GenAI engineers in the services world

A role across the services world, every Company Tier together, Accenture left out. Accenture posts more than twice as much GenAI work as the rest of the services world put together, and its GenAI hiring is described with the rest of its hiring on its own.

The short answer

GenAI engineers build software on top of large language models, the assistants, agents and search tools that companies now want inside their own systems. Every client of a services firm is asking for this, and the services world has been hiring for it fastest of any role. Across the whole record GenAI was a middle-sized role, a little below its weight at product companies. But its share of services hiring doubled through 2026, and by July to September 2026 it had caught up with product companies. GenAI openings fell into April to June 2026 with everything else and then rose sharply in July to September 2026, while services hiring overall kept falling. The consulting arms, like Deloitte, PwC and McKinsey, lead, at close to three times their services weight, with the small shops, like AgileEngine and GeekyAnts, and the BPO and KPO firms, like TELUS Digital and EXL, also well above theirs. The giants, like IBM, Infosys and Wipro, posted the most early on but at well below their weight, and most of them pulled back in 2026, as Deloitte's GenAI hiring surged. The postings require Python above all, LangChain and LangGraph, and the language-model services of the major clouds. The door leans senior, and the junior openings that existed in the winter of 2025 thinned out through 2026.

Where the work sits

The consulting arms lead. In Company Tier 7, The consulting firms' engineering arms, GenAI is close to three times its services weight, and the consulting arms hold more GenAI openings than any tier but the giants. The engagement a consulting firm now wins most often puts an AI model into a client's process, and its Indian arm builds it. Deloitte and PwC hire the most, with McKinsey, BCG and Sia beside them.

The giants hold the most, at well below their weight.

Company Tier 1, The IT services giants, posted the largest share of GenAI openings across the record, at IBM and Infosys above all, then TCS and Wipro. But GenAI is a smaller part of the giants' hiring than of services as a whole.

The small shops and the BPO and KPO firms are above their weight. In the small shops, GenAI is well above the services rate, and at AgileEngine it is a large part of what the firm posts. Among the BPO and KPO firms, TELUS Digital and EXL hire GenAI engineers to put AI into the customer service and back-office work they run for clients.

About their weight. The build firms, at EPAM above all, the specialist firms, where Gruve, Impetus and Fractal hire it heavily, the mid-sized firms, at Virtusa and Happiest Minds, and the quiet foreign majors, at NTT DATA and CGI, hire GenAI at about the services rate.

Below it. The engineering-services firms, like Cyient and Quest Global, and the infrastructure firms, like Insight and SoftwareOne, hire GenAI at well under their services weight.

What the work is

Most GenAI openings are for building applications on top of language models, assistants, agents and search over a company's own documents. A smaller part is putting AI into existing software, adding a model to a system that already runs, and that part is largest at the giants and the BPO firms. A handful are conversational AI, voice and chat assistants.

The postings require Python above everything else. LangChain is required in many of them, and LangGraph and CrewAI for agents that carry out several steps on their own. The model services named are OpenAI's, Azure OpenAI, Anthropic's Claude, Google's Vertex AI and Amazon Bedrock, with Pinecone and other vector databases for search over documents. FastAPI or Flask serves the model, Docker and Kubernetes run it, and Azure is named slightly more often than AWS. Many postings also require building the screen around the model in React or Angular, and some require training or tuning models with PyTorch, TensorFlow and Hugging Face. Some ask for AI governance, checks on what a model says and does.

The emphasis differs by kind of firm.

The consulting arms lean to Google's Vertex AI, Node.js and React, building whole products for a client.

The quiet foreign majors lean to Microsoft's Semantic Kernel, CrewAI and Pinecone, for agents on a client's Microsoft estate.

The mid-sized firms lean to Azure OpenAI, OpenAI's API and Anthropic's Claude.

The build firms pair GenAI with Java and Angular, putting models inside the large systems they build.

Bengaluru has most of the work, though less of it than for services hiring as a whole. Hyderabad holds a larger share of GenAI than of services overall, much of it at Deloitte.

What is changing

These readings follow the record quarter by quarter, from October to December 2025, the third quarter of the financial year 2025-26, to July to September 2026, the second quarter of 2026-27. A longer record will show whether they last.

GenAI bucked the fall

Its openings slipped into April to June 2026 like everything else, then rose sharply in July to September 2026 while services hiring overall kept falling. Its share of services hiring doubled across the record and by July to September 2026 matched its share at product companies.

The giants pulled back and Deloitte surged

IBM, Infosys and Wipro posted far fewer GenAI jobs through 2026, while Deloitte went from almost none in the winter of 2025 to the largest GenAI employer in services by July to September 2026. TCS came back in that quarter, and Virtusa grew.

Agents took over

LangGraph, the tool for building agents, was required in twice the share of openings in July to September 2026 as in the winter of 2025. MCP, the standard that lets a model use outside tools, went from rare to common from April 2026. Vertex AI, Amazon Bedrock and Azure OpenAI all rose in July to September 2026.

The work became whole products

Building applications on models grew to the great majority of the role, while putting models into existing software shrank, and React was required more often in every quarter of 2026.

Persistently open roles

GenAI jobs are persistently open about as often as services jobs overall. When one is posted again, its first and last postings are typically about three weeks apart, and rarely more than two months.

Among the larger employers, the quiet foreign majors keep GenAI jobs open longest. NTT DATA posted most of its GenAI jobs more than once, and IBM, Infosys and PwC post the same job again more often than the services average. The small shops and the specialist firms rarely do.

Model builders and AI safety are the hardest skills to find. The skills found most often in persistently open GenAI jobs are those of training models, PyTorch, TensorFlow, Keras and Hugging Face, and of keeping models safe, AI ethics and Guardrails AI, along with Cohere's models and Semantic Kernel.

Junior and staff-level GenAI jobs stay open more often than mid-level and senior ones.

The door

The GenAI door leans senior. Senior and staff openings are both more common than the services rate, with staff openings close to twice it, while mid-level openings are fewer. Entry and junior openings are close to the services rate across the record, but they thinned out sharply from April 2026, as the firms that hired juniors posted less.

The consulting arms take the most juniors, PwC above all, and the mid-sized firms, at Happiest Minds, and the BPO firms, at TELUS Digital, take them most readily in proportion.

The giants take juniors mostly at IBM, which posted far fewer GenAI jobs in 2026.

The build firms are the most senior, with senior and staff openings the large majority there.

In one line

GenAI is the services world's growing role, the one that doubled its share while hiring fell around it, and if you build with Python, LangChain and LangGraph on a major cloud, the consulting arms, Deloitte above all, are where the hiring has gone, though most of it now wants experience.

Terms used on this page

Large language model

An AI model that reads and writes text, such as OpenAI's GPT or Anthropic's Claude, on which assistants and agents are built.

Agent

Software built on a language model that carries out several steps on its own, using tools and data.

LangChain, LangGraph and CrewAI

Toolkits for building applications and agents on top of language models.

MCP

The Model Context Protocol, a standard way for a language model to use outside tools and data.

Vertex AI, Azure OpenAI and Amazon Bedrock

Google's, Microsoft's and Amazon's cloud services for using language models.

Semantic Kernel

Microsoft's toolkit for building AI features into applications.

Vector database and Pinecone

A database that finds text by meaning rather than exact words, used for search over a company's documents.

Hugging Face

A library and hub of open AI models.

AI governance and Guardrails AI

Checks and controls on what a model says and does, and a tool for setting them.

Conversational AI

Voice and chat assistants that hold a conversation.

Back to the story

This page reads the openings. The story of the companies is on IT services firms in India.