He joined a mid-sized consulting firm in Singapore straight out of one of the country’s top universities, first-class honours, the quiet confidence of someone who had always been told he was among the best. His first two years were unglamorous — slide decks at midnight, industry benchmarking reports that took a week to produce, competitive landscapes that his managers would glance at for thirty seconds before a client call. He did not complain. That was the deal. You paid your dues at the bottom, and the ladder rewarded patience.

By year four, he was a consultant. By year six, a senior consultant. Clients started asking for him by name. He could see the next rung clearly — manager, then principal, then perhaps, one day, partner.

Then, in early 2026, a new analyst joined the team. Fresh out of university, unremarkable in every way Marcus could initially see. One Thursday afternoon, she handed him a competitive analysis. It was the kind of report Marcus would have needed two full days to produce in his first year. She had done it in three hours — with an AI tool and a monthly subscription that costs less than a weekday lunch at Raffles Place.

It was not perfect. But it was 80 per cent of the way there. And 80 per cent by Thursday afternoon is worth considerably more than 100 per cent on Monday morning.

Marcus looked at that document for a long time.

What he was really looking at was this: she had not just done the work faster. She had done his work. The work that justified his salary. The work that seven years of experience had taught him to do well. And she had done it on her second month on the job, at a fraction of his cost, using a tool that anyone can access.

He is still looking.

The Story Nobody Is Telling

There is a story that has been circulating on Chinese social media recently. It is funny, in the way that things are funny when they are also slightly terrifying.

A senior employee, drowning in work, finally asked his boss for help. “We need to hire someone,” he said. “I genuinely can’t keep up.”

The boss agreed. “You can sit in on the interviews.”

Feeling magnanimous, the senior employee pulled the most promising candidate aside before the session. “Don’t worry,” he said. “The boss doesn’t really understand the technical side. Just sound confident. The more impressive you seem, the better.”

The candidate — fresh out of university, fluent in every major AI tool, and interviewing at half the senior employee’s salary — nodded, walked in, and when the boss asked whether he could handle the entire workload alone, said: “Absolutely. With the AI systems I use, I could probably streamline the whole operation while I’m at it.”

The boss looked at the candidate. Then at the senior employee. Then back.

The next morning, the senior employee was called in. “Since one person can handle it,” the boss said, “there’s no need for two.” A pause. “And he’s cheaper.”

The senior employee had coached his own replacement into the room.

This is not a story about a new hire. It is a story about what happens when the new hire comes equipped with AI — capable of producing a senior employee’s output, at a junior employee’s wage, with a confidence that years of experience used to be the only way to earn.

AI Doesn’t Start at the Bottom

This is not the version of the AI jobs story that most people are telling.

The version most people are telling goes like this: AI will first come for the lowest rungs — the data entry clerks, the call centre agents, the back-office processors. Knowledge workers are safer. Their work is too complex, too contextual, too human. Give it time. But not yet.

That story is not entirely wrong. It is just aimed at the wrong floor.

What AI does well — remarkably well — is the kind of work that can be broken into steps, thought through carefully, and written up in structured form. Research. Analysis. Synthesis. Drafting. Spotting patterns across large amounts of information. These are not low-skill tasks. They are the core of what professional services firms have charged premium rates for, for decades. They are what junior lawyers, analysts, consultants, and accountants spend their early years learning to do.

And it turns out that this kind of work, however demanding it feels, can be learned. By a machine that has ingested more text and data than any human could get through in ten lifetimes. For a monthly subscription fee.

The junior associate reviewing contracts does not lose her position because her work is simple. She loses it because her work is knowable — and once something is knowable, it can be learned. By anything. By anyone willing to spend a weekend figuring out the right approach.

Which is exactly what this year’s graduates have done.

