Malaysian professional starting a focused AI-assisted work session at his morning desk

10 Intermediate AI Habits That Separate Serious Users From Casual Ones

Short answer: The AI habits that matter at the intermediate level are not about better prompts. They are about how you brief the tool, check its work, protect your data and measure whether it is actually saving you time. Ten of them, built into an ordinary workweek, are below.

Most Malaysian professionals are past the starting line.

The EY 2025 Work Reimagined Survey, reported by The Star in January 2026, found 93% of Malaysian employees already using generative AI at work, with daily usage nine percentage points above the global average.

Microsoft and LinkedIn’s 2024 Work Trend Index put the global figure at 75% of knowledge workers, nearly double what it had been six months earlier.

Using AI is no longer unusual.

The difference between people now is not whether they use it, but how they use it on an ordinary Tuesday.

Beginners ask what AI can do.

Intermediate users have answered that question and moved on to a better one: what is worth doing with it every week, and what discipline keeps it useful.

That is a question of habits, not features.

These ten are the ones worth building once the novelty has worn off.

What intermediate actually means

Beginner use is exploration: trying the tool on everything to find out what it can do.

That phase is normal, and it is how everyone starts.

Intermediate use is quieter and more specific.

You know which of your own tasks the tool helps with and which it quietly damages.

You have a standard way of working with it, refined by your own misses rather than borrowed from a thread of tips.

One framing note before the list: the studies and company reports cited below show these principles working at scale, in controlled experiments or large organisations.

The weekly routines suggested here are practice recommendations built on those principles, not exact prescriptions the studies themselves tested.

1. Brief AI the way you would brief a colleague

The most common early pattern is the two-line prompt with no context.

The intermediate habit is to treat every request like a work brief: who the output is for, what it is for, what tone fits, what to avoid and what a good result looks like.

A colleague handed a task with no background would produce something generic.

So does a model.

At the intermediate level this goes further than a one-off careful prompt.

  • Brief templates per work type. A stored brief for client emails, another for internal reports, another for analysis, each with the audience and constraints already filled in.
  • Negative constraints. What the output must not do: no invented figures, no hedging filler, no sign-off you would not use.
  • An exemplar. A short sample of what good looks like in your voice, pasted in, so the tool matches rather than guesses.
  • Material to reason from. The notes, figures or draft the answer should be built on, so it works from your facts rather than its guesses.

An illustrative example: ‘Write an email about the delayed shipment’ produces a generic apology. ‘Write an email to a long-standing Malaysian client about a two-week shipment delay, calm and accountable tone, offer a revised date and a named contact, under 120 words’ produces something much closer to sendable. The difference is the brief, not the tool.

Writing the brief also forces you to clarify what you actually want, which is often the real bottleneck.

Some requests turn out not to need AI at all once the brief is written clearly.

Infographic showing five parts of a complete AI work brief combining into one stored template

2. Draft with AI, own the final edit

Some of the clearest controlled evidence for AI at work concerns drafting.

In an MIT study published in Science in 2023, researchers Shakked Noy and Whitney Zhang gave 453 college-educated professionals, including marketers, grant writers, consultants and managers, two writing tasks specific to their occupations.

Those given access to ChatGPT for the second task finished 40% faster, and independent evaluators rated their output 18% higher in quality, as reported by MIT News in July 2023.

The study’s own caveat matters as much as the headline.

The tasks did not require precise factual accuracy or knowledge of a company’s goals or a customer’s preferences.

Real work does.

The intermediate habit is to use AI for the first draft and to keep ownership of the last one, editing in two separate passes.

The first pass is accuracy: figures, names, dates, claims.

The second is voice: whether it sounds like you and fits the reader.

Where people get into trouble is skipping both passes because the draft reads smoothly.

A useful self-test: if you would be embarrassed to defend a paragraph in front of the reader, it is not your draft yet.

3. Verify everything that matters, every time

Verification is the habit that separates competent use from confident use, and at the intermediate level it becomes selective rather than uniform.

The caution comes from a field experiment run by Harvard Business School researchers with Boston Consulting Group involving 758 consultants, some working with GPT-4 and a control group working without it, across realistic tasks, a study known as the jagged technological frontier.

