HomeThere is a quiet experiment running inside every professional-services firm, bank and consultancy in Singapore right now, and almost nobody has named it out loud. The experiment is this: we have built a machine that is genuinely excellent at the work we used to hand to twenty-three-year-olds, and we have not decided what to do about it.
For a century, the bottom rung of the white-collar career ladder looked roughly the same. A bright graduate joined as a junior analyst, a research associate, a trainee auditor, a paralegal. They were handed the unglamorous interior of the profession — pulling data, reconciling figures, formatting decks, drafting first-pass memos, summarising documents nobody senior had time to read. It was tedious, it was formative, and it was the price of admission. You did the grunt work, you absorbed how the work actually worked, and three years later you had quietly become someone whose judgment was worth paying for.
Now a model does the grunt work in seconds. It pulls the data, reconciles the figures, drafts the memo, summarises the document, and formats the deck — faster, cheaper, and at three in the morning if you ask it to. The economic logic that put a graduate in that seat has, almost overnight, stopped adding up. And so the question landing on desks across the island is brutally simple: if AI can do what junior analysts did, why hire junior analysts at all?
The lazy answer is to stop. Quietly thin the graduate intake, let the AI absorb the entry-level load, and book the saving. It is already happening, in firms that would never admit it. And it is one of the most short-sighted moves a knowledge business can make — because it confuses the task the junior performed with the reason the junior existed. The grunt work was never the point. The grunt work was the apprenticeship.
This is the real story of junior analysts versus AI. Not a replacement. A redesign that Singapore is uniquely equipped to get right — or, if it sleepwalks, to get expensively wrong.
The core shift: the bottom rung was always an apprenticeship in disguise
To understand why "just stop hiring juniors" is a trap, you have to understand what the junior role was actually for — because it was never what the job description said.
On paper, a junior analyst exists to produce output: the model, the memo, the reconciliation, the research note. That is the visible function, and it is exactly the function AI now performs. If that were the whole story, removing the junior would be a clean efficiency win, and we could all go home.
But the visible function was never the real one. The bottom rung was an apprenticeship wearing the costume of a job. The output the junior produced was almost incidental; the point was what producing it did to the junior. By pulling the data themselves, they learned where the data lies and where it breaks. By drafting the memo badly and being corrected, they learned what a good memo argues and why. By sitting in on the meeting taking notes, they absorbed how a senior partner reads a room, handles a difficult client, decides which number to trust. The grunt work was the medium through which judgment was transmitted from one generation of professionals to the next. You cannot download that. You earn it, task by tedious task.
This is the shift the AI conversation keeps missing. Everyone is asking whether AI can do the junior's tasks. The answer is increasingly yes. But that is the wrong question, because the tasks were a side effect. The right question is: if AI now does the apprenticeship work, how does anyone become senior?
Consider the mechanics of how expertise actually forms. A senior credit analyst does not know how to price risk because someone taught them the formula — the formula is in every textbook. They know how to price risk because they spent years building thousands of small models, each one slightly wrong, each correction adding a layer of intuition that no formula captures. The intuition is the asset. The thousands of models were the gym. Remove the gym and you do not get a leaner athlete. You get someone who never learned to lift.
The vanishing first decade
There is a temporal cruelty to this that compounds the danger. The damage from cutting the bottom rung is not visible for years.
A firm that thins its graduate intake today will look more efficient immediately. Costs fall, the AI handles the load, margins improve, and the quarterly numbers reward the decision. The seniors are still there, still excellent, still carrying the firm on the judgment they accumulated back when the apprenticeship still existed. Everything looks fine. The pipeline does not announce that it is broken. It simply, silently, stops producing.
Then five, seven, ten years pass. The senior cohort retires, moves on, burns out. And the firm reaches for the next generation of judgment — the people who should by now have graduated from juniors into the experienced professionals who run the place — and finds the bench empty. Not because those people left, but because they were never hired, never trained, never given the thousands of small reps that turn a graduate into a partner. The firm optimised away its own future and booked it as a cost saving. By the time the gap is visible, it is a decade too late to close. You cannot conjure a senior analyst on demand; the lead time on judgment is measured in years, and the firms that severed the pipeline will be the ones bidding frantically for the few seniors the market still has.
This is the heart of the matter, and it is why "stop hiring juniors" is not a clever efficiency move but a slow-motion act of self-harm. AI did not just automate the junior's tasks. It dissolved the apprenticeship those tasks delivered — and unless someone deliberately rebuilds that apprenticeship in a new form, the entire mechanism by which a profession renews itself quietly fails.
Why this is a world-class problem, not a Singapore one
It helps to see how large this actually is before localising it. The World Economic Forum's Future of Jobs research has projected something like 170 million new roles created and 92 million displaced globally by 2030 — a net positive of roughly 78 million — alongside a finding that the overwhelming majority of employers, on the order of 86%, expect AI and information-processing technology to transform their business within the same window. Treat those figures as reported and approximate; the precise digits matter less than the shape they describe.
And the shape is unambiguous: this is not a story of mass disappearance. It is a story of churn and recomposition. Roles vanish and roles appear, and the net is positive — but only in aggregate, and only over time. The aggregate hides a brutal distributional truth. The roles being displaced skew toward the routine, repeatable, entry-level work that historically sat at the bottom of career ladders. The roles being created skew toward judgment, oversight, and the building and governing of the very systems doing the displacing — work that sits higher up. The net number is positive. The path from the displaced bottom to the created top is exactly the apprenticeship that AI is dissolving. Which means the global "net +78 million" is not a reassurance; it is a redesign challenge in disguise. The jobs will exist. The question is whether anyone will be qualified to fill them, and that depends entirely on whether we rebuild the on-ramp.
There is a related signal worth naming. Microsoft's recent Work Trend research described what it called a "redesign gap" — the observation that AI-driven productivity gains are sharply outpacing the rate at which organisations actually redesign how work gets done. Firms are bolting AI onto old org charts and old role definitions and wondering why the promised transformation feels hollow. The redesign gap is precisely where the junior-analyst problem lives: the technology has changed what the bottom rung should be, and almost no one has changed the bottom rung.
A young analyst at a desk in a modern Singapore office, AI dashboards glowing alongside printed work, conveying apprenticeship in transition
The misread: replacement versus task-automation
Now to the mistake itself, because it is being made in real boardrooms this quarter, and it is expensive in a way that does not show up for years.
The misread runs like this. A managing partner or a head of operations looks at what AI can now do, looks at what their junior cohort costs, and reasons backward to a headcount. The model writes a better first-draft memo than our first-years. It pulls data faster than our analysts. So why are we paying for forty graduates a year? The finance team models the saving. It is large and immediate. A target is set: cut the intake, let AI absorb the load. The whole exercise is framed, from its opening slide, as a cost-reduction programme dressed in AI clothing.
This fails for a reason that is almost mechanical, and it is the same reason it fails everywhere AI meets a workforce: AI does not replace jobs. It replaces tasks.
A job is a bundle of tasks — some routine, some judgment-heavy, some relational, some regulated. A junior analyst's week is not one homogeneous block of automatable work. It is, say, 65% retrieval, formatting, reconciliation and first-draft drudgery — genuinely automatable — and 35% something else entirely: learning to spot when a number feels wrong, sitting in the room where judgment is exercised, building the relationships and the instincts that compound into seniority. When you point AI at the role, it does not vaporise the role. It dissolves the automatable 65% and leaves the 35% standing — except that 35% was the part that mattered, and you have just removed the seat from which it was learned.
The leader who frames this as headcount reduction makes two errors at once. First, they cut for the wrong number — chasing the salary saving rather than the compounding value of a renewed talent pipeline. The salary of forty graduates is a visible, modelable, satisfying number. The cost of an empty senior bench in 2034 is diffuse, deferred, and absent from the spreadsheet entirely. A cost-first analysis structurally overcounts the saving and is blind to the loss.
Second, and worse, they automate the wrong thing. Impatient to bank the saving, they remove the role rather than redesigning it — and in doing so they sever the apprenticeship along with the drudgery. The intelligent move was to automate the 65% and intensify the 35%, accelerating how fast a junior reaches judgment. The cost-first move automates the 65% and deletes the 35% by removing the human who would have grown into it. Same technology, opposite outcome. One firm builds a faster pipeline; the other quietly snaps it off and books the broken machinery as efficiency.
