The next outcome is the real test.

Start with the work and its owner.

Define what must be completed, the business baseline, the people accountable and the consequences of failure. Choose the appropriate mix of reasoning, deterministic execution and human decisions. More agents is not the goal.

Build across four layers.

Foundation: ownership, decision rights, system responsibilities and the operating model. Execution: approved connections, reliable actions, exception handling and recovery. Intelligence: reasoning, evaluation, routing and risk-proportionate permissions. Interface: the usable experience accountable teams need to direct the work.

Inherit the approved foundation. Assess the new risk.

The next use case should reuse what is already sound. New data access, greater autonomy, different business consequences or new connections may require additional assessment. Reuse should reduce duplicated work, not bypass necessary review.

Measure the second outcome honestly.

Track the incremental engineering effort, time to value and the controls and connections reused. Measure operating cost per completed workflow, including inference, infrastructure, human review, exceptions and recovery.

Make ownership usable.

Source access alone does not create independence. The client needs the agreed rights, documentation, operational knowledge, controls and ability to evolve the capability. An ongoing Active service can remain the best choice without being the only possible choice.

Why the second outcome decides the economics.

A first deployment carries the cost of discovery: mapping the work, connecting systems, agreeing controls and earning the trust of the people who will use it. If the second deployment repeats all of that, the enterprise has bought two projects, not a foundation. The Active Harness exists so the second outcome starts further along than the first did, with approved connections, tested controls and an operating model the team already knows.

The market is moving the same way. Sequoia Capital, Services: The New Software describes the shift from selling tools to delivering the work itself. McKinsey’s State of AI finds that high performers redesign workflows rather than insert AI into existing ones, and back their deployments with leadership commitment and operational rigor. A reusable harness is how that discipline carries from one outcome to the next.

What the next use case should inherit.

Connections: the approved integrations, credentials and data contracts already in production. Controls: the evaluations, permissions, approval thresholds and audit trails that remain valid for the new work. Operating knowledge: runbooks, exception patterns and the decisions the accountable team has already learned to make. Interface patterns: the ways people direct, inspect and correct the work. Whatever the new use case introduces, such as new data, greater autonomy or different consequences, is assessed on its own terms.

Ask before the next deployment.

Which connections and controls will this use case reuse, and which are genuinely new? Who owns the outcome, and what can they change without waiting on a provider? What does a completed workflow cost once inference, infrastructure, human review, exceptions and recovery are counted? What will the use case after this one inherit? If those answers are unclear, the foundation is not yet doing its job.