On tasks inside AI’s capability frontier, consultants completed 12.2% more tasks, worked about 25% faster and produced results rated roughly 40% higher in quality than the control group, as VentureBeat reported when the working paper was released.

On a task the researchers had designed to sit outside that frontier, consultants using AI were less likely to produce correct solutions than those working without it.

The tool that makes you faster on one task can quietly make you wrong on another, and the output looks identical either way.

That is why fluent output is not proof of accuracy, a point this site has made before in the difference between AI confidence and AI competence.

Intermediate verification works by stakes, not by habit alone.

  • High stakes, hard check. Anything client-facing, financial, legal or public gets every figure and name checked against a primary source.
  • Medium stakes, spot check. Internal reports get a sample of claims verified, especially the ones that look most precise.
  • Low stakes, read-through. Drafts for your own use get a careful read for sense and tone.

Two Malaysian colleagues checking a printed report against a laptop with a red pen

4. Summarize first, read second

Intermediate users rarely read long documents from the top any more.

The habit is to have AI produce a short brief of a long report, proposal or thread first, decide whether it deserves a full read, then read the original for the parts that matter.

This is how some large firms now handle internal knowledge at scale.

Morgan Stanley built an AI assistant with OpenAI so its financial advisors can query the firm’s research library in plain language instead of hunting through documents, as CNBC reported in September 2023.

The intermediate version of this habit asks for more than a summary.

Ask what the document does not say: its assumptions, its missing evidence, the sections where the argument is thin.

A summary tells you what is there; a gap brief tells you what to distrust.

If a figure or claim from the summary will be used, it gets checked against the original.

Summaries compress, and compression is where nuance quietly dies.

5. Keep a prompt library and improve it

Early users rewrite the same instructions from memory every week.

Intermediate users keep the instructions that worked.

The habit is a small personal library: the meeting-notes brief that produces clean action items, the report structure that fits your manager, the tone settings that sound like your company.

Each time an output misses, the fix goes back into the stored version, not just into today’s session.

What makes it intermediate rather than a folder of saved text is maintenance.

Old entries get retired when a model update changes what works, and each entry carries a one-line note on why it works, so a colleague can adapt it rather than copy it blindly.

Over a few months this becomes a genuine asset.

It is also how the quality gap between two people using the same tool opens up: one starts from zero each time, the other starts from everything they have already learned.

6. Use AI as a critic before you commit

Most people use AI to produce.

Fewer use it to attack their own work, and that second use is often more valuable.

The habit: before an important proposal, plan or email goes out, ask the tool to argue against it.

Intermediate use gives the critic a standard to argue from.

A sceptical client who has been burned before.

An auditor who only cares about evidence.

A competitor who wants your budget.

Where is the reasoning weak, what would that reader ask, what has been assumed without support.

The tool does not know your situation, so some objections will miss.

But it can surface questions you were too close to see, and it usually takes only a few minutes before the decision is real.

Because it does not know your office politics or your client’s history, its objections stay at the level of pure argument, which is often where blind spots live.

Infographic of the critic loop: draft the work, hand AI a standard to argue from, it argues against it, you revise, then commit

7. Batch the recurring work

Steady gains come from work that repeats.

Weekly status reports, meeting minutes, first-pass replies, data clean-up, slide outlines: anything with the same shape every week is a candidate for a standing AI workflow.

The intermediate habit is to stop treating each recurrence as a new task.

Set the workflow up once, refine it with the prompt library habit, then run it as a routine with a quality gate.

A good candidate for batching has three traits.

  • Regular input. The raw material arrives on a predictable schedule.
  • Fixed output shape. The result looks the same every week, so a template holds it.
  • A review gate. A mistake would be caught in review before it causes harm, because a wrong report produced every week is worse than a slow one produced once.

Tasks missing any of the three stay manual.

8. Decide in advance what never goes into the tool

This habit is a boundary, set once and kept always.

Client lists, personal data, unpublished financials, confidential terms: each professional needs a clear personal line about what does not get pasted into which tool, drawn before the deadline pressure arrives.

The pressure always arrives.

The intermediate version matches the tool to the data.

An enterprise agreement alone does not make a tool safe for confidential data. Approval needs to come from your organisation for that specific data, tool and use, on top of company policy and any client contracts.

This site has covered the practical side in using AI at work safely in Malaysia, including the PDPA considerations that apply here.

