement begins from capability already in hand. Where the path is clear an integrator will do; Arc takes the work whose crux is whether an AI system can be trusted.

See how Arc has reframed real problems — the Casebook →

START HERE

Start with a Diagnostic.

Arc accepts a small number of selected engagements where the problem genuinely requires its research, engineering, intelligence, or governance. Most begin the same way — a short, low-risk diagnostic that finds what is real, what is failing, and what must be built before deployment.