One of the biggest changes AI has brought to my work is that I can do more. I am no longer as daunted by the mundane work involved in getting something started.

A research idea, teaching session or presentation comes with preparation: organising notes, building a structure, condensing information and revising the wording. These tasks are necessary, but the effort can become a barrier.

AI makes that preparation feel more manageable.

Across clinical practice, research, teaching and healthcare innovation, I use it to synthesise evidence, challenge ideas, review documents and prepare for discussions.

The benefit goes beyond completing an existing task faster. It changes how willing I am to begin. But being able to produce more makes deciding what deserves my attention more important.

Start with a specific task

Recently, I asked AI to tighten a healthcare innovation presentation to ten minutes. My instruction was simple: less is more.

Producing material was easy. Deciding what deserved the audience’s attention was harder.

The task involved taking material developed for an innovation discussion and reducing it to fit the time available. AI could help reorganise the argument, simplify explanations and suggest shorter headings.

However, shortening a deck requires more than removing words. Which points support the discussion? Which can be dropped? What should the audience understand or decide afterwards?

Those questions provide the standard against which to assess the output.

This is why defining the task matters. “Improve this presentation” leaves considerable room for interpretation. A specific audience, purpose and time limit give the work clearer boundaries.

For founders and operators, the equivalent might be an investor update or management paper. Before generating the document, identify the decision it needs to support.

A small error in this article

Preparing this article provided a concrete example of why checking matters.

My original instructions about the presentation were “Tighten for 10 mins presentation” and “Less is more”.

AI rewrote them as “Tighten for a 10-minute presentation. Less is more.” It improved the grammar but retained quotation marks, presenting edited wording as a direct quotation.

When I subsequently asked for verification and a word count, the review identified the discrepancy. The revised opening paraphrases the request.

The meaning had barely changed. Yet quotation marks make a specific claim: these were the words used.

Also Read: The sovereign threshold: Why airtight AI decisions are paralysing executive teams

When AI transforms source material, it can preserve the general meaning while changing a detail that matters. In research, a more consequential example would be turning “associated with” into “causes”: a small wording change with a substantial change in meaning.

Asking AI to check its work helped surface the quotation issue. However, another AI response is not independent evidence. The original wording made the discrepancy verifiable.

Use summaries to begin the review

Another recent task involved asking AI about the study design and appropriate critical appraisal tool for a national linked-data study of out-of-pocket healthcare costs in Australia.

The question was more specific than “summarise this paper”. I wanted to understand how to assess the evidence.

AI can explain terminology and compare appraisal approaches. Selecting an appropriate tool, however, depends on what the researchers actually did.

Was an intervention evaluated? How were participants selected? How were variables measured? Which biases could affect the findings?

The answers must come from the methods.

The practical lesson is to use the summary as an entry point, then return to the source for claims that determine the appraisal.

The same applies to market reports and customer research. Understanding the conclusion is insufficient if the method cannot support it.

Challenge polished but empty answers

Some output needs correction because it says too little.

I have asked AI to make writing less like a “motherhood statement”: something agreeable that offers limited practical direction.

Consider: “Healthcare organisations should embrace innovation to improve care.”

What should the reader do after reading that sentence? Which problem deserves attention? What would change? How would improvement be assessed?

The sentence is not necessarily false. It is insufficiently specific to guide action.

AI can make this harder to notice because fluent language gives an argument the appearance of completion.

A useful test is whether the paragraph still works after replacing the organisation’s name with another. If it does, it may need more context.

For a founder, “we use AI to improve efficiency” raises similar questions. Which task changes? Who benefits? What happens when the output is wrong?

Also Read: The AI told me to raise my prices by 40 percent but it had never been to a wet market

More capacity still requires judgement

I have not stopped reviewing or editing because AI is available. What has changed is my willingness to tackle work that includes substantial routine preparation.

I can ask it to organise material, propose an outline or shorten a draft, then examine and improve what it produces. The initial work feels less daunting.

I think of this as a cognitive multiplier. It makes it easier to develop ideas and explore alternatives.

However, generating a draft is only part of the workflow. Reviewing, correcting and adapting it also count.

A field experiment involving 758 Boston Consulting Group consultants found that participants using GPT-4 completed tasks within its capabilities 25.1 per cent more quickly on average. On a task outside those capabilities, AI users were 19 percentage points less likely to produce the correct solution.

This was a consulting experiment, not a clinical evaluation. It does not establish how much time AI saves in my work.

It illustrates why benefits should be assessed task by task. A quick draft that needs extensive repair may offer less value than its initial speed suggests.

The relevant questions are whether the finished work is useful and whether it deserved to be done.

Keep ownership of the decision

AI can help develop a coherent research proposal. Whether the study should proceed depends on its importance, feasibility, participant burden and potential to change practice.

Operational decisions require similar judgement.

Consider an appointment no-show prediction system. Identifying someone as likely to miss an appointment does not determine the appropriate response.

A reminder, telephone call and help with rescheduling involve different resources and assumptions. Missed appointments might reflect transport difficulties, caregiving responsibilities or an unsuitable appointment time.

Someone must decide which response addresses the problem, who will deliver it and how its effects will be assessed.

Also Read: The 70/30 rule: Why ASEAN banks still need humans as AI takes over

Clinical decisions involve additional considerations, including the patient’s circumstances, preferences and the consequences of acting or waiting.

AI can organise those considerations. The decision requires someone who understands the context and accepts responsibility for the action.

For founders, this matters whenever AI informs hiring, customer support or resource allocation. A recommendation needs an accountable owner.

Practice makes permanent

I believe professionals should use AI appropriately and often. Familiarity develops through working with it, examining the results and adjusting the approach.

But practice does not automatically make perfect. It can make our habits permanent—including the wrong ones.

Repeatedly accepting plausible answers can establish a habit of uncritical reliance. Repeatedly checking important claims, comparing outputs with sources and correcting errors can reinforce a more disciplined approach.

Frequency alone is insufficient. The quality of the feedback matters.

Start with a bounded task where the output is possible to assess. Define what success looks like. Identify which details require verification. Review where the tool helped and where it introduced additional work.

In healthcare, appropriate use also includes respecting confidentiality and approved arrangements for handling information.

The aim is to become better at judging the output, including recognising when additional expertise is needed.

AI has made me less reluctant to tackle the mundane work. It helps me reach a workable draft, a sharper question or a clearer structure more quickly.

Before using the result, I still need to answer three questions: Is it supported? Does it fit this situation? Can I take responsibility for acting on it?

AI helps me get to those questions faster. I still have to answer them.

—

Editor’s note: e27 aims to foster thought leadership by publishing views from the community. You can also share your perspective by submitting an article, video, podcast, or infographic.

The views expressed in this article are those of the author and do not necessarily reflect the official policy or position of e27.

Join us on WhatsApp, Instagram, Facebook, X, and LinkedIn to stay connected.

The post AI can help me think faster, but I still have to decide how I actually use AI in healthcare appeared first on e27.