The new health-tech race is not summarisation — it is anticipation

For the last wave of healthcare AI, summarisation was the feature everyone could understand immediately.

A note became shorter. A chart became easier to scan. A long interaction became a neat paragraph. A complex source became a digest. A patient conversation became a draft.

That was useful.

It still is.

But in 2026, summarisation no longer feels like enough.

Not because it has stopped being valuable, but because it is increasingly becoming visible rather than decisive.

It is the feature clinicians see. It is not necessarily the feature that creates the deepest operational value.

That is the shift.

The new health-tech race is moving beyond passive recap and toward something more consequential:

anticipation.

In other words, the next useful healthcare AI will not simply tell you what happened.

It will increasingly help tell you what now needs to happen.

That might mean:

surfacing the next administrative task
flagging missing follow-up steps
preparing the referral package
drafting the after-visit summary
suggesting coding opportunities
routing a message to the right queue
identifying discharge actions
preparing handoff content
surfacing risk signals that require response

That is a very different product layer.

Summarisation looks backwards.

Anticipation looks forwards.

And in a workflow-driven health system, forward-looking support is often where the larger economic value sits.

That is why this is becoming the real battleground.

The short answer

Summarisation is becoming easier to imitate and harder to defend as the main differentiator in healthcare AI.

The more valuable layer is increasingly anticipation: the ability of an AI system to identify, prepare, and tee up the next administrative or clinical task in workflow.

That is where the current market hints are pointing:

Dragon Copilot is positioning itself around not just documentation, but also coding suggestions, referral letters, after-visit summaries, and workflow assistance
Amazon Connect Health is explicitly aimed at administrative workflows such as patient verification, scheduling, histories, documentation, and coding
Epic’s broader AI strategy is increasingly framed around handoffs, patient responses, provider insight, and workflow assistance rather than pure recap

That does not mean summarisation disappears.

It means summarisation is increasingly becoming the visible feature, while anticipation becomes the economic feature.

The new moat may not be what the AI can summarise.

It may be what the AI can foresee and tee up safely.

Why summarisation no longer feels like enough

Healthcare AI needed a first wedge.

Summarisation was a perfect one.

It was easy to demonstrate, easy to understand, and easy to connect to a real pain point. Clinicians are overwhelmed by documentation, notes, inboxes, messages, and information overload. A system that can turn a messy interaction into a cleaner output is immediately appealing.

That is why the category took off.

But first wedges rarely remain the whole market.

Over time, two things happen.

First, more competitors become able to offer reasonably strong summarisation. The novelty fades.

Second, buyers begin to ask a harder question:

What is the summarisation actually enabling?

That is the key.

A summary by itself may save some time.

But the larger operational value usually appears only when the summary becomes the bridge to something else:

an action
a routing step
a follow-up task
a documentation improvement
a communication output
a coding event
a discharge or referral process

That is why summarisation no longer feels sufficient as the main product story.

It is still useful.

But on its own, it risks becoming closer to a feature than a durable strategic moat.

What anticipation means in healthcare

Anticipation in healthcare should not be understood as science fiction.

It is not about omniscient AI predicting every clinical twist.

It is about something more practical:

recognising what needs to happen next and preparing it before a human has to manually build it from scratch.

That next step may be administrative, operational, or clinical-adjacent.

1. Next tasks

After a note, visit, message, or handoff, there is usually more work to do.

A more anticipatory system can identify the next likely task and tee it up:

patient needs follow-up in 2 weeks
referral letter needs preparing
medication reconciliation is incomplete
discharge instructions are still missing
coding detail is insufficient
a message should be routed to a clinician rather than admin

2. Missing orders or missing workflow elements

Anticipation can also mean noticing that something expected in the workflow is absent.

That does not necessarily mean auto-ordering in the strongest sense. It may mean identifying what still needs clinician attention, what part of the packet is incomplete, or what operational step has not yet been addressed.

3. Needed follow-up

Many healthcare workflows break not because the first decision is wrong, but because the follow-up work is fragmented or delayed.

A system that can surface follow-up needs proactively may create more value than a system that merely restates the encounter.

4. Coding prompts and documentation-linked operations

Documentation is not only clinical memory. It is also operational substrate.

A more anticipatory system can move beyond summarising the note to identifying what the note implies for coding, reimbursement, or task routing.

5. Referral preparation and discharge support

These are powerful anticipation layers because they convert information into movement.