It also helps to know your organisation’s line, because company policy and client contracts often restrict more than personal caution does.

When the two disagree, the stricter one wins.

Infographic of a three-tier data line for AI tools: public information is fine to use, internal material needs identifiers stripped, confidential and personal data is never pasted by default

9. Schedule deliberate practice, not just usage

There is a difference between using AI a lot and getting better at it.

Usage repeats what you already know.

Practice pushes into tasks you have not tried, with attention to what worked.

The companies taking this seriously treat it as culture, not software rollout.

Moderna launched its own internal ChatGPT instance, mChat, in 2023, then deployed ChatGPT Enterprise across its business functions, and the company has said publicly that embedding AI into everyday work requires real-time, on-the-job AI training rather than technology alone, as described in its April 2024 collaboration announcement with OpenAI.

The individual version is modest by comparison.

An hour a week spent on one new task type, reviewing honestly where the tool helped and where it misled, builds judgment that routine use alone tends not to.

The EY survey adds a Malaysian reason to take this seriously: only 12% of Malaysian employees report receiving sufficient AI training to fully benefit from it, according to The Star’s January 2026 report.

Nobody’s schedule is coming to save them.

The practice has to be self-scheduled, which is the kind of structural question this site examines in where AI adoption gets stuck, from the organisational side.

10. Measure whether it is actually saving you time

The final habit closes the loop.

Headline numbers about AI productivity are everywhere, and they deserve careful reading.

A widely cited GitHub Copilot experiment found developers completed a timed programming task 55.8% faster with the AI assistant.

The scope matters: that was 95 developers, one specific task, measured in a controlled experiment, as the researchers themselves describe it.

It is real evidence, for that task, not a promise about your workload.

Even company-reported wins get revised.

Klarna said its AI assistant was doing the work of about 700 customer service staff, as Reuters reported in 2025, but Bloomberg reported in May 2025 that the company was slowing its AI-driven job cuts and calling for real people in customer service again.

The useful response to all of this is not cynicism.

It is measurement.

Pick two or three recurring tasks, note how long they take with AI assistance, and compare honestly with before.

The EY survey found 81% of Malaysian employees report significant time savings from AI, yet two-thirds also say their workload has increased.

Saved time has a way of being silently reabsorbed.

The intermediate user notices where the hours actually went.

Where to start

Ten habits is a list, not a homework assignment.

The workable approach is to pick two: one that saves time immediately, usually the briefing habit or the batching habit, and one that protects you, usually verification or the data boundary.

Run those for a month until they are unremarkable.

Then add the next.

Habits compound quietly, which is why they tend to outlast enthusiasm.

It also helps to say them out loud.

A team that can name its shared AI habits can also spot when one has slipped, which is how standards survive a busy quarter.

Infographic of 10 intermediate AI habits arranged around a weekly brief, draft, verify and measure loop

Frequently asked questions

What are good AI habits for beginners?

Start with two: always give the tool context about audience and purpose, and always check factual claims before using them. Those two carry much of the value and prepare you for the intermediate habits above.

How long does it take to build an AI habit?

It varies from person to person. Give a new habit several weeks of deliberate repetition, attached to an existing routine such as Monday planning or Friday reporting, before judging whether it has stuck.

Do AI habits matter if my company already has AI tools?

Yes. Company tools set what is available; personal habits set what actually happens. The EY 2025 Work Reimagined Survey found 93% of Malaysian employees use generative AI, but only 12% feel they have had sufficient training to benefit fully, so individual practice is where the gap gets closed.

Which AI habit gives the fastest payoff?

Briefing the tool properly, the first habit on this list. It needs no new software or budget, only the discipline to write the brief before the request.

Can AI habits replace formal AI training?

No, and they work best together. Habits keep skills alive between sessions; structured training compresses the learning curve and corrects bad patterns early. Self-directed practice is a complement to training, not a substitute for it.

Building habits that outlast the hype

Tools will keep changing every quarter.

The professionals who benefit steadily are the ones whose habits survive the tool changes: brief well, verify always, protect data, measure honestly.

For teams that want to build these habits together rather than alone, Mike Cheong’s AI training for workplace productivity builds exactly these habits with Malaysian teams, on their own real work.

Details are on the AI training course page.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *