Vibe Coding, Clinical Tool Building & Agentic Workflows

The Diagnostic Prompt IMI Applied to System Design + AI-Assisted Tool Development — Concept, Method & First Build

Opens with a 15–20 minute formal revisit of The Diagnostic Prompt IMI, now applied to system design — its most sophisticated application: before specifying a clinical tool, let AI interrogate the problem itself. Live demonstration: a real clinical tool idea is brought to AI — the AI's questions surface assumptions about the user, edge cases in clinical workflow, failure modes under stress, governance requirements, and scope creep the clinician had not articulated. The resulting specification is demonstrably more precise and safer than what was initially described. Remainder of session: AI-assisted development defined as a clinical skill; why clinical domain expertise is the primary skill required, not programming; a five-question clinical tool specification method; live demonstration building a validated clinical scoring tool from a plain-language description; participants begin building their first tool during the session with facilitator guidance.

First Clinical Tool Build — Applied Self-Directed Learning with The Diagnostic Prompt IMI

Participant begins by applying The Diagnostic Prompt IMI to their own clinical tool idea — before writing any specification or opening any development environment. The AI's questions are recorded verbatim. Using the specification that emerges from that interrogation, participant builds their first independently developed clinical tool using a browser-based AI development environment, addressing a genuine clinical or administrative problem in their practice. Participant shares the tool with two clinical colleagues and records their feedback verbatim.

Clinical Tool Specification Framework & Formal Usability Testing

Structured online teaching session covering: a four-component structured specification framework — (1) clinical situation: who is using this tool, in what context, under what constraints; (2) purpose: what decision or action should it support; (3) worked examples: three to five realistic cases defining correct output; (4) constraints: what the tool must never do, what populations it does not apply to, what information it requires before it can operate; a five-step iterative refinement loop for improving a tool through directed prompting; plain-language debugging — when the tool produces wrong output, how to describe the problem to AI in terms that generate a fix; a validated ten-item usability assessment instrument — administration protocol, scoring method, and interpretation benchmarks for clinical tool readiness.

Formal Usability Testing — Reviewing Tool Performance Against a Validated Standard

Participant applies the four-component specification framework to implement at least one substantive improvement to their Week 1 tool. They then formally administer the validated usability scale to a minimum of two clinical colleagues (not involved in the tool's development) and calculate usability scores. Structured reflection on what the usability data reveals about the tool's performance and readiness — reviewing clinical tool performance against a validated standard.

Agentic Clinical Workflows — Scheduled Agents with AI Integration

Structured online teaching session with live build covering: the distinction between a triggered automation and a scheduled autonomous agent; connecting Make.com to an AI reasoning engine via direct API integration — without coding; scheduling an agent to run at defined intervals without manual initiation; three high-value clinical agent patterns with live demonstrations — (1) operational data anomaly detection: agent reviews department data and flags outliers for human review; (2) literature monitoring: agent checks for new publications matching defined clinical criteria; (3) audit data processing: agent collects, formats, and summarises recurring audit data; mandatory human approval checkpoint design — no agent in a clinical setting should send external outputs without a human review step; a pre-deployment safety checklist for clinical agents.

Reflection on Clinical Agent Deployment — Measuring Agentic Workflow Performance

Participant builds and deploys a clinical agent using a no-code automation platform integrated with an AI reasoning engine. They run the agent on real or de-identified data a minimum of three times and document what it produced, where it was accurate, where it required correction, and what oversight mechanisms were required. Prospective evaluation of a newly deployed agentic clinical workflow, including reflection on what conditions would need to be met before it could be trusted to run autonomously.

Deployment, Operational Intelligence & Clinical Governance

Structured online capstone session covering: deploying a built clinical tool to a publicly accessible URL; building a live operational intelligence dashboard from department data — live demonstration transforming a raw audit data file into an interactive dashboard using Claude (personal device) and the equivalent workflow in Microsoft Copilot in a hospital Microsoft 365 environment; structured peer review of each participant's deployed tool (one specific strength, one specific improvement, one governance question); a Clinical AI Tool Governance Dossier — the eight required sections for hospital and university governance committee submission, drafted in-session with facilitator guidance.

Capstone Reflection — Governance Dossier & Full Lifecycle Measurement

Participant completes their Clinical AI Tool Governance Dossier, collects post-deployment feedback from additional clinical colleagues, and completes their capstone reflection covering the full lifecycle: problem identification (via The Diagnostic Prompt IMI), specification, build, usability testing, deployment, agent design, and governance. Structured evaluation of the complete lifecycle of clinical AI tool development and deployment.