Representative engagements, told as reframes — what a client came in wanting, the misread that would have produced the wrong build, and the reframe that reordered the problem. The most telling line in each is what Arc did not do. Anonymised.

Field Practice
Read asReframes, not wins
The load-bearing lineWhat Arc did not do
Cases04 · anonymised

AI-NATIVE

The production-trust boundary.

Clients already building AI, who hit the boundary where a demo must become something trustworthy in production.

FDE·An enterprise AI / agent-workflow company

From enterprise agent to business-process actor

Came in wanting
Better agent orchestration.
The misread
That an enterprise agent is a chat window or a CRM wrapper, and that more tools wired in means more capability.
Arc's reframe
The unit of value is a business-process actor — something that enters a real process, carries context, triggers actions, leaves evidence, and accepts responsibility. The work is process infrastructure, not integration.
What Arc did
A process-agent architecture: action boundaries, an evidence path, role-based decision rights, an audit trail, and a fallback/escalation design.
What Arc did not do
Did not bolt on more tool integrations, and did not treat orchestration as if it were capability.
Sediment
Process-architecture method and the action-boundary pattern.
Why it's Arc
The move from integration to process infrastructure is exactly the production-realness boundary.
FDE·An agent-builder / workflow-automation platform

From agent orchestration to agent governance

Came in wanting
To answer “which agent should make the decision?”
The misread
That orchestration is governance — that a permission toggle in the UI is authority.
Arc's reframe
Orchestration answers who acts; governance answers who is allowed to decide — under what evidence, what authority, and what accountability. The market solves the coordination problem; Arc solves the authority problem.
What Arc did
A governance model: policy-bound execution, a decision-rights matrix, approval gates, and evaluation/audit hooks.
What Arc did not do
Did not reduce governance to UI permissions, and did not equate “A calls B” with the right to decide.
Sediment
The governance-substrate direction behind Cawsal — a deterministic supervisor over non-deterministic action.
Why it's Arc
Responsibility-bearing structure, not capability, is the real blocker when AI enters an organisation.

DOMAIN-NATIVE

Forming the capability.

Clients holding real knowledge, workflows, or market position — but no path into AI, or no path up from a commodity.

Strategic reframe / Feasibility·An Australian solar-hardware retailer — not an AI business

From selling inverters to selling verified truth

Came in wanting
A way out of a commodity squeeze: caught between upstream manufacturers and downstream price-cutting resellers, with thinning margins on the boxes they sold.
The misread
Compete harder on price and volume, or “add an app” — either of which keeps them a reseller, only busier.
Arc's reframe
Their real asset was not the hardware; it was a position of independence and trust. The move: stop selling equipment and start selling verified truth about energy systems — independent third-party testing and certification of installed inverters (is output as claimed, is power being drawn off, is generation over-reported), and virtual-power-plant readiness assessment. The squeezed middleman becomes the trusted verifier both ends need — squeezed at both ends becomes earning from both ends.
What Arc did
The software-service and certification business model, the independent testing-and-certification product, the VPP-readiness service, and the evidence pipeline that makes a certification report defensible.
What Arc did not do
Did not build a generic e-commerce or CRM app, and did not advise a price war.
Sediment
The certification-as-moat pattern — Arc's verification thesis transplanted into a physical domain (the independence logic behind Lancet and the Arc Mark).
Why it's Arc
Arc sells trust, evidence, and independent judgement; here it taught a client to do the same, in solar. The think-with-you band creating real value before any build.

WHAT THE SET PROVES

Read end to end, four things hold.

  • Arc does deep-water AI, not integration.
  • Arc serves AI-native and domain-native alike.
  • Arc chooses the system shape from the problem — rarely “an agent” by default.
  • Arc has clear boundaries — it does not take whatever is asked.

START HERE

Bring the problem behind your demo.

Most engagements begin the same way — a short, low-risk diagnostic that finds what is real, what is failing, and what must be built before deployment.

Apply for a DiagnosticAll of Field Practice →