Is Generative AI really replacing Indian IT jobs?

It is dissolving a category of work, not causing a temporary slowdown. Net hiring at India’s top IT firms fell 86% between FY22 and FY26 — from 600,000 engineers added in a year to 140,000 — and roughly 80,000 mid-level engineers at TCS, Wipro, HCL and Tech Mahindra were laid off over two years as routine coding moved into AI models. (Sources: NASSCOM, EY, Deccan Herald.)

Why are experienced engineers being laid off before fresh graduates?

For thirty years seniority was a moat — experienced engineers knew which work to skip. AI made it cheap to explore every option, so that judgment of when to do nothing now hides possibilities the model can surface. The first layoffs at TCS, Wipro, HCL and Tech Mahindra targeted engineers with 12–15 years of experience. Experience became a liability.

How many Indian IT jobs will AI affect?

On roughly the same $250 billion in revenue, the sector employed about 5 million people in the mid-1990s and is projected to need around 2 million by 2030 — the same output produced by building the systems that write routine code rather than writing it by hand.

What skills still pay in an AI-first IT industry?

Two. Workflow thinking — seeing the whole system of who feeds in, who approves and what gets remembered, since single-task execution is now nearly free. And proof of work — shipped systems, open-source contributions and public case studies, which replace credentials that AI made cheap to fake.

The Bifurcation: AI and Indian IT Jobs

> Rehearsal Research · Visual Story
dossier_02 · indian_workforce · 2026-05
WHAT IS
GENERATIVE AI
DOING TO INDIAN
IT?
A six-minute dossier on the hiring engine that just stopped — and what it means for the next decade of Indian work.
anchored on: NASSCOM FY26 [1] · NASSCOM × Deloitte [2] · NITI Aayog [3] · EY 2025 [5]
▼begin scan

▼continue scan
[STAT_01]
0%
drop_net_hiring · FY22 → FY26
src: NASSCOM [1]
[STAT_02]
0
laid_off · India IT · 2025-26
src: TCS · Wipro · HCL · Tech-M [1][7]
[STAT_03]
0:1
ai_demand : supply · 2026
src: NASSCOM × Deloitte [2]
▒░narrator_track░▒
Net hiring at the top IT firms has fallen 86%. In FY22, they collectively added 600,000 engineers in a year. In FY26, that number is 140,000. Roughly 80,000 mid-level engineers at TCS, Wipro, HCL and Tech Mahindra have been laid off in two years.
For thirty years, Indian IT services was the country's largest middle-class engine. Every year, 1.5 million engineering graduates joined it. No more.
This is not a slowdown. A category of work — the kind of code the entire IT services industry was built on — is being dissolved by Generative AI.

▼render bifurcation
> resolve_premise()
> claim = "industry did not shrink — it split"
> proof.length = 5 panels
> proof.anchor = NASSCOM × Deloitte × NITI_Aayog
▒░narrator_track░▒
Something like this has happened before. In the 1990s, factory work moved overseas — to Vietnam, to China, eventually to Tamil Nadu. The salary followed wherever the labour was cheaper.
This wave is the same shape, one layer up. Routine code work has moved into the model itself. The salary has moved with it — to the people building the models, not the people running their output.
What follows is a map of which work fell on which side.
                                
                  ▒             
              ▒  ▒█▒            
             ▒█▒ ███     ░      
         ▒  ▒███░███ ░  ░█▒     
        ░█▒ ████ ███░█▒ ███     
     ▒  ████ ████████████████   
    ▒█▒ ████ ████████████████   
   ▒███ █████████████████████   
████████████████████████████████
whitefield · hinjewadi · hitec city
built 1995–2020s · constraint = geographic
↓ split ↓
┌─┬─┬─┬─┬─┬─┬─┬─┬─┬─┐
│0│0│0│1│0│1│0│1│1│0│├─┼─┼─┼─┼─┼─┼─┼─┼─┼─┤
│0│1│0│0│0│0│1│0│0│1│├─┼─┼─┼─┼─┼─┼─┼─┼─┼─┤
│0│0│0│0│0│1│0│0│1│0│├─┼─┼─┼─┼─┼─┼─┼─┼─┼─┤
│0│1│0│0│1│1│1│0│0│1│
└─┴─┴─┴─┴─┴─┴─┴─┴─┴─┘
model weights
built 2022→ · constraint = cognitive

