Module 1 · 30 min
From Copilots to Agents
What 'agentic' means operationally, and how much of it you need
The hook · a fictional referral desk
The work is rarely the sentence. It is everything that happens after it.
A coordinator moves one referral between five systems, waits, re-checks and remembers what is outstanding. A copilot helps write. An agentic system may carry bounded coordination across those steps — while clinical judgement stays with a person.
The cost is not the writing. It is the traversal between systems and the waiting in between.
- 01
Referral inbox
New referral arrives
- 02
EHR check
Record, consent, prior imaging
- 03
Message practice
Chase a missing document
- 04
Queue update
Mark as waiting, move on
- 05
Scheduling
Check a matching slot
- 06
Clinician handoff
Anything ambiguous
Read left to right, then down. Each step is a system switch, and most of them are re-establishing context rather than making a decision. The final step is amber because it is a clinical judgement, not coordination.
What good could look like
- Coordination and chase work. Following up incomplete referrals, missing documentation, unanswered requests and stalled queues. High volume, largely administrative, and heavily rule-describable — the clearest early candidate.
- Multi-system handoffs. Work that requires reading one system, deciding, and writing to another. The value is not the decision but the traversal, which is exactly what a bounded agent can carry.
- Queue triage and preparation. Getting a case to the point where a human decision takes two minutes instead of fifteen: gathering, checking against a defined policy, and presenting what is missing.
- Exception surfacing. Reliably noticing what does not fit the policy and escalating it early, rather than discovering it at the point of appointment. Often the largest quality gain, and it needs less autonomy than people assume.
Boundary
Clinical judgement — urgency, diagnosis, treatment — stays a human responsibility. Delegation here is operational coordination only.
Potential value is a hypothesis to test in your own setting, not a guaranteed clinical or financial outcome.
The constructive question
Where could delegation reduce coordination burden while leaving clinical judgement and consequential choices with people?
Fictional educational framing: this module uses invented healthcare workflows to teach operational distinctions.
Sources & evidence · 4 sources
This module cites public or consensus guidance, vendor documentation.
Content reviewed: September 2026. Publication dates of the individual sources are shown in each citation.
OpenAI, A practical guide to building agents (2025).
A vendor engineering guide describing the loop, tool design and guardrail patterns in concrete terms. Useful as a description of how these systems are commonly built; it is not a neutral definition of what an agent is.
Open sourceOpenAI, New tools for building agents (11 March 2025).
Product announcement showing what commercial agent tooling now provides by default — tracing and evaluation among it. Read as evidence of where the industry is heading, not as an endorsement of any architecture.
Open sourceAnthropic, Building effective agents (19 December 2024).
Argues for the simplest architecture that works and distinguishes workflows from agents. The clearest published statement of the point this module makes about not reaching for agency by default; again, a vendor perspective.
Open sourceNIST AI 600-1, Artificial Intelligence Risk Management Framework: Generative AI Profile.
General risk-management background for generative systems. It is not agent-specific and does not address action-taking directly, so treat it as a framing document rather than a control set for agentic deployments.
Open source