They are not waiting patiently at the bottom of the pyramid. They have looked at where Marcus is standing, worked out what they can already produce with the tools on their laptops, and concluded there is no reason to start below it — and no reason for their employers to pay a senior salary when a junior with AI delivers comparable output. They are not looking for a place to start. They are looking at where you are standing, and asking whether you still need to be there.

So the middle of the pyramid is being squeezed from two sides at once. The AI tools pressing down, making accumulated knowledge less exclusive. The AI-equipped graduates pushing up from below, willing to do the same work for significantly less. The senior employee in the story did not see it coming because he was watching the door, not the tool the person he let in was carrying.

The Reassurance That Doesn’t Quite Reassure

At this point, someone usually steps in with comfort. Let me give that version its due.

Teo Ser Luck, outgoing president of the Institute of Singapore Chartered Accountants, offered one in a recent Straits Times interview. AI changes the game, he said, but it does not end it. He described how an auditor typically works with about 60 per cent of available data. AI can push that to 80 per cent. The final 20 per cent — the judgment call, the professional responsibility, the signature on the document — stays human. “Someone must be responsible,” he said, “and it can’t be a machine.”

He is right. And he is right that the work becomes more interesting — junior accountants who once spent years on ledger reconciliation will now engage in analysis and strategy much earlier in their careers.

But sit with that 80 per cent number for a moment.

If AI now handles 80 per cent of the work that junior staff were hired to do, the question is not whether the remaining work is more fulfilling. The question is: how many junior staff does the firm actually need? If one AI-equipped graduate can do what three junior accountants once did, the firm does not need three junior accountants. It needs one — the sharpest, the most capable with the tools — and a much smaller payroll.

There is a second problem inside the optimistic framing. The career ladder worked because the early years taught you something. Marcus became a senior consultant worth listening to because he spent years doing the groundwork that nobody glamorous wanted to touch. That groundwork built the judgment he now sells. Take it away, and you do not just lose the entry-level roles. You lose the experience that produces the senior ones. Mr Teo’s reassurance contains, quietly buried inside it, a question it does not answer: if graduates must enter already performing at a higher level, who — or what — taught them how to get there?

The One Thing That Cannot Be Subscribed To

So which jobs are actually safe?

Look at the roles that feel genuinely secure and ask what they share. The founding partner who puts her name on a legal opinion. The editor-in-chief who decides what the paper says tomorrow. The board director who votes on the acquisition. The entrepreneur who remortgages her apartment to back a bet she believes in.

None of these people are safe because they know more than the AI. In many areas, they know less — the machine’s recall is sharper, its reading list longer. They are safe because they carry responsibility that cannot be passed on. When things go wrong, they cannot point at a subscription service. They answer for it personally, professionally, and sometimes legally.

That kind of accountability, in a world where knowledge has become cheap, is the one thing that has become genuinely rare.

The professionals in the middle — producing the research, writing the first draft, building the models, preparing the analysis — have been well paid because what they knew was hard to acquire. That edge is shrinking. The competitive analysis that once took Marcus two days now takes a second-month analyst three hours. The contract review that needed three junior associates now takes an AI tool four minutes. The knowledge that justified the salary is, increasingly, available on demand.

What has not changed — what no tool can replicate — is the value of the person who stands behind the output. Who owns it. Who answers for it. Who sits across the table from a client or a regulator and says, without flinching: I am responsible for this.

Marcus is a good consultant. His clients trust him. But if he is honest with himself, much of what fills his week is knowledge work, not accountability work. Analysis he synthesises. Frameworks he applies. First drafts he polishes. A significant portion of it is precisely the kind of work his junior colleague, armed with AI, is already replicating — at a fraction of his cost.

The question he needs to ask himself is not will AI take my job? It is: which parts of what I do involve genuine responsibility, and which parts are simply knowledge that has now become free?

What This Means for Singapore

For Singapore, this is not just a personal question. It is a city-sized one.