Every firm cutting its graduate intake to AI is making the same bet: that it will never again need the seniors those graduates would have become. It is a bet against your own future, and you will not learn you lost it until it is far too late to re-enter.
There is a third error, quieter and arguably the most corrosive. A replacement frame poisons the very people it keeps. The remaining juniors — and the seniors watching how the firm treats its young — read the signal clearly: this organisation views talent as a cost to be automated, not an asset to be grown. The best of them leave. The rest disengage. And the tacit knowledge transfer that made the apprenticeship work in the first place — the senior who took time to explain why a memo was wrong, the analyst who flagged the number that looked off — quietly stops, because nobody invests in training a cohort the firm has signalled it intends to shrink. The replacement framing sabotages the culture that produces judgment. The frame you choose is not a communications footnote. It is an input to whether your firm still has a future inside it.
Redesign, not replacement: the three-bucket model
If "cut the juniors" is the wrong frame, what is the right one? It begins with a far better question — the most useful question any leader can ask of AI:
"If the machine clears the routine work, what could our people finally learn and do with the time?"
That reframing produces a concrete, repeatable operating model. You do not start with job titles or headcount targets. You start with tasks. Take the junior analyst role — or the trainee auditor, the research associate, the paralegal — and decompose it into the discrete tasks the person actually performs across a representative month. Then sort every task into one of three buckets.
Bucket one — what machines do better
These are the tasks where a capable model genuinely outperforms a graduate on speed, consistency and cost. Pulling and cleaning data. Reconciling figures across sources. Drafting the first pass of a memo or research note. Summarising a hundred-page filing into a brief. Formatting the deck. Building the standard model from the standard template. Surfacing the anomalies in a dataset. In a junior role, this bucket is large — frequently the majority of the hours — and it is exactly the layer AI is now ready to absorb. Route this work to the machine without sentimentality. It is genuinely better at it, and clinging to it as "how juniors learn" is mistaking the scaffolding for the building.
Bucket two — what humans do better
These are the tasks where the human is not merely preferable but load-bearing, and where, crucially under Singapore's regulatory frame, a human must own the outcome. Judging when a clean-looking number is quietly wrong. Reading a client's hesitation in a meeting. Owning a recommendation and being accountable for it. Making — or being trained toward — the consequential call that a model can inform but cannot own. This bucket is small in volume and enormous in value, and it is where every senior professional ultimately lives. The entire purpose of the redesigned junior role is to get the graduate to this bucket faster than the old apprenticeship ever could — not to wall them off from it.
Bucket three — what they do better together
This is the bucket most firms forget exists, and it is where the upside hides. It is the junior who, freed from sixteen hours of weekly reconciliation, now sits in three more client meetings and builds judgment in months instead of years. It is the analyst whose AI handles the model build, so their human energy goes to interrogating the assumptions — the exact skill that makes a senior. It is the trainee who reviews and corrects the AI's first-draft memo, learning what "good" looks like by editing rather than by staring at a blank page. Together they are not a smaller cohort doing the same work. They are a faster-maturing cohort doing higher-value work sooner. The AI becomes the apprenticeship's accelerant rather than its executioner — if you design the role to make it so.
The reason bucket three is so easy to miss is that it never appears in the first round of cost modelling. A spreadsheet asking "how many juniors can we cut?" finds bucket one and stops. It has no column for "judgment formed faster," "pipeline renewed," or "senior bench filled in 2032," because those values are diffuse, deferred, and land on the revenue and resilience lines rather than the cost line. The cost-first analysis sees the headcount you can delete and is structurally blind to the talent you could compound. The redesign-first analysis inverts it: freed hours become accelerated apprenticeship, and the entry-level role is rebuilt around judgment-from-day-one.
When you run this honestly, the right move stops looking like "hire fewer graduates" and starts looking like "hire graduates into a fundamentally redesigned role." This is the same redesign discipline that separates firms compounding gains from AI from the ones quietly degrading their own capability while congratulating themselves on a leaner intake. The house rule is simple enough to put on a wall: redesign before you reduce. Reduce first and you sever the pipeline blind. Redesign first and the right headcount — which may even be more graduates, each far more productive — reveals itself.
