PCI: turning useful AI work into a repeatable capability

Process-Centric Intelligence, or PCI, is Senjen’s framework for designing work that can produce valuable results repeatedly, with clear authority and evidence. It starts with the outcome a person or organization needs, then asks how process and judgment should work together to deliver it.

This introduction summarizes PCI v1.1, the conceptual baseline dated July 13, 2026. PCI is a design framework and a set of hypotheses to test. It is not a claim that Amphra has already demonstrated better reliability, lower costs, or a complete governance system.

Start with the result someone needs

A team does not ultimately need an agent to produce more messages. It might need an accurate vendor record, a research brief that supports a decision, or a weekly report that exposes unresolved work.

Define that outcome before choosing the technology. Name the beneficiary, the conditions for success, the evidence that will be checked, and the tradeoffs that are unacceptable. A report completed on time but built on invented sources has not met the same outcome as a trustworthy report.

Look at the work through five responsibilities

Outcome

The valuable result and how its success will be judged.

Process

The known steps, constraints, state, handoffs, and feedback that help produce it.

Intelligence

The human or AI judgment needed where the right action cannot be fully specified in advance.

Authority

Who may observe, recommend, approve, execute, reverse, or change the work.

Evidence

The traceable facts that support evaluation: inputs, results, interventions, failures, time, and cost.

These are perspectives for assigning responsibility, not five services that every application must build. Evidence supports a judgment; collecting a log does not by itself prove that a customer benefited or that one method caused an improvement.

Use intelligence where it earns its place

For a competitor pricing brief, a known arithmetic conversion can be a formula. Extracting a clearly labeled date may need a simple rule. Interpreting an ambiguous product promise may require human or model judgment.

PCI calls this Minimum Necessary Intelligence: allocate judgment to the parts that need it, while preserving decisions the team already understands. It does not mean minimizing AI use regardless of quality. The right allocation depends on the outcome and the working conditions.

Keep the structure proportional to the work

A one-off, low-risk task may be well served by a prompt, a checklist, or a portable skill. Repeated work with several dependencies may need explicit inputs, versioned instructions, result checks, and a clear owner. A consequential action may also need enforced permissions and a recovery plan.

This is PCI’s Minimum Sufficient Structure principle. More structure is justified when it solves a real coordination, reliability, or authority problem. It is overhead when it does not.

Repeatability does not require identical wording or an identical path. In open-ended work, it may mean a stable standard of evidence and review, with an acceptable distribution of outcomes under declared conditions.

Make improvement something the organization can keep

When a reviewer corrects a recurring omission, capture the lesson in the relevant method and evaluation. Test the proposed revision, assign the authority to approve it, and keep its effects visible. PCI calls this Governed Evolution.

The framework builds on established process management, workflow, case management, and agent orchestration ideas. Its claims need comparison: does a revised method reduce failures or review effort on representative work without unacceptable costs? If not, simplify it or revise the hypothesis.

This is the connection to work that compounds. A useful result serves today’s need. A tested lesson can improve the next attempt. Amphra’s direction is to give that work a durable home; its current preview remains a canvas and bounded text-work experience. The changelog separates current progress from that longer-term direction.