About two-thirds of Singapore’s workforce holds a white-collar role. The city has built its economic identity around this: that skilled professional work commands a premium, that the people who do it are worth attracting from across the region, and that the whole ecosystem built around them — the rental condos, the international schools, the private hospitals, the restaurants — makes sense at the scale it currently exists.

That foundation is under pressure in ways that do not yet show up neatly in any single set of statistics. Companies do not put out press releases saying that AI has changed their hiring calculus. They simply slow down recruitment. Restructure quietly. Let contracts run out. The Employment Pass holder who might have filled the regional analyst role at a bank never arrives. The condo in Tanjong Pagar sits vacant a little longer than it used to. The effect accumulates in numbers that will not cite automation as their cause.

Budget 2026 points in the right direction — the National AI Council, the raised EP salary thresholds, the SkillsFuture credits being steered toward AI skills, NTUC’s AI-Ready SG platform launched in February. These are real efforts. But they share one assumption: that reskilling can move fast enough, and that the higher-value layer professionals are meant to step into is large enough to absorb everyone being eased out of the one below.

Nobody has shown convincingly, yet, that the new layer is as big as the one being hollowed out. And the tools are not slowing down to wait.

The View From Both Ends of the Ladder

Back to Marcus.

He is thirty-four. Seven years of solid work, a good reputation, a mortgage in Bishan, a child starting primary school next year. He is not panicking. But he is thinking, with a sharpness he did not feel two years ago, about what he is actually selling — whether the expertise he has spent his career building still commands what the market has been paying for it.

He does not yet have an answer.

Across the office, the analyst who handed him that Thursday report has already moved on to the next question. She has looked at what Marcus does, thought about what her tools can already handle, and concluded that the gap between them — measured in output, not in years — is narrower than either of them would find comfortable to admit.

She is not wrong.

The professional world was built on a straightforward story: that knowledge, built up over years, was hard to come by, and that rarity had a price. That story is not ending dramatically. It is ending in small, undramatic moments — a report produced in three hours instead of two days, a contract not renewed, a headcount request quietly declined.

The ladder has not disappeared. But the rungs in the middle — the ones that a generation of Singapore professionals is standing on right now — are being removed, one subscription at a time.

The graduates entering the workforce this year already know this. They did not come to wait at the bottom.

They came with the tool that makes waiting unnecessary.

The writer is a Singapore-based entrepreneur and observer of the intersection of technology, business, and society.

I have hundreds of notes in Google Keep. Quick thoughts captured on the MRT, snippets from articles, voice memos from late-night ideas, photos of whiteboard scribbles. For years, they sat there — a graveyard of good intentions.

The problem was never capturing ideas. Keep makes that effortless: it lives inside Gmail, clips from Chrome, transcribes voice on the go. The problem was everything that comes after. Those fragments never connected to each other, never turned into anything. I’d capture a brilliant insight on a Tuesday and forget it existed by Thursday. Sound familiar?

This is the trap most note-takers fall into. We optimise for frictionless input, then dump everything into a passive storage box with no mechanism to surface, connect, or act on what we’ve saved. The cognitive burden of organising all that chaos falls right back on us — which means it never gets done. Our “idea bank” quietly becomes an idea landfill.

Then Gemini arrived inside Google Workspace, and something clicked — literally.

The Quiet Revolution Inside Google Keep

Most coverage of Gemini in Workspace focuses on Gmail summaries or Docs drafts. But its integration with Keep is, I think, the most underrated development in the entire suite. It transforms Keep from a passive notebook into something closer to a thinking partner — one that can retrieve, synthesise, and execute on your behalf.

Here’s what that looks like in practice. I’ve grouped the key capabilities into three layers, from basic retrieval all the way to automated execution.

Layer 1: Retrieval — Finding What You Forgot You Knew

Semantic search across your entire archive. Traditional note search is keyword-dependent. If you can’t remember the exact words you used, the note might as well not exist. Gemini changes this fundamentally. You can ask something vague — “that idea I had about improving workshop productivity” — and it will surface a note from six months ago that never contained the word “productivity” at all. It understands meaning, not just keywords.