A clinician often does not only need to know what happened. They need help preparing what comes next.

That is where anticipation becomes economically meaningful.

Summarisation is the visible feature; anticipation is the economic feature

This is probably the sharpest way to understand the shift.

Summarisation is easy to see.

It produces a tangible artefact:

a summary
a draft note
a concise recap
a shortened message

That makes it excellent for demos and early adoption.

But the economic value of healthcare workflows often sits further downstream.

The expensive problems are not only that information is messy.

They are that after information is produced, organisations still need to:

schedule correctly
code accurately
follow up reliably
route tasks efficiently
avoid denials
close loops
reduce delays
keep patients moving through care safely

That is why anticipation may matter more economically.

A good summary is helpful.

A system that uses the content of the encounter to tee up the next operational step can change throughput, staffing burden, follow-up quality, revenue capture, and care continuity.

That is a much stronger value story.

Why the market is moving there

The shift toward anticipation is not arbitrary.

It is being driven by product economics, workflow logic, and the maturing of the AI market.

1. Summarisation is easier to imitate

As large models improve and more vendors gain access to similar capabilities, summarisation quality alone becomes a weaker differentiator.

Not irrelevant — but less defensible.

If many systems can produce a decent note or a decent summary, buyers start looking for a deeper source of value.

2. Anticipation ties more tightly to workflow outcomes

Healthcare buyers increasingly care about what the product changes, not only what it generates.

Anticipatory systems tie more directly to outcomes such as:

reduced administrative burden
fewer missed follow-ups
faster discharge processing
better coding capture
more efficient inbox triage
smoother referral workflows

That makes the commercial case much stronger.

3. Embedded AI creates the conditions for next-step support

Once AI sits inside workflow surfaces rather than separate browser tabs, it becomes much easier for the system to help with what comes next.

This is one of the reasons workflow placement and anticipation are tightly linked.

If the AI is only a destination search box, it mostly answers questions.

If it sits inside the documentation layer, inbox, handoff, or referral process, it can begin to support next-step generation.

4. Health systems care about movement, not just interpretation

Healthcare organisations do not only need information interpreted.

They need care, admin, and communication workflows to move.

That is why the industry is increasingly interested in systems that can reduce friction after the summary is produced, not only in systems that are good at producing the summary.

Current hints of that shift

The market signals are now clear enough to support this thesis in a grounded way.

1. Dragon Copilot’s workflow outputs

Dragon Copilot is a strong example because its positioning now extends well beyond note-taking. Microsoft’s own materials explicitly frame it around coding suggestions, referral letters, after-visit summaries, and evidence-backed support alongside documentation.

That matters because it reflects a move from recap toward task generation and workflow participation.

The product is not only summarising the encounter.

It is helping prepare what follows the encounter.

That is anticipation in a practical form.

2. AWS’s new health administration platform

Amazon Connect Health is another strong proof point.

Its public framing is not centred on a better note summary. It is centred on reducing administrative burden across workflows such as patient verification, appointment scheduling, medical histories, documentation, and coding.

This is exactly the pattern the market should be expected to reward.

The value is not “the AI understood the interaction.”

The value is “the AI helped move the workflow forward.”

3. Epic’s broader workflow AI

Epic’s AI direction also supports the same argument. Its public materials increasingly emphasise patient responses, handoff summaries, provider insights, charting support, and the embedding of generative AI directly into the EHR workflow.

Again, the important thing is not merely that summarisation exists.

It is that intelligence is being placed where it can connect directly to the next operational step.

The move from passive recap to active next-step generation

This is the deeper product shift.

Passive recap is useful because it reduces reading and writing burden.

Active next-step generation is useful because it reduces coordination burden.

That is a much more strategic layer.

A healthcare workflow often fails not because nobody knows what happened, but because the next step is:

delayed
forgotten
routed badly
incompletely prepared
left for someone else to reconstruct manually

That is why active next-step generation may be the more powerful product shift.

The system no longer merely compresses the past.

It begins to prepare the future.

Where anticipation lands first

The earliest landing zones are unsurprising once you think in workflow terms.

1. Admin

Administrative workflows are the most obvious first destination because they are repetitive, expensive, and often rules-driven.

Anticipatory systems can help identify what needs to happen next without stepping directly into high-risk autonomous clinical decision-making.