> render: bifurcation.svg
[FIG_01]·bifurcation_paths · Y0 → Y10salary_lpa
0₹40LY0Y5Y10work AIcan't dowork AInow does
▒░narrator_track░▒
Twenty years ago, a fresh graduate and a fifteen-year senior developer did the same kind of code. The senior was just faster. Both translated business requirements into APIs, CRUD apps, status reports. The bigger the company, the more such people it needed.
That work now happens inside the model. The chart shows what's left. The green line is the people who build the model. The red line is everyone else — the work the model already does.
The salary doesn't follow the engineer. It follows what's still scarce.
▼read the news

▒░narrator_track░▒
These are not the kind of layoffs that come back next quarter.
For thirty years, Indian IT services hired one and a half million graduates a year because the demand for routine code kept growing. The cities downstream were built on that demand. Whitefield. Hinjewadi. HITEC City. EMIs, college admissions, marriage markets — all of it.
The demand just stopped existing. The cities will catch up to that, on a delay.
> tail -f news.live · path_a.collapse
// headlines from the last 9 months — chronological. each entry verified.
[FEED_01]·2025-07-28·DECCAN HERALD01 / 05
TCS to lay off 12,000 employees amid AI shiftsrc: [7]
[FEED_02]·2025-08-04·CNBC02 / 05
Why India's IT sector is shedding jobssrc: [6]
[FEED_03]·2025-10-12·STORYBOARD18 · EY ANALYSIS03 / 05
Entry-level IT jobs shrink 20-25% as AI reshapes hiringsrc: [5]
[FEED_04]·2026-01-22·RAYSOLUTE WORKFORCE INTEL04 / 05
TCS · Wipro · HCL · Tech-M added just 3,910 staff in a yearsrc: [4]
[FEED_05]·2026-04-08·NASSCOM TECH WORKFORCE FY2605 / 05
Net hiring drops 86% — FY22 600K to FY26 140Ksrc: [1]
▼the other side

> query: path_b.ai_specialist
// live listings · indian market · all verifiable via company career pages.
$ grep -rE "ML Engineer|AI Specialist" jobs.live | head -8
──────────────────────────────────────────────────────────────────────
razorpay/careers·ML Engineer·BLR · hybrid
₹38-55 LPA
cred/jobs·Senior AI Engineer·BLR · onsite
₹52-72 LPA
atlan/openings·ML Platform Lead·remote-IN
₹65-90 LPA
together.ai/in·Research Engineer·BLR · onsite
₹70+ LPA
sarvam.ai/team·AI Specialist·BLR · onsite
₹42-60 LPA
ola.electric/careers·Senior ML·BLR · hybrid
₹48-68 LPA
zerodha/jobs·ML/AI Lead·BLR · onsite
₹55-80 LPA
krutrim/careers·AI Engineer·BLR · onsite
₹45-65 LPA
──────────────────────────────────────────────────────────────────────
8 results·listing_date ≤ 7d·verified · company career pages
"Specialists in GenAI and LLMs realistically reach 40-60 LPA in 5 years."
src: NASSCOM × Deloitte (2025) [2] · demand:supply 5:1
▒░narrator_track░▒
On the other side of the split: eight Indian startups. About 600 engineers between them.
Each one earns more per engineer than the entire IT services industry did at its peak. That's where the wage premium comes from. It's not compensation for being rare. It's the productivity ratio — what one engineer captures when the model has absorbed the labour of ninety-nine others.
This is concentration, not abundance.
▼do the math

▒░narrator_track░▒
Same $250 billion in revenue. 5 million people in 1995. 2 million projected by 2030.
What 5 million used to do by writing routine code, 2 million now do by building the systems that write routine code.
The other 3 million weren't worse at their jobs. They were aligned with the old engine. The engine just stopped needing them.
> compute: bifurcation_delta · section_06
[LOG]build · trajectory-engine · 20y horizon00 / 20
$ npm run bifurcation_delta.compute
 
> rehearsal-research@1.0.0 bifurcation_delta.compute
> tsx ./lib/trajectory-engine.ts --paths=both --years=20
 
