Quick answer: In FY2026, TCS is releasing around 12,000 employees — about 2% of its 613,000-person workforce — with mid and senior grades hit hardest, and unallocated engineers now reportedly get 35 days on bench before retraining or exit. Moving from TCS to AI engineer roles is genuinely possible in 6–9 months of focused preparation, through two routes: TCS’s own AI Careers assessments, or an external switch to product companies and GCCs. This guide covers both, honestly.
If you work at TCS, this year probably felt different.
Not because of one announcement. Because of what came after it. The company confirmed it will let go of around 12,000 people through FY2026, mostly mid and senior grades. The CEO went out of his way to say the reason was a skill mismatch, not AI taking jobs. And then the detail that actually changed daily life on the ground: reports of a tightened bench policy, where an engineer without a project has roughly 35 days to get redeployed or move on.
Here’s the part worth sitting with. TCS says it has trained over five and a half lakh employees in basic AI skills and more than a lakh in advanced ones. Twelve thousand people are still going. Which tells you something most upskilling ads won’t: a certificate of completion and being deployable on an AI project are two different things. The company itself has effectively admitted that gap.
I spent years on Google’s AI Platform team, and I’ve sat on the hiring side of the table for AI roles. I now run a training company, so read everything below knowing that — I’ve linked sources so you can check my claims rather than trust my incentives. This is the guide I wish someone had written for the engineers messaging me since July.
What actually changed at TCS this year
Three things, and they compound.
First, the layoffs target experience, not freshers. That breaks the old assumption that surviving long enough makes you safe. Senior delivery roles built on coordination, status calls and legacy maintenance are exactly the profile being “realigned.”
Second, the bench went from waiting room to countdown timer. Under the reported policy, 35 days of non-allocation forces a decision. If your skills only match projects that are shrinking, the countdown isn’t really about allocation. It’s about your resume.
Third — and this is the one almost nobody talks about — TCS opened a fast lane for the skills it actually wants. Its AI Careers program invites engineers with hands-on experience in Java, Python, .NET and similar stacks to sit capability-first assessments for AI, data and cloud roles. Read that carefully: your own employer has moved from “years of experience” to “prove it in an assessment.” That’s the direction the whole industry is moving. The assessment culture is coming to you either way; the only choice is whether you prepare for it on your timeline or the bench’s.
You have two exits, not one
Most “switch to AI” advice assumes you have to resign. You don’t, at least not immediately. Play both routes in parallel.
Route 1: The internal move (TCS AI Careers)
If you have two to four years of genuine hands-on coding — not config-and-ticket work, actual code — the internal assessments are the lowest-risk first move. No notice period, no interview gauntlet, no salary-band reset while you learn. The catch is that “capability-first” means precisely that: a Java developer who hasn’t written Python in three years will not pass an AI capability assessment on service-project experience alone. Treat it as a real exam with a syllabus, which the second half of this guide is.
The internal route has a ceiling, though. You’ll likely land on AI-adjacent delivery work first, and pay revisions inside services move the way they always have. Think of it as a paid bridge, not a destination.
Route 2: The external switch (product companies and GCCs)
This is the route with the bigger payoff and the higher bar. Product companies and global capability centres in Bangalore, Pune and Hyderabad are hiring for GenAI roles, and they don’t much care which company badge you carry. They care what you’ve shipped. Which brings us to what these roles actually test — because it’s not what most courses teach.
What “AI engineer” means in hiring rooms right now
Forget the model-building fantasy. In 2026, the majority of open AI engineer roles in India are applied roles: take foundation models that already exist and make them work reliably inside a product. When I interviewed candidates for platform roles, and in every hiring conversation I have now, the same five things come up:
- Python, fluently. Not “familiar with.” Your Java or C# background helps you with structure and systems thinking, but Python is the working language, and interviewers can tell within ten minutes whether you live in it.
- RAG systems that survive contact with real data. Anyone can wire a vector database to an API in a weekend. The interview questions are about chunking choices, retrieval failures and grounding — the messy parts.
- Agents, treated with suspicion. Companies are burning money on agentic systems that fail silently. Engineers who can explain when not to use an agent stand out more than the ones who demo one.
- Evaluation. This is the great filter right now. “How do you know your system got better?” ends more interviews than any coding round. If you learn one unfashionable thing, learn evals.
- Enough MLOps to ship. Docker, monitoring, cost control, versioning. Your services background is a genuine advantage here — you’ve lived production discipline that fresh ML graduates haven’t.
Notice what’s absent: deep learning math, research papers, Kaggle medals. Useful, not gating. The bar is an engineer who can build, measure and maintain — which is closer to what you already do than you think.
A plan that survives a 9-to-7 job
The honest timeline for a working TCS engineer is six to nine months of consistent evenings and weekends, or faster with structured full-pace support. Anyone promising four weekends is selling something. The shape that works:
- Months 1–2: Python until it’s boring. Rebuild something you’ve already built in Java, in Python. Then data handling, APIs, and enough of the ML vocabulary to read job descriptions without googling every second word.
- Months 3–4: Build one real RAG system. Not a tutorial clone — point it at genuinely messy documents (policy PDFs are perfect) and keep a log of what broke and what you changed. That log becomes interview material.
- Months 5–6: Evals and deployment. Put your project behind an API, containerise it, and build a small evaluation harness that scores its answers. Now you have the thing 90% of applicants don’t: evidence.
- Throughout: one visible artifact per month on GitHub. Three imperfect, documented projects beat ten certificates. Time your exit around your appraisal cycle and the 90-day notice period — recruiters at product companies are used to it, but your preparation shouldn’t start when the notice does.
I’ve written separately about the general path in our software engineer to AI engineer guide; the difference for TCS folks is mostly the two-route strategy and the calendar above.
The money, without the fantasy numbers
I won’t quote you a salary you’ll definitely get, because nobody can. Directionally: listings and reported compensation on AmbitionBox and Glassdoor show applied-AI roles at product companies and GCCs paying meaningful multiples of equivalent-experience services packages, and engineers with GenAI, LLM and RAG skills command a 20–40% premium over traditional ML roles in current market analyses. The honest caveat: those numbers describe people who completed the switch, not everyone who attempted it. If a course quotes you an average hike without showing the denominator, ask for the denominator. (That applies to us too — ask.)
Who shouldn’t attempt this yet
Being straight with you: if you can’t protect eight to ten hours a week for six months, fix that first, because a half-done switch shows in interviews. If your coding has fully atrophied — no code at all for 3+ years — spend two months just programming before touching anything AI. And if you’re currently on bench with the clock running, your priority is the internal assessment route first; it buys you time and salary while you build toward the external one.
FAQ
Is 30 or 35 too late to switch from TCS to an AI engineer role?
No. The layoffs skew senior precisely because experience without current skills lost value — but experience with current skills is what GCCs pay for. Your production discipline and domain knowledge transfer; the stack is what needs rebuilding.
Can I prepare during my notice period?
You can polish, but 90 days isn’t enough to start from zero. Begin while employed; use the notice period for interviews, not fundamentals.
Is TCS’s internal AI training enough to get an AI engineer job?
By the company’s own numbers — 5.5 lakh trained, 12,000 still released — completion alone isn’t the bar. Whether internal or external, deployable skills mean built projects, not attendance.
Do I need a master’s degree?
For applied AI engineer roles, no. Hiring has moved to capability assessments and portfolio evidence. Research scientist roles are a different track with different rules.