A clean conceptual diagram of three buckets — machine tasks, human judgment, and the two working together — rendered as elegant interlocking forms
What this means for Singapore
Singapore feels this more acutely than almost anywhere, and for reasons that are structural rather than incidental. This is a small, dense, talent-scarce economy whose entire competitive position rests on the quality of its knowledge workforce. We do not win on cheap labour or vast scale. We win on trust, precision and the calibre of the people in the room — in the banks, the law firms, the consultancies, the regulators. Snap the talent pipeline here and you are not trimming a cost line. You are eroding the national asset.
Look at the institutions where this plays out most sharply: the major banks — DBS, OCBC, UOB — and the professional-services and financial ecosystem around them. These are firms that have spent decades building a reputation for rigour, and that reputation is carried entirely by senior judgment. A senior credit officer who knows when a deal smells wrong. A senior auditor who knows where the bodies are buried in a set of accounts. A senior relationship manager whom a family office trusts with three generations of wealth. Every one of those people was a junior analyst once. They became what they are through exactly the apprenticeship that AI is now positioned to dissolve. The question for Singapore's flagship institutions is not whether AI can do junior work — it plainly can — but whether they will redesign the junior role to keep producing seniors, or quietly let the on-ramp erode and discover the shortage a decade out.
There is a sharper edge to this in finance specifically, because of how the regulator thinks. The Monetary Authority of Singapore's FEAT principles — Fairness, Ethics, Accountability and Transparency — require that consequential, customer-impacting decisions remain owned by a human who can explain and answer for them. Read correctly, FEAT is not a constraint on AI so much as a permanent guarantee of human judgment in the financial system. The model can prepare, surface, draft and flag; a human must decide and be accountable. That guarantee is precisely the senior capability the apprenticeship exists to produce. So in Singapore finance, the regulator has effectively mandated a permanent home for the very judgment that junior analysts must be trained toward. A bank that automates away its junior pipeline is, in a real sense, automating away its future capacity to satisfy FEAT — because in ten years it will lack the trained humans the principles require to be in the loop.
The graduate's-eye view
It is worth standing in the graduate's shoes, because Singapore has a particular sensitivity here. This is a society that has made a generational promise around education and meritocracy: study hard, earn the credential, and a ladder will be there to climb. A cohort of bright graduates is now walking out of NUS, NTU, SMU and the polytechnics into a market where the bottom rung of that ladder is visibly thinning. The anxiety is real and it is rational. If the entry-level job is the apprenticeship, and the apprenticeship is being automated, what exactly is a twenty-three-year-old supposed to climb?
The honest answer — and honesty matters more than comfort here — is that the entry-level job is not disappearing, but it is being rewritten, and the graduates who thrive will be the ones who understand the rewrite. The old entry-level value proposition was "I can do the routine work." That proposition is now worth very little, because the machine does it for a fraction of the cost. The new proposition is "I can direct, interrogate and improve the machine's work, exercise judgment earlier, and own the relationship the machine cannot." The graduate who treats AI as a threat to hide from will lose. The graduate who treats it as the tool that lets them skip three years of drudgery and reach real judgment faster will be more valuable, sooner, than any junior analyst in history. Singapore's education and training system has a direct interest in teaching that rewrite — and the firms have a direct interest in designing roles that reward it.
This is also where the firms that get the redesign right will win the war for talent. The best graduates are not naive; they can see which firms are thinning the intake to bank a saving and which are building a genuinely redesigned, judgment-first early career. The redesigners will attract the strongest cohort precisely because they are offering a future rather than a holding pattern before automation. In a market as reputation-dense as Singapore's, that signal travels fast.
The Singapore enablers: a system built for redesign
Here is the part that should change the calculus for every Singapore leader weighing "cut" against "redesign" — because Singapore, almost uniquely, has built the institutional machinery to make redesign the cheaper, easier, better-supported path. Most firms are simply not using it.
Start with the labour model itself. Singapore runs on tripartism — government, employers and unions moving in concert — and it has navigated every prior wave of disruption, from manufacturing offshoring to the financial crisis, through coordinated redesign rather than raw market churn. AI is the next wave, and the same machinery is pointed at it. This matters enormously for the junior-analyst question, because it means a firm choosing to redesign rather than reduce is moving with the national grain, not against it — and is eligible for real, funded support in doing so.