[14:08:01]  loading: traditional_path.salaries  ........ ok (21y)
[14:08:01]  loading: ai_path.salaries           ........ ok (21y)
[14:08:01]  inflation_assumption: 5.0% · constant
[14:08:01]  discount_rate: 0.0% · nominal-only
[14:08:02]  computing: cumulative_earnings(traditional)
            → ₹ 1,35,80,000
[14:08:02]  computing: cumulative_earnings(ai_path)
            → ₹ 7,01,00,000
[14:08:02]  delta = ai_path - traditional
            → ₹ 5,65,20,000
[14:08:02]  formatting: ₹5.65 Cr
 
[14:08:02]  ✓ build successful · 0 warnings · 0 errors
 
> next: render hero_number()
● computingelapsed: 00:00:00s
▼the pivot

> what.most_analysis_gets_wrong()
> assumption.deprecated = "experience is protection"
> rendering: binary....
EXPERIENCE
IS A LIABILITY
> first_wave.cohort = "mid_management_12-15yr_xp"
> the_moat.role_in_new_system = "the_wall"
> confidence = 0.81
src: NASSCOM FY26 [1] · Deccan Herald [7] · RAYSolute [4]
▒░narrator_track░▒
Here is what most of the analysis is missing. The first wave of layoffs at TCS, Wipro, HCL and Tech Mahindra wasn't fresh graduates. It was mid-management. Engineers with 12-15 years of experience.
For thirty years, that experience was a moat. The senior person knew which client emails to ignore, which APIs never to touch, which escalations to skip. The judgment of when to do nothing was the seniority.
Then AI made it cheap to explore everything. The same judgment that earned the seniority now hides the options the model can surface.
Experience used to be the moat. It is now the wall.

▒░narrator_track░▒
If experience no longer protects you, what does?
Two things. The first is workflow thinking . Single-task execution is now free — drafting, summarising, classifying. What is still expensive is seeing the whole picture: who feeds in, who approves, what gets remembered. That picture is the new scarcity.
The second is proof of work . Credentials worked because the artifacts they required — a thesis, a code submission, a presentation — were expensive to fake. AI made them cheap. The proof itself — open-source contributions, shipped systems, public case studies — is the new credential.
> recategorise: evidence · certificate → artifact
[FRAME_01]· signal_stack · what counts as evidence×
OLD_SIGNAL_STACK
certificate-based · decaying
[ CV ]
↓
[ DEGREE ]
↓
[ INTERVIEW ]
↓
[ JOB ]
NEW_SIGNAL_STACK
artifact-based · ascending
[ PORTFOLIO ]
↓
[ SHIPPED WORK ]
↓
[ OPEN-SOURCE TRACE ]
↓
[ JOB ]
evidence.recategorise()certificate  →  artifact· 2024-2030
frame01 / 03
frame: Rehearsal Research · 2026-04 → 2026-05
▼measure yours

> what_to_actually_do()
[STUDENT]
// not which company to join. what to ship.

  ship_in_public()
  contribute_to_open_source()
  document_the_reasoning()
[MID_CAREER]
// not which tool to learn. which workflow to redesign.

  identify_costliest_workflow()
  redesign_around_ai()
  document_before_and_after()
[LEADER]
// not which employees to keep. which workflows to protect.

  audit_workflows_for_ai_displacement()
  protect_judgment_accumulating_workflows()
  reward_documentation_over_attendance()
> not_promise. proof.
▒░narrator_track░▒
So what does that look like, in practice?
If you're a student: stop optimising your CV. Start shipping things online. The first stranger who reads your work will teach you more than four years of feedback from teachers and parents combined.
If you're mid-career: the question isn't which tool to learn. It's which workflow in your week costs the most human hours. Redesign it around AI. Document before and after. Ship the redesign.
If you're a leader: some workflows accumulate judgment AI cannot replicate. Others repeat work AI now does cheaper. Protect the first. Reskill people out of the second.
Not promise. Proof.
▼take the audit

[AUDIT_TERMINAL]proof_of_work · 5_questions
booting
▼see sources

The Bifurcation: What Generative AI Is Doing to Indian IT Jobs

Sources

WHAT ISGENERATIVE AIDOING TO INDIANIT?

TCS to lay off 12,000 employees amid AI shift

Why India's IT sector is shedding jobs

Entry-level IT jobs shrink 20-25% as AI reshapes hiring

TCS · Wipro · HCL · Tech-M added just 3,910 staff in a year

Net hiring drops 86% — FY22 600K to FY26 140K

WHAT IS
GENERATIVE AI
DOING TO INDIAN
IT?