Bulk synthesis across time. Thanks to Gemini’s large context window, you can ask it to read through hundreds of untagged notes at once. I tested this by asking it to summarise everything I’d noted about kitchen renovation colours over the past year. It pulled together scattered entries — paint codes, brand names, mood-board thoughts from different months — into a single coherent brief. What would have taken me an hour of scrolling took about ten seconds.

Layer 2: Creative Synthesis — Connecting Ideas You Didn’t Know Were Related

This is where things get genuinely interesting, and where most people haven’t explored yet.

Forced idea collisions. You can ask Gemini to take two seemingly unrelated notes and find connections between them. A note about meditation techniques and another about software development workflow? Gemini might propose a “mindfulness-based sprint framework.” Not every combination yields gold, but the practice of forcing cross-pollination is how breakthrough ideas surface. It’s like having a creative collaborator who has read everything you’ve ever written.

Beating the blank page. When you have a vague intention but no structure — say, planning a home DIY project or sketching a content calendar — Gemini can generate a structured starting point directly inside Keep: steps, checklists, resource lists. It acts as a creative ignition switch, not a replacement for your thinking, but a way to bypass the paralysis of starting from zero.

Multimodal understanding. This extends beyond text. Take a photo of a hand-drawn business model on a whiteboard and save it to Keep. Gemini doesn’t just OCR the text — it interprets the diagram: the relationships between components, the implied priorities, even potential gaps. Similarly, a rambling voice memo captured while driving can be distilled into three structured action points without you ever listening to it again.

Layer 3: Execution — From Notes to Action, Automatically

This is the layer that transforms Keep from a thinking tool into a doing tool, and it’s what separates this from every other AI note-taking solution.

Extracting action items into Tasks and Calendar. After dumping a messy meeting debrief into Keep, you can instruct Gemini: “Find the three things I committed to, add them to Google Tasks, and set them due by Wednesday.” It generates the structured tasks and pushes them into your workflow — no switching between apps, no manual re-entry. When a note mentions a specific time — “client dinner Friday 7pm” — Gemini can create the calendar event and attach the Keep note as context. That throwaway line in your notes becomes a real commitment on your schedule.

Triggering automated workflows. With Google Workspace’s automation capabilities, Keep becomes a starting point for business processes rather than an endpoint. Imagine this: you jot down an expense in Keep and tag it “expense.” That triggers an automated flow — Gemini extracts the amount into a Google Sheet, and if it exceeds a threshold, sends an approval request via Google Chat to your manager. No code required.

Activating custom AI agents (Gems). If you’ve set up a custom Gem — say, a “Marketing Strategist” — you can point it at a rough product idea captured in Keep. The Gem reads your note, cross-references it with strategy documents in your Google Drive, and produces a preliminary marketing proposal. Your fleeting thought at 11pm just became a first draft by morning.

Feeding deeper research pipelines. For projects that need serious depth, your accumulated Keep notes — the fragments, the half-thoughts, the bookmarked quotes — can be imported as source material into NotebookLM. This creates a seamless chain from casual capture to deep research output, all within Google’s ecosystem.

The Second Brain, Properly Built

Tiago Forte popularised the “Second Brain” concept: the idea that we need an external system to offload our memory so our actual brain can focus on thinking and deciding. It’s a powerful framework, but its original implementation still relied heavily on manual organisation — tagging, filing, periodic reviews.

What Gemini adds to Keep is the missing piece: intelligence. Your Second Brain now has a division of labour. Keep handles frictionless input and storage — it’s the memory. Gemini handles connection, analysis, and output — it’s the logic. You no longer need to be disciplined about organising your notes, because an AI layer now does that cognitive work for you.