The concrete instruments are already in place. Workforce Singapore (WSG) and e2i run Career Conversion Programmes and Jobs Redesign support designed to do precisely what the three-bucket model demands: help employers restructure roles around higher-value work and fund the reskilling that moves a person from a shrinking task-set into a growing one. SkillsFuture underwrites the continuous reskilling that keeps a workforce current. These are not abstractions; they are budgeted, operational programmes built on the explicit philosophy that people should be redesigned into new roles, not discarded. A firm that frames its AI shift as "we are reskilling our juniors into AI-augmented judgment roles" can tap this support. A firm that frames it as "we are cutting our intake" forfeits it — and absorbs the reputational cost on top.
There is a quiet brilliance to how this aligns the incentives. In most economies, when a firm automates, the displaced worker is the firm's liability or the individual's misfortune, and the path of least resistance is simply to cut. In Singapore, the path of least resistance has been deliberately engineered to be redesign, because the funding, the institutions and the social expectation all push that way. The state has, in effect, subsidised the better strategy. A leader who reaches reflexively for the headcount cut is not just making the strategically weaker choice — they are leaving real public support on the table and swimming against a current built specifically to carry them.
And then there is trust, which in Singapore is both a moat and a constraint. This is a high-trust, reputation-dense market where word travels and institutions are judged over decades. A bank that is seen to handle its AI transition with care — reskilling its young, engaging its unions, redesigning rather than discarding — protects something larger than its margin. It protects the trust that is the actual basis of its franchise. Getting that posture right — defensibly, in line with FEAT, and in a way the regulator and the public will respect — is exactly the governance and grants work our sister practice FMC Collective exists to build into a transformation from the first day rather than bolt on after the headlines. The firms that treat governance and reskilling as the architecture of the redesign, not the cleanup after a cut, are the ones that will keep both the regulator's confidence and the public's.
A Singapore tripartite scene blending public institutions, employers and young workers with subtle technology motifs, conveying coordinated transition
The operator's playbook: five moves to run now
Strategy is only as good as the next action it produces. If you lead a bank, a professional-services firm, a consultancy or any judgment-heavy business in Singapore, the junior-analyst problem compresses into five concrete moves. Run them in order — the order matters.
1. Map the apprenticeship, not just the tasks
Pull a representative month of a junior's actual work and tag every task: routine, judgment, relational, regulated. You will almost always find that the majority is genuinely automatable — the retrieval, the reconciliation, the first drafts. But do a second pass the cost-cutters skip: for each routine task, ask what was the junior learning by doing it? That second map is the one that matters. It tells you not just what to automate, but what learning you must rebuild elsewhere once the task is gone. Skip this and you will automate the drudgery and accidentally automate the education with it.
2. Automate the routine — and redirect the freed hours into judgment
Deploy AI against bucket one without apology. Then make the decisive move almost everyone misses: explicitly reinvest the reclaimed hours into accelerated apprenticeship. More client exposure, sooner. More time interrogating the AI's output rather than producing the input. More seats in the rooms where senior judgment is exercised. The reclaimed time is not a saving to bank; it is fuel for faster talent development. A junior freed from sixteen hours of reconciliation should be in three more meetings, not made redundant. The firms that win treat freed capacity as an investment in capability, not a cost to delete.
3. Redesign the entry-level role upward
Rewrite the graduate job description around its new core: directing AI, interrogating its output, exercising judgment early, owning relationships. The role did not get smaller. It got harder, and more valuable, sooner. Title, expectations and development path should reflect that. A graduate who reviews and corrects an AI's first-draft memo on day one is learning what "good" looks like faster than one who spent eighteen months producing first drafts from scratch. Design the role so AI compresses the years-to-judgment rather than removing the seat where judgment was learned.
4. Keep the human in the loop where it counts — and train toward it deliberately
Bucket two is sacred, and under MAS FEAT it is non-negotiable in finance. Consequential decisions — credit, risk, anything carrying regulatory or client-trust weight — are owned by an accountable human. So build the junior's development path explicitly toward that ownership. Make the apprenticeship's destination clear: you are being trained to be the human FEAT requires in the loop. This reframes the entire early career from "do the routine work" to "become the judgment the firm and the regulator cannot do without." It is both better training and a more honest promise to the graduate.
