100,000 free AI seats. Here is how to make them count

 From Merdeka Day, 100,000 young Malaysians can earn three months of free AI software use, under a new government initiative. Designed well, that is not a giveaway but seed capital for an industry.

Somewhere in Kuantan this September, a 22-year-old will finish a short online module and unlock three months of an AI assistant that retails at around RM100 a month. What happens next is a design choice.

Built one way, she summarises her lecture notes and three months later, her subscription quietly lapses. Built another way, she ships a stock-tracking tool for her father’s hardware shop, and by December three other shops pay to use it.

The second is within reach.

On Tuesday, Prime Minister Anwar Ibrahim announced that some 100,000 Malaysians aged 18 to 30 will get free three-month AI subscriptions after completing training on the government’s Rakyat Digital platform from Aug 31. The digital ministry, under Gobind Singh Deo, has set modules on generative AI, agentic AI, cloud, cybersecurity and AI safety.

Those who finish, Anwar said, “will then be able to subscribe freely (for three months) to popular AI apps”. Right instinct, right moment.

Start with the arithmetic. No provider has been named, but ChatGPT Plus retails here at RM99.90 a month and its Go tier at RM24, so three months for 100,000 people lands between RM7 million and RM30 million, less if bought in bulk. Against Malaysia Digital 2030 and its 500,000 high-value digital jobs, that is under RM60 a job.

The money is not the hard part. Three design choices are.

First, give them agents, not a chat window

A subscription buys a conversation. What changes output is an agentic harness: Anthropic’s Claude Code and Cowork, OpenAI’s Codex, or the open-source OpenClaw. You give these a goal and controlled access to your files or browser, and they do the work.

A chat window helps a clerk write an email faster. An agent pointed at 400 supplier invoices extracts them, reconciles them against the ledger and pushes what it cannot match to a human, overnight, every night.

So, teach the harness, not the prompt. The durable skills are scoping a task, giving an agent the least access it needs, and checking its work before trusting it. That middle one carries the risk. A chatbot that gets something wrong wastes a minute. An agent loose in your drive can send the wrong file to the wrong client at three in the morning, and never stop to wonder if it should.

Second, point them at real problems

Every ministry and state should publish ten operational problems it wants solved, each with an officer authorised to answer questions, the data that exists, and what a fix is worth. Problems like these:

• Small firms keying tax invoices into a government portal one at a time;

• Road complaints closed as “selesai” with no photograph proving the repair;

• Business licences that renew differently in every council;

• Subsidy forms demanding documents another ministry already holds; and

• Contract awards scattered across hundreds of agency sites with no common feed.

Judge on a working demo, not a slide deck. Pay winners for a pilot small enough to sign without a tender, and make the department say publicly whether it kept the thing. Publish it under an open licence, as GovTech Malaysia does with its Hansard tools, so the next council with the same problem reuses it instead of paying to build it again.

Third, copy what is already working abroad

Two countries got the sequencing right.

Estonia’s AI Leap gave premium AI tools to roughly 4,800 upper-secondary teachers in August 2025, and only opened a learning app to 20,000 students five months later. Do that with polytechnic and vocational lecturers and this becomes curriculum, not a campaign.

Singapore’s TechSkills Accelerator goes further. On its company-led track, firms rather than portals deliver the training, the trainee signs a contract before it starts, and the employer draws salary support for a year. Nearly 18,000 Singaporeans had been placed and trained across the scheme by 2024. Do that here and the pipeline ends in a payslip.

Then measure what they build, not what they finish

Completion rates are the least interesting number here. Capture a baseline at sign-up: what each entrant does now, and how long it takes. At six months and twelve, publish:

• What they shipped, with a live link or listing;

• Revenue booked – spot-audited not self-declared;

• Hours a week saved in the job they already had;

• New companies registered, and what they billed; and

• How many still pay for the tools themselves.

Set those targets before launch, on a public dashboard, or they get invented later.

Malaysia has spent fifty years assembling other people’s technology for a modest share of the value. This is the first wave where the tools are cheap enough for us to be on the other side.

The figure of RM30 million and 90 days is a small bet. Placed well, it is the one we will talk about in 2036. - FMT

The writer can be contacted at kathirgugan@protonmail.com.

The views expressed are those of the writer and do not necessarily reflect those of MMKtT.