This reframes what notes even are. They stop being static records and start becoming raw material — ore waiting to be refined. Every photo, voice clip, text snippet, and half-formed idea you toss into Keep is now a potential input for structured business decisions, creative projects, or strategic planning. It’s note-taking as alchemy: turning base fragments into something valuable.

How to Start

If you’re already in the Google ecosystem, the friction to try this is almost zero:

This week, stop worrying about organising your notes. Just capture freely — in Keep, on your phone, through Chrome, via voice. Don’t tag anything. Don’t create folders. Just capture.

This weekend, open Gemini and ask it to review your recent Keep notes. Try prompts like: “What are the three most actionable ideas from my notes this week?” or “Find connections between my notes about [topic A] and [topic B].”

Next week, try the execution layer. After your next meeting, dump your raw notes into Keep and ask Gemini to extract action items into Tasks with deadlines.

You’ll feel the shift immediately. The ideas you used to lose will start working for you.

In an age of information overload, your ability to capture, connect, and convert ideas into action is your most durable competitive advantage. The tools to do it are already sitting in your Google account.

All your notes need is a brain. Now they have one.

What I’ve observed over the past two months is something different: most Chinese AI apps are simply “wrapper” applications—thin interfaces layered on top of large language models. These lack true innovation or defensibility and are destined to struggle. This illustrates a deeper shift: AI has itself become the new programming layer, making traditional software roles and approaches increasingly obsolete.

This challenge with AI wrapper apps isn’t exclusive to China; it’s part of a wider transformation affecting enterprise software across industries and geographies.

The Fall of the Giants

SAP (ERP) and Salesforce (CRM) are two classic pillars of enterprise software. Yet both stocks have fallen sharply this year—not because they rejected AI, but because AI is redefining what software must be. The core logic of enterprise software is being transformed, upending the old model and requiring a different response.

I once helped implement SAP at a company. The system was expensive and attempted to cover every business process, but no pre-built system can fit every company’s unique needs.

It’s like wanting chopsticks, but getting a full banquet—and being charged for it. Worse, the banquet lacks chopsticks, so you’ll need to bring your own.

A Small Experiment

Recently, my small business needed a basic workflow: assign tasks and send daily email reminders. Nothing fancy.

I searched for solutions, only to be pointed toward ERP or CRM systems. Really? For task reminders?

Then I asked AI. Since we use Microsoft 365, AI helped design a Power Automate workflow connecting Planner and Outlook. Now, tasks run smoothly—no more delays.

Previously, this would have required hiring a programmer or buying a SaaS solution. Now? No extra costs. No frustration.

I wondered if I was overreacting. At dinner, a Singapore government manager was surprised: “You did this yourself?”

The Real Revolution

This small case shows the real revolution: software is moving from pre-made products to AI-driven, tailored solution creation. It’s not about adopting AI wrappers, but about AI fundamentally altering how software is conceived and used.

When I need a digital solution — for work or personal matters — I no longer search for apps or software packages. I collaborate with AI to create something immediately usable, entirely tailored to my needs. You might call it “software” or an “app,” but it’s fundamentally different. It’s built for one purpose: mine.

This is the underlying reason SAP and Salesforce stocks are dropping: businesses no longer need rigid, expensive, one-size-fits-all platforms when they can use AI to create instant, bespoke software solutions.

I’ve used these enterprise platforms. They’re not just expensive — their rigidity is staggering. Business innovation gets strangled by programmers’ if-else statements. Companies pay enormous fees, then hire teams just to manage the software.

The new wave of Chinese AI “wrapper apps” follows the same outdated SaaS logic. Their quick obsolescence isn’t true innovation—it underscores why only truly AI-driven, customised solutions will thrive going forward.

What Comes Next

For programmers, this shift raises the bar. The future may belong to systems architects — people who understand how to orchestrate AI capabilities — rather than those who write conventional code.

As the Chinese rock musician Cui Jian once sang: “It’s not that I don’t understand — the world is just changing too fast.”

不是我不明白，这世界变化快。

Subscribe now