5. Reskill with the system, not against it
Tap the machinery Singapore built for exactly this. Career Conversion Programmes, Workforce Singapore and e2i Jobs Redesign support, SkillsFuture — these fund the move from a routine role to a redesigned one. A firm that runs its junior redesign through these programmes lowers its own cost, signals good faith to the regulator and the public, and moves with the tripartite grain. A firm that cuts instead banks a one-time saving and forfeits all of it. Redesign, don't release — and let the national system carry part of the load it was built to carry.
Run these five and the AI shift stops being something that happens to your talent pipeline and becomes something you steer deliberately — building a faster, sharper generation of professionals while the firms around you quietly hollow out their own benches. This is the end-to-end redesign we help Singapore businesses run: finding the real automation surface and building the AI safely with Freemansland, and locking the governance, grants and reskilling posture in place with FMC Collective so the transition is as defensible as it is efficient.
The investor's close: the operating leverage hidden in the pipeline
Now for anyone allocating capital, because this is where the argument cashes out on an income statement — and where the market is currently mispricing the difference between firms.
The naïve reading is that a professional-services firm cutting its graduate intake to AI will show lower costs and fatter margins. In the short run, it will. And it is exactly the wrong number to back. Watching it will lead you to reward the firms quietly dismantling their own future and to overlook the ones building durable advantage.
Here is the mechanism that actually matters. A knowledge business has always scaled the way a galley scaled — more output meant more juniors at more oars. Revenue and headcount marched in lockstep. AI breaks that chain: when the routine work migrates to a machine that costs a fraction of a graduate's salary and scales without hiring, the link between growth and headcount finally decouples, and a service firm can begin to exhibit something closer to software economics. The number to watch is revenue per employee, and beneath it, the operating leverage of the whole firm.
But — and this is the crux — that leverage only materialises if the firm is redesigned to capture it, and the talent pipeline is the hidden variable that determines which way it breaks. Two firms can buy the identical AI and end up in opposite places. The firm that cuts its juniors shows a flattering margin for a few years, then hits a wall: an empty senior bench, a war for the few experienced professionals left, soaring senior compensation, and a quiet erosion of the judgment-driven quality that justified its fees in the first place. Its revenue per employee looks great until the engine of revenue — senior judgment — has no one left to replace it. It optimised the present by mortgaging the future, and the bill arrives precisely when it can no longer be paid quickly.
The firm that redesigns shows something categorically different and far more durable: a leaner, AI-augmented junior cohort that matures into senior judgment faster than the old model allowed, a continuously renewed pipeline, rising revenue per employee that is sustainable because the talent engine keeps running, and quality that holds because the seniors of 2034 are being trained today. Same technology. Same starting point. Completely different trajectory — and the difference is invisible on this quarter's income statement and decisive on the next decade's.
The question for an investor is no longer "is this firm using AI to cut costs?" Nearly all of them are. The question is "is this firm redesigning its talent pipeline around AI, or quietly cannibalising it?" Only one of those is durable operating leverage. The other is a margin that is borrowing against a future it cannot repay.
So the signal to read in any knowledge business is not the size of the cost cut. It is whether the freed capacity is being reinvested into faster talent development or simply booked and gone. The firms that redesign the bottom rung will compound capability for years. The firms that snap it off will look brilliant until, suddenly, they cannot field a senior team — and in a market as talent-scarce and trust-dense as Singapore's, that reckoning arrives faster and bites harder than anywhere else.
AI did not come for the junior analyst. It came for the junior analyst's tasks — and in doing so, it put a profession's entire renewal mechanism on the table. The firms that read the headline will thin their intake, bank the saving, and quietly break their own pipeline. The firms that read the design will rebuild the apprenticeship in a new form, reach judgment faster than anyone before them, and own the next generation of talent. The lesson is not in the cut. It is in the redesign — and Singapore, of all places, has built the system to get it right. For more on how the AI workforce shift is being decoded for Singapore, explore the rest of our Insights.