On the morning of 20 October 2026, two numbers about the same labour market arrived in the same inbox, and they do not agree.
The first is 1,23,000. That is the count of active technology openings in India in October 2026 — the highest in 18 months, up about 17% from a year earlier and about 5% from September, according to Xpheno's Active Tech Jobs Outlook. IT services alone accounts for 66,000 of those openings, up 57% year on year. In June the same measure touched a 28-month low of roughly 93,000. From there to here is what a recovery looks like on a chart.
The second number is 15,000. That is how many of those openings are entry-level. Roughly one in eight. And it fell 6% from September while the headline rose 5%.
That is the entire story, and it fits in a headline I could not have written better myself: the market is healing at the top and closing at the bottom. Not because fewer jobs exist, but because the first rung of the ladder has been quietly removed while everyone was looking at the top of the building.
The stat is not the story
Before we go further, a small demonstration of why I refuse to write posts that begin and end with one number.
On the same day that Xpheno reported openings up 17%, Naukri's JobSpeak index reported hiring in IT and software services down 4% year on year. Two credible firms, two opposite moods, one industry, one month.
Both are right. Xpheno counts active openings — advertisements live on portals and company sites on a given day. Naukri measures new listings — the flow of fresh advertisements in the month. A job posted in July and still unfilled in October counts once for Xpheno and not at all for JobSpeak. When companies slow down new postings but take months to fill the ones they have, the two series diverge exactly as they did this week.
Which is why the useful question is never "is hiring up or down?" It is "up or down for whom, counted how?"
| Xpheno, October 2026 | Openings | Share | Trend |
|---|---|---|---|
| All active tech openings | ~1,23,000 | 100% | +17% YoY, +5% MoM |
| Mid-senior | ~68,000 | 55% | +19% YoY |
| Entry-level (freshers) | ~15,000 | ~12% | +15% YoY, −6% MoM |
| IT services | ~66,000 | 54% | +57% YoY |
| Global capability centres | ~18,000 | 15% | +64% YoY |
I should also record the counterweight, because the temptation in pieces like this is to blame everything on the machine. Xpheno co-founder Kamal Karanth told reporters that most layoffs so far trace back to margin pressure, cost-cutting and restructuring under adverse market conditions — not directly to AI. Mainstream enterprise AI deployment, in his reading, is not yet mature enough to have replaced engineering roles at scale. I have quoted Karanth in this blog before and I quote him again because he keeps declining the easy headline. No claims, just comparison.
But notice what his caution does not explain. It explains why jobs disappeared. It does not explain why the jobs coming back are for people who already have ten years of experience.
The pay ladder has snapped in two, inside one company
Here is the same story told in rupees.
HCLTech has revised fresher salaries this year. Its "Elite Engineer" cadre — graduates from the IITs and NITs hired for frontier technology roles — starts at ₹18 lakh, rising to ₹22 lakh. The bulk of its campus intake comes through the Primus programme, where pay starts at ₹4.5 lakh in the Ascend tier and tops out at ₹10 lakh in the Polaris tier. Total fresher intake is expected to be around 11,000.
Read that again. The best-paid ordinary campus recruit at HCLTech earns less than the worst-paid Elite Engineer. Within one company, in one hiring season, the entry-level distribution is not a distribution any more. It is two separate staircases.
TeamLease Digital's Digital Skills & Salary Primer for FY27, drawn from 37,000 technology and digital roles, shows the same fracture across the industry:
- AI architect, 15 years' experience: median ₹10.86 million a year
- Entry-level AI-core roles: median ₹1.32 million
- AI-adjacent roles: ₹9–10.2 lakh
- AI-support roles: ₹4.8–5.4 lakh
A note on units, because this matters and it spread widely in the wrong form this week: ₹10.86 million is ₹1.086 crore, not ₹10.86 crore. The difference is a factor of ten, and at least one early summary of the primer got it wrong. Set the architect's ₹1.086 crore beside HCLTech's ₹4.5 lakh Ascend floor and you have a 24-fold spread between the top and bottom of a single industry, in a single year, in a single country. That is striking enough without inflating it. (A tenfold error in a salary number is itself a small lesson in how statistics travel.)
TeamLease also reports that only about 16% of India's IT professionals have AI skills, and projects FY28 salary growth of 16–21% for AI-core roles against 7–8% for AI-support roles. The ladder is not merely split. The two halves are moving apart.
The proof from the company that builds the machines
If you suspect this is an India-only artifact of our campus system, look at Anthropic's own internal workforce study, published on 16 October.
The share of junior employees in Anthropic's technical departments fell from 21% in early 2025 to 14% in 2026. The share of junior hires among new technical hires fell from 27% to 9%. By August 2026, AI agents were logging six times as many working hours as their human handlers — a ratio reported at 19:1 inside the research department. Engineers were submitting 5.1 times as many code changes as before.
Two honest caveats, because Anthropic's own data does not say what the headlines said. First, the company's technical headcount kept growing in absolute terms; it is the proportion of juniors that collapsed. Second, independent analysis of Anthropic's engineering roster suggests the firm never hired many juniors to begin with — median prior experience of 12.2 years, campus recruitment close to non-existent. The trend is real. The causation is messier than the graph.
Then, this week, Siemens said the quiet part out loud. Judith Wiese, its chief people and sustainability officer, warned against eliminating entry-level jobs as AI takes over routine tasks. Her example was software developers: if AI writes the simple code, the junior's job becomes critically examining it — and you cannot learn to judge code you have never written. Remove the entry level, and in ten years you have no one able to supervise the machine.
I have been groping for a name for this and I think I have one: the tragedy of the cognitive commons. Every firm is rationally economising on juniors, and every firm is thereby depleting the shared stock of experienced people on which all of them will depend. Nobody is doing anything wrong. Everybody is making the same mistake.