2. Coding

Coding is a strong early landing area because it depends on what the documentation implies operationally.

A more anticipatory system can surface opportunities, missing detail, or likely next steps in the coding and reimbursement workflow rather than simply summarising the encounter text.

3. Inbox

The inbox is increasingly a major site of low-grade but costly work.

A system that can anticipate routing, prepare drafts, identify urgency, and reduce manual triage burden may be much more valuable than one that only summarises incoming messages.

4. Referrals

Referral preparation is another natural fit. A more anticipatory system can recognise that a referral is likely, prepare the relevant information, and reduce the manual work between the clinical decision and the downstream operational action.

5. Discharge and handoff

Discharge and handoff are rich areas for anticipation because they are transition-heavy and prone to omission. A system that can prepare summaries plus next actions, continuity information, and follow-up elements can create real workflow value.

6. Scheduling and follow-up orchestration

Scheduling is not just an admin problem. It is a continuity-of-care problem.

A more anticipatory system can connect the clinical encounter to the right next follow-up step with less manual reconstruction.

Why anticipation is harder

This is not an easy frontier.

If it were, everyone would already be doing it well.

1. Liability

A system that proposes next actions carries more responsibility than a system that merely summarises the past.

That increases the stakes.

2. False positives

Anticipatory systems can create clutter if they tee up actions that are unnecessary, irrelevant, or poorly timed.

In healthcare, too many incorrect prompts are not merely annoying. They can erode trust quickly.

3. Workflow interruption

An anticipatory system must be helpful without becoming intrusive. If it interrupts the clinician too often, or at the wrong moment, it can create more burden rather than less.

4. Alert fatigue

This is a major risk.

Healthcare already suffers from alert overload. Any anticipatory system that behaves like another noisy alert layer will struggle.

5. The need for safe tee-up rather than unsafe autonomy

This is a key design principle.

The best near-term systems may not be those that act autonomously in strong ways. They may be the ones that tee up the next step safely for human confirmation, correction, or routing.

That is a much more realistic and governable model.

What founders should learn

This trend matters because it changes how product value should be designed.

1. Move beyond recap-only thinking

If your product story stops at “we summarise”, it may become increasingly vulnerable unless summarisation is only the visible layer of something deeper.

2. Design around the next action

The more useful question is not only “What did the AI generate?”

It is “What workflow moved because of it?”

3. Build for safe anticipation, not theatrical autonomy

Healthcare does not need more magical claims. It needs systems that can anticipate the next useful step in bounded, auditable ways.

4. Tie value to operational outcomes

Products that can show effects on routing, follow-up, discharge flow, coding quality, inbox burden, or admin throughput will increasingly have a stronger economic story than products that only demonstrate elegant summaries.

Bottom line

The new health-tech race is not summarisation.

It is anticipation.

Summarisation will remain useful, but it is increasingly becoming a commoditising layer.

The higher-value product shift is from passive recap to active next-step generation.

That means the next useful healthcare AI will not just tell you what happened.

It will increasingly help tell you what now needs to happen:

the next task
the next route
the next follow-up
the next document
the next handoff
the next operational step

That is why the next AI moat may not be what a system can summarise.

It may be what it can foresee and tee up safely.

Frequently asked questions

What does “anticipation” mean in healthcare AI?

It means AI systems helping identify, prepare, or surface the next useful administrative or clinical-adjacent task in workflow, rather than only summarising past interactions.

Why is summarisation becoming less defensible as a moat?

Because summarisation is increasingly easier to imitate as model quality improves and becomes more widely accessible. Buyers then look for value tied more directly to workflow outcomes and operational movement.

What are examples of anticipation in healthcare?

Examples include coding prompts, follow-up reminders, referral preparation, discharge task support, inbox routing, handoff preparation, and surfacing missing workflow steps after an encounter.

Why might anticipation be more valuable than summarisation?

Because healthcare value often sits in what happens next: the routing, coding, communication, follow-up, and coordination tasks that determine whether care and operations move efficiently.

Why is anticipation harder to build safely?

Because it can create liability, false positives, workflow interruption, and alert fatigue if designed poorly. The most useful systems will usually tee up next steps safely rather than acting autonomously without oversight.

What should founders do differently if this shift is real?

They should design beyond recap and ask how their product changes the next operational step, ties to measurable workflow outcomes, and supports safe anticipation rather than only elegant summarisation.