1 AI-first workflows in Biometrics
AI adoption adds AI to an existing task. An AI-first workflow redesigns a bounded process so an agent can perform structured work under explicit rules, evidence requirements, and human decision rights.
Clinical trial development is already organized as a chain of connected workflows. The study protocol informs trial operations and the statistical analysis plan (SAP). Collected data move through the Study Data Tabulation Model (SDTM) and Analysis Data Model (ADaM) into tables, listings, and figures (TLFs). Those results support the clinical study report and regulatory submission. Every step has inputs, rules, handoffs, reviews, and accountable owners.
AI can do more than accelerate one isolated task. An agent can inspect artifacts, use tools, execute several steps, observe results, and return evidence-backed findings or recommendations. Using that capability reliably requires the surrounding workflow to be designed for human-agent collaboration.
In this book, an AI-first workflow is a bounded, repeatable process designed for an AI agent to perform structured work with tools, while explicit artifacts, rules, events, benchmarks, and human decision rights govern its execution and improvement.
1.1 From AI adoption to AI-first
AI adoption begins when an individual uses AI for an isolated task inside an existing workflow. A person decides when to invoke AI, assembles the context, reviews the response, and moves the work forward. This can improve personal productivity, but the workflow itself remains unchanged. The KEYNOTE-189 demonstration in Chapter 3 uses this pattern to show what a single prompt can accomplish and why a repeatable workflow needs more than a prompt.
An AI-first workflow begins with the work rather than the model. The team defines the objective, boundary, evidence, success measures, and decision rights, then assigns each step to the component best suited to perform it. AI does not need to perform every task or make final decisions.
| Characteristic | AI adoption | AI-first workflow |
|---|---|---|
| Unit of work | An isolated task | A bounded, repeatable process |
| Invocation | A person starts a prompt | A defined event or rule starts the work |
| Context | Assembled for the interaction | Supplied through controlled, versioned artifacts |
| Output | A response for a person to interpret | A structured artifact with evidence, coverage, and uncertainty |
| Success | Immediate usefulness | Performance against a defined benchmark |
| Continuity | The next interaction starts largely anew | State and human decisions inform the next run |
1.2 How an AI-first workflow works
An AI-first workflow does not begin and end with a prompt. It connects AI to a repeatable process:
Event -> Task -> Tools and agent -> Evidence -> Human decision -> Recorded state
An event starts the workflow, such as a change to an SAP, dataset, program, or output. The task definition states the objective, source artifacts, allowed tools, expected result, conditions for stopping, and responsible reviewer.
The workflow gives the agent current, versioned artifacts instead of relying on whatever happens to be available in a chat. The agent returns a structured finding that states what was checked, the supporting evidence, any uncertainty, and what it could not cover. A human reviews the finding, makes the decision, and records that decision for the next run.
The design separates three kinds of work:
| Work | Best-suited component |
|---|---|
| Apply an exact rule, calculation, or comparison | Deterministic software |
| Interpret written requirements, trace context, or plan variable steps | AI agent |
| Exercise domain judgment, approve policy, or accept accountability | Human role |
The simplest capable component should perform each step. An AI agent should not replace a deterministic test, and access to a tool should never be confused with approval authority. Chapter 2 develops the design principles that make this division of work inspectable and trustworthy.
1.3 Human accountability is not optional
In an AI-first workflow, the human role shifts from manually executing every step toward defining the work, supervising agents and tools, reviewing exceptions, challenging evidence, and improving the workflow. The agent can retrieve, compare, draft, run approved tools, and monitor routine conditions. The accountable human retains responsibility for the decision and its consequences.
This makes three human capabilities especially valuable: systems thinking, end-to-end knowledge, and exception handling.
An April 2026 FDA warning letter to Purolea Cosmetics Lab described a drug manufacturer that used AI agents to create specifications, procedures, and production records. FDA found that the firm had not adequately reviewed the generated documents for accuracy and Current Good Manufacturing Practice (CGMP) compliance. The firm also reportedly relied on the agent’s failure to mention a requirement as a reason for not knowing that the requirement applied.
The letter concerns pharmaceutical manufacturing and CGMP, not clinical study reporting. Its workflow lesson still transfers: AI-generated content remains a draft until an authorized person verifies and approves it. The workflow must check authoritative requirements, disclose what it did not cover, and preserve the evidence behind the decision. An agent cannot be the source of regulatory truth or its own approver.
1.4 Where AI-first workflows apply in Biometrics
AI-first workflows can support each Biometrics pillar without changing who is accountable for the work.
| Pillar | Useful first pass | Human decision |
|---|---|---|
| Data management | Compare edit checks, draft queries, trace unresolved issues | Approve queries and database readiness |
| Statistical science | Find inconsistencies across the protocol and SAP, assemble evidence, draft questions | Approve design, methods, and interpretation |
| Analysis and reporting | Check traceability, execute code, compare TLFs, draft findings | Validate results and approve deliverables |
Many valuable workflows cross these boundaries. Reviewing a TLF may require the SAP, ADaM metadata, executed code, rendered output, and prior decisions. Domain expertise and end-to-end knowledge therefore become more important, not less.
1.5 Start with one bounded workflow
An organization does not need to begin with autonomous, end-to-end execution. A useful first workflow is small enough to observe closely but valuable enough to improve real work. It has a clear beginning and end, accessible source artifacts, a reviewable output, and a named person responsible for the decision.
A workflow is bounded when the team can state:
- what event starts it;
- which artifacts and tools it may use;
- which actions the agent may take;
- what output it must return;
- when it must stop or escalate; and
- who reviews the result and makes the decision.
The initial implementation should include a task contract, a deterministic foundation wherever exact rules can be encoded, and a reviewable result. It can begin in advisory mode and expand one boundary at a time. The applied examples develop these implementations through the shared AI-first SDLC rather than introducing a separate development sequence for each workflow.
For example, a first workflow for reviewing a changed table could be defined as follows:
| Element | Definition |
|---|---|
| Trigger | A pull request changes a table program or rendered output |
| Inputs | Current SAP, output specification, analysis metadata, source code, and rendered table |
| Agent task | Compare the change with the requirements and identify inconsistencies or missing evidence |
| Required output | Structured findings with source references, coverage, uncertainty, and recommended follow-up |
| Stop conditions | A required artifact is missing, versions conflict, or the finding requires statistical judgment |
| Human decision | The accountable reviewer accepts, rejects, or investigates each finding |
This workflow is deliberately narrow, but it establishes reusable capabilities: artifact retrieval, traceable review, structured findings, escalation, and a recorded human decision. Those capabilities become the foundation for broader AI-first workflows and for the lifecycle described in Chapter 4.