Infosys added about 5,000 employees in FY2026 — and still terminated over 300 trainees who couldn’t clear internal assessments in three attempts, with more rounds since. Moving from Infosys to AI engineer roles takes a realistic 6–9 months of preparation, through two routes: internal mobility into AI-tagged work, or an external switch to product companies and GCCs. Here’s both, honestly.

Here’s a strange pair of facts to hold in your head at the same time.

Infosys grew. Headcount went from 323,578 to 328,594 over the year, up around five thousand people. And in the same stretch, more than 300 freshers were let go from the Mysuru campus for one reason: they failed an internal assessment three times. Not a business downturn. Not a project ramping down. A test.

That combination tells you more about your future than any layoff headline does. TCS’s story this year was about releasing senior people. The Infosys story is different and, I’d argue, more important: the company is growing, and it still removes people who can’t clear assessments. The filter isn’t headcount. The filter is skills, checked on a schedule, with limited attempts.

If that’s the culture at your employer, the question isn’t whether you’ll be assessed on AI skills. It’s whether you’ll be ready when it happens — and whether you want your ceiling to be an Infosys assessment or a product-company offer.

I run an AI training company and I used to work on Google’s AI Platform team, so I have an obvious interest in your answer. Read this with that in mind; I’ve cited sources throughout so you can verify rather than trust.

Assessment culture arrived at Infosys first

The mechanics are worth knowing even if you’re years past training. Freshers sign contracts that require clearing foundational assessments within three attempts. Fail, and the options presented have been a move to the BPM division or an exit with a month’s ex-gratia. The terminations drew enough heat that the Labour Ministry ordered a probe and Infosys postponed a later assessment round — but the model itself hasn’t changed, and there’s no sign it will.

Now connect that to what’s happening at your level. Appraisals increasingly reference skill tags. Project allocations quietly check certification status. The same logic that removed those trainees — prove the skill or lose the seat — moves up the pyramid every year, because the repetitive delivery work that used to absorb everyone is exactly what’s being automated across the sector.

You can resent that or use it. This guide is about using it.

“AI-aware” is not “AI-deployable” — and Infosys proves it

Infosys says it has built a community of more than 250,000 “AI-aware” employees. You may be one of them. You did the Lex modules, maybe picked up an internal certificate or two, possibly touched a Topaz engagement.

Be honest with yourself about what that means, because a hiring manager at a product company will be. “AI-aware” is a literacy tier. It means you can talk about prompts and models sensibly in a meeting. It does not mean you’ve chunked a messy document corpus, debugged why retrieval returns garbage for 15% of queries, or defended an evaluation metric in a design review. Those are the things AI engineer interviews actually test — I’ve written about the full skill bar in the TCS switching guide, and it applies here unchanged: Python fluency, RAG systems that survive real data, healthy suspicion about agents, evaluation above all, and enough MLOps to ship.

Lex certificates have real value inside Infosys — they feed internal mobility, which is route one below. Outside, they’re a line on a resume that every other services applicant also has. What separates you outside is built projects with visible code and a log of what broke. There is no shortcut around that, from Infosys or anywhere else.

Route one: move internally before you move externally

The under-used option. Infosys is genuinely winning AI-tagged work — Topaz engagements need people, and staffing them from inside is cheaper than hiring. If you’re at JL4 or JL5 with decent internal standing, a deliberate campaign works better than waiting for allocation luck: finish the relevant Lex certifications fast (they gate the conversation, not the job), then directly message delivery managers on AI-tagged accounts — internal mobility at Infosys still runs on who knows you’re looking. Six months on a real Topaz project gives you production stories for external interviews later, at zero risk, on full salary.

Treat it as a bridge. Internal AI work at a services company is mostly integration — valuable experience, but the compensation ceiling stays a services ceiling.

Route two: the external switch

Product companies and GCCs don’t care about your JL band. They care what you can demonstrate. The preparation is the same 6–9-month arc I laid out in detail in the TCS guide — Python until it’s boring, one genuinely messy RAG project, then evaluation and deployment — so I won’t repeat it here. The Infosys-specific adjustments:

  • Time it around your appraisal and the 90-day notice. Start preparation at least two quarters before you intend to resign. Use the notice period for interviews, never for fundamentals.
  • Use the bench properly if you’re on it. Bench time at Infosys is study time with salary. The trainees who got terminated had no warning system; you do. Don’t waste it on anxiety-scrolling.
  • Mine your project for domain stories. A banking-domain Infosys engineer who builds a compliance-document RAG system has an interview narrative a fresh graduate cannot match. Your “boring” domain knowledge is a multiplier — pick portfolio projects that use it.

If you were let go after an assessment

Some readers are here because of the terminations, so let’s handle the interview question directly. Do not hide it and do not over-explain it. One calm sentence: “Infosys requires clearing internal assessments in three attempts during training; I didn’t clear mine, and it showed me I needed a fundamentally different preparation approach — here’s what I built since.” Then show the projects. I’ve watched interviewers respect that answer, because it demonstrates the exact thing the failure suggested was missing: the ability to diagnose a gap and close it. A GitHub with three documented projects erases a training-period stumble almost completely. What doesn’t erase it is a gap year explained with “I was applying.”

The money, briefly and honestly

Same discipline as always: I won’t promise you a number. Current listings on AmbitionBox and Glassdoor show applied-AI roles at product companies and GCCs paying meaningful multiples of services packages at equal experience, with GenAI and RAG skills carrying a well-documented premium over generic ML roles. The people earning those numbers completed the preparation; plenty who started casually didn’t. Any course that quotes you an average outcome without showing how many people it’s averaged over — ask for the denominator. Ours included; our fee terms are public precisely so you can do that math.

Who should wait

If you can’t hold eight hours a week for six months, fix your schedule before your skills. If you’re mid-appraisal-cycle with a promotion realistically weeks away, collect it first — the title helps externally. And if you haven’t written code in years, spend two months purely programming before touching anything with “AI” in the name. A rushed switch fails in interviews, and interview failures compound psychologically. Slow is fine. Stalled is the only real failure.

FAQ

Do Infosys Lex certifications count outside Infosys?
As resume lines, marginally — every services applicant has them. As preparation, somewhat. External interviews are won by demonstrable projects, not internal certificates. Use Lex to unlock internal AI work; use projects to unlock external offers.

Can I switch while on bench?
Yes — bench is the best preparation window you’ll get: full salary, low workload. Follow the monthly plan in the TCS guide and route one and two simultaneously.

Does my JL band matter for an external switch?
Product companies map you by demonstrated skill and years of hands-on work, not JL bands. A JL4 with three shipped AI projects outcompetes a JL6 with none.

Is the internal Topaz route or the external route better?
Internal is lower risk and slower money; external is the opposite. The smart sequence for most people: internal AI work first if you can get it within a quarter, external interviews six months later with those production stories.

Prateek Jain is the founder of Nuviq AI and previously worked on Google’s AI Platform. Nuviq runs a 12-week applied AI engineering program with pay-after-placement terms — fee structure published here. If you’re not sure you’re ready, start with a month of route one above and find out for free.