Why this is structural, not a bad quarter
Three independent measurements, all published in the last few days, say the same thing from three directions.
1. The IMF's 40 million postings
The IMF's October 2026 Regional Economic Outlook: Asia and Pacific examined 40 million online job postings across 19 economies through April 2026. Employer demand in occupations more exposed to AI fell by nearly 10% relative to less-exposed occupations, within the same economy, industry and month.
The most interesting finding is what happened next. As AI got better — from tasks taking seconds, to an hour, to a full day — the gap did not widen. It held roughly steady. Because a new demand was rising underneath it: prompt engineering, retrieval-augmented generation, building AI agents. These skills barely existed before 2023 and are now the fastest-growing requirement inside the very occupations being hollowed out. But only about 6% of firms in Asia ask for AI skills at all, which tells you how early in the transition we are, and how much of the displacement is still ahead.
2. CMIE: fewer graduates looking, not more graduates working
India's graduate labour participation rate fell to 52.26% in September — the lowest in CMIE data going back at least to October 2021. Over three years the graduate population grew 28.6%. Total employment of graduates grew 26.8%. Salaried jobs grew 11.9%.
That is the arithmetic of discouragement. It is not that graduates stopped wanting work; it is that the specific kind of work a degree was supposed to buy grew at less than half the rate of the supply of degrees. The number of graduates actively seeking work fell by over 3 million in a year.
3. NITI Aayog: the degree no longer filters
NITI Aayog's August 2026 report, Reimagining Skilling for Viksit Bharat, puts graduate unemployment at 13% against a national rate of 3.2%. Only 8.25% of graduates work in a job that matches their qualification. And 2 crore of the 3.4 crore students in undergraduate programmes are enrolled in B.A., B.Sc. and B.Com. — degrees that, as currently taught, lead nowhere in particular.
Meanwhile the demand side is telling us exactly what it wants. Talent500's survey of 2,000 Indian tech professionals finds 71% want hands-on AI upskilling and 29% say their employer provides it. And 79% would drop an application, or turn down an offer, if there were no clear growth path. TCS now reports 2,70,000 "AI-proficient" staff while cutting 23,460 jobs. Both facts are true at once. That is what a workforce being reclassified looks like.
So what?
The answer cannot be "more degrees." We have run that experiment for twenty years and the CMIE numbers are the result. Nor is the answer "learn AI" as a slogan — 71% of people already want to, and the employers who could give them that experience are the same ones not hiring them.
Three things would actually move the needle, and I have argued all three before.
One: proof of work must replace the credential as the unit of hiring. Employers are already behaving this way — the whole reason a ₹18 lakh Elite Engineer and a ₹4.5 lakh Ascend hire can sit in the same batch is that one has demonstrable frontier skills and the other has a degree. The policy task is to make proof of work producible at scale: credit-bearing projects, apprenticeships that end in a real assessment, open-source and internship records that a recruiter can verify in minutes. I wrote about how badly India underuses apprenticeships in Discovering the Obvious? — the TeamLease NETAP findings then were that apprentices turned productive in 45 days and delivered 100–150% returns to employers. Nothing about that arithmetic has weakened since.
Two: a matching rail, not a search engine. When one in eight openings is for a fresher, the cost of a bad match rises for both sides, and the burden falls hardest on the candidate with the least information. I set out the case in my National Employment Grid whitepaper: resumes uploaded once, vacancies posted once, AI scoring every pair daily, InterviewBots doing first-round screening, alerts pushed to both sides every morning. Job search itself is obsolete. The reason it persists is that no one has built the rail.
Three: publish AI-exposure in every job description. This is the cheapest of the three and the one I would legislate first. If an employer must state, in the advertisement, which parts of the role AI already performs and which parts are reserved for the human, then three things happen at once. Candidates stop applying blind. Colleges get a live map of which curricula are selling. And firms are forced to notice, in writing, that they have stopped hiring juniors — which is the first step towards not doing it.
The National Job Skilling Policy work already points this way: an annual skills and employment survey, a live demand-mapping portal, employer incentives for formal training. My Graduates Without Jobs post in March made the supply-side case. This is the same argument with the year's receipts attached.
What to watch on 23 October
Infosys reports Q2 FY27 on Thursday. The previews expect organic constant-currency growth of about 1.1% sequentially, EBIT margin expansion from Project Maximus, deal TCV in the $2.5–3.5 billion band, an LTM attrition rate of 13%, and a headcount that fell by 532 last quarter to 3,28,062. What I will be looking for is not revenue. It is the net addition to headcount, and any sentence at all about fresher intake.
If Infosys confirms the pattern — more revenue, fewer people, no freshers — then October 2026 will be remembered as the month the contradiction became the norm. If it complicates the pattern, I will happily write that too. That is the point of keeping receipts.
The bookend
In March 2005 I wrote a note called Problem Statement. It counted over 100 million jobseekers in India, 44 million of them registered with some 945 government employment exchanges, which between them had placed 1.5 lakh people in 2003. My conclusion then was that the problem was not a shortage of jobs or of candidates but the absence of a fast, accurate, automatic matching system between them.
Twenty-one years later, the exchanges are digital, the resumes are counted in crores, and the matching problem has been solved many times over — for engineers with eight years of experience. For the 21-year-old with a degree and a project on GitHub, it has been solved in the wrong direction. The market now knows precisely who it wants, and it has decided it does not want to train anyone to become that person.
India has more openings and fewer doors. That is not a paradox. It is a choice, made 15,000 times at the bottom and 1,08,000 times at the top, in a single month.
Regards, Hemen Parekh (hcp@recruitguru.com)