GovTech Just Let Go of 93 People, and AI Didn't Take Their Jobs

A Singapore government job is about as close as this country gets to the proverbial iron rice bowl. The pay may not beat the private sector, but the bowl doesn't crack. That deal has held for so long that most people would struggle to name a single public-sector retrenchment.

Last week, GovTech retrenched 93 people in one announcement.

Around 300 will go over the next two years.

So I understand why the group-chat reaction was some version of "even the government can't protect jobs from AI anymore". It's the obvious reading. It's also the wrong one, and the actual story is far more useful to anyone running a business in Singapore right now.

What was actually announced

On 15 July, GovTech announced it would cut 7 to 9 per cent of its workforce over two years, in three phases. The first phase put 305 officers under review, and a review isn't an exit; here's how it split.

102 kept their existing roles.

110 are being retrained into new roles through what GovTech calls full-salary apprenticeships.

93 were retrenched, 36 of them immediately.

The official framing is that GovTech is moving from a "one-off project-delivery model" to a "continuous product-ownership model", which is the kind of phrase that makes eyes slide off the page. The translation: instead of paying vendors to build government systems and managing the contracts, GovTech wants to build and run more of those systems itself.

Look at which roles are going and which are being hired, because that's where the story lives. The cuts fall mainly on project and vendor management. Meanwhile the agency says it's still actively hiring software engineers, product managers and designers, and the new operating model leans hard into engineering, data and cybersecurity capability.

GovTech also stated, flatly, that the move was not driven by artificial intelligence and wasn't a cost-cutting exercise. Chairman Chng Kai Fong put it better than the press release did: "GovTech is changing shape, not shrinking... It will be a different GovTech, and not every current role maps onto it."

The part that's easy to misread

I'll concede the sceptical reading has a point. "Not driven by AI" is doing some heavy lifting in that statement, because AI is a big part of why building software in-house has become cheap and fast enough for this reorganisation to make sense at all. The honest version: AI made building things cheap enough that GovTech now needs a different set of jobs, and it decided to say so out loud.

And notice the direction of the change, because it runs opposite to the popular fear. If AI were simply replacing people, you'd expect the builders to go and the coordinators to stay.

GovTech did the reverse.

The coordination layer (managing vendors, running one-off projects) is what got cut, and the agency is paying full salaries to retrain people toward owning systems: the code, the data, the security posture, the 3am incident when something breaks.

We run a small version of this shift at Magnified. Our monthly client reports used to be an afternoon of copy-pasting per client from GA4, Google Ads and Search Console. That part is now a pipeline an AI agent runs. What surprised me was the proportions.

Automating the report was maybe a third of the work.

The other two thirds went into rules nobody puts in a product demo: making sure every client's analytics is queried with that client's own credentials so keys never cross accounts, and making sure nothing reaches a client until a human has read it. Boring governance work. It's also the only reason I can let the agent run without watching it.

That experience is why I think GovTech's restructuring is the most instructive AI story Singapore has produced this year. Anyone can use AI to generate things (your secondary-school nephew included). The scarce skills are deciding what AI is allowed to touch, who checks its output, where the data lives, and who answers for it when it's wrong. Which is exactly the mix GovTech is hiring for, cybersecurity included, while it cuts the roles that existed to supervise outsourced work.

What this looks like at SME scale

Most SME owners I speak to have settled into one of two camps. The freeze camp: "AI is hype, we'll wait." The fantasy camp: "AI will let me halve my headcount."

GovTech is quietly demonstrating a third option, and I'd argue it's the only one that ages well: keep roughly the same people, change what they do, and treat AI capability and data security as things the organisation itself understands, even when it hires outside help to build them.

A ten-person company doesn't need a Smart Nation budget for this. "Changing shape" might mean the ops manager learns to run and check an AI workflow instead of retyping invoices, and a named person (not "the vendor") knows which customer data goes into which tools. That last one matters more than most owners realise, because pasting a customer database into a random AI tool is a data breach with extra steps.

The same story, from the employee's seat

Most of the commentary treats this as a management story. It's just as much a story for anyone who draws a salary, because GovTech won't be the last organisation to change shape. Companies restructure around new tools, and when they do, jobs and responsibilities get redrawn along with the org chart.

This has happened before. The pattern is kinder than the headlines suggest.

When word processors arrived, the typists who learned them early stopped being typists and became the people who ran the office. When spreadsheets arrived, the clerks who picked up Lotus 1-2-3 and then Excel stopped totting up columns by hand and started building the models their bosses depended on.

The tools ate the task. The people who picked up the tools got promoted with them.

Economist James Bessen documented a version of this with ATMs: the machines took over cash handling, banks opened more branches because each one got cheaper to run, and tellers shifted toward relationship and advisory work the machines couldn't do.

Look back at GovTech's phase one with that history in mind. 110 officers are living the same trade right now, paid full salaries to swap their old roles for the skills the new operating model runs on. The retraining is the deal on offer, and the officers who take it seriously will come out the other side more valuable than they went in.

If the organisation around you starts changing shape, the losing move is guarding the old task. The winning move is the one the early Excel adopters made: get to the new tool first, learn to run it, and become the person who can check its output when it's wrong.

The typing pool didn't survive. The typists who moved first did better than survive.

To the 93 officers, none of this is comfort, and I won't dress it up; some of them, by their chairman's own admission, performed well in roles the organisation simply needs less of. That's the uncomfortable deal the AI era puts on the table: being good at a role no longer guarantees the role.

GovTech looked at its own org chart and asked a blunt question: which of these roles only exist because we pay outsiders to build things we don't understand ourselves? Then it acted on the answer, publicly, with 93 retrenchments' worth of conviction.

Full disclosure before the closing thought: I run an agency. I'm the vendor in this story, so read the rest of this knowing where I sit.

GovTech's answer was to bring the understanding in-house, and it paid 305 reviews and 93 retrenchments to do it. Most organisations won't go that far, and don't need to. The workable middle ground I see is an arrangement where the work can stay outside while the understanding still comes home: the vendor builds, and the client's team learns enough to question what's been built, check it, and eventually redirect it.

That's a negotiation, and both sides have to want it. Vendors have to be willing to explain themselves out of their own mystique. Clients have to put someone in the room who actually wants to learn.

So here's the question I'd leave both sides with. The next time work goes outside the building, what's the plan for the understanding to come back in?