Cognizant is placing more than 18,000 employees behind Google Cloud’s Gemini platform as part of a new unit designed to build AI agents for enterprise clients. The IT services company said the Gemini Frontier Capability Group will focus on four initial areas: modernising legacy systems, improving developer tools, automating business processes and strengthening cyber defence. The unit is part of a broader agreement with Google Cloud. The companies described the arrangement as a substantial joint investment, but they did not disclose financial terms.
The new group is intended to help companies move AI agents from pilot projects into daily use. That distinction matters because many enterprises have tested generative AI tools in controlled settings but have struggled to embed them into core workflows. AI agents are software systems that can carry out multi-step tasks, call on other tools and make decisions within boundaries set by their operators. Moving them into production requires more than a model. It requires data plumbing, security controls, monitoring, employee training and a clear process for handling errors. Cognizant’s unit is positioned at that implementation layer, where strategy, engineering and operations meet.
Cognizant has been building toward this announcement. In April, the company established a Gemini practice, and the new group expands that effort into a larger dedicated organisation. Ravi Kumar S, Cognizant’s chief executive, described the unit as a way for the company’s most advanced engineering talent to focus on putting frontier AI into production for clients. The phrasing reflects the commercial challenge facing IT services firms: clients are no longer asking only for experiments or demonstrations. They want working systems that can be measured, governed and maintained.
A dedicated Gemini unit
The Gemini Frontier Capability Group will sit within Cognizant’s wider relationship with Google Cloud. The two companies have not said how many people will be assigned to specific industries or how the unit will be staffed across countries. They have also not disclosed the size of the joint investment. What is clear is the scale of the commitment. More than 18,000 staff is a significant slice of Cognizant’s workforce, which stood at about 357,000 people at the end of June. Dedicated teams of that size can support large client programmes, but they also create pressure to keep skills current as Gemini models and Google Cloud services change.
The unit will begin with four areas. Legacy system modernisation is among the most labour-intensive. Many large companies still run critical processes on older applications, and they want to use AI without rebuilding everything at once. Developer tools are another priority. AI assistants can help engineers write, review and test code, but enterprises need controls around intellectual property, code quality and security. Business process automation covers tasks such as claims handling, invoice processing and customer service workflows, where agents may be able to route work, retrieve information and take routine actions. Cyber defence is the fourth focus. Security teams are exploring AI for threat detection, alert triage and incident response, though the same technology can also widen the attack surface.
Training at scale
Cognizant said more than 55,000 of its staff have completed a Gemini training workshop with Google Cloud, which it called the largest of its kind so far. The company is also rolling out Gemini to 100,000 of its own employees this year, with plans to reach 200,000. Internal adoption can serve two purposes. It can improve productivity inside Cognizant, and it can give consultants hands-on experience that they can carry into client engagements. The numbers also signal how seriously the company is treating AI fluency as a baseline skill rather than a specialist niche.
Cognizant offered one example of client work. It said it helped an unnamed retailer raise developer productivity by 40% and cut costs by 35% using Gemini. The figures have not been independently checked, so they should be treated as company claims rather than audited outcomes. Still, the example points to the kind of return that IT services providers are trying to sell: faster software delivery and lower operating costs, achieved through AI assistance embedded in existing teams and tools. For enterprise buyers, the hard questions will be how those gains are measured, whether they persist after pilots end and what happens when models change.
A broader partner race
Google has been courting consulting firms and systems integrators to sell its AI. In April, it set up a 750 million dollar partner fund to help them build agents on Google Cloud. In September, Accenture and Google formed a unit of up to 1,000 engineers to install Gemini inside client companies. The strategy is familiar from earlier waves of cloud computing. Platform providers need partners who can translate technical capabilities into industry-specific solutions, handle integration work and manage long-term operations. For Google, a larger partner ecosystem can help Gemini reach enterprises that are not ready to build AI systems on their own.
Cognizant is not tied to a single AI provider. It was also among the first firms to join OpenAI’s Codex partner programme in April. That multi-provider stance is likely to appeal to clients that do not want to depend on one model vendor. It also means Cognizant must maintain expertise across competing platforms while helping clients choose the right tool for each job. In practice, many enterprises are expected to use a mix of models, cloud services and internal systems. The integration layer, rather than any single model, may become the most valuable part of the stack.
Why system integrators matter
The announcement highlights a wider shift in the AI market. The first phase of generative AI was dominated by model launches and public experimentation. The next phase is about deployment: connecting models to company data, building guardrails, measuring results and changing how teams work. System integrators are well placed for that work because they already manage large technology estates, understand regulatory constraints and have relationships with business leaders. They can also absorb the complexity of stitching together software from multiple vendors.
AI agents add a further layer of difficulty. An agent that can take actions needs clear permissions, audit trails and fallback procedures. It must know when to stop and ask for human help. It must be tested against unusual inputs and monitored after launch. In customer service, for example, an agent might resolve a simple request, escalate a complex case and log every step for compliance. In cyber defence, an agent might prioritise alerts, but a wrong action could disrupt operations. These are not purely model problems. They are process, governance and engineering problems, which is why services firms see an opening.
The commercial stakes are high for Cognizant and its peers. Traditional IT services growth has slowed in some markets, and generative AI is both an opportunity and a threat. It can create new work in data preparation, integration and change management. It can also automate tasks that services firms once billed by the hour. By dedicating a large team to Gemini, Cognizant is making a bet that enterprise demand for AI implementation will be deep enough to support specialised capacity. The undisclosed financial terms make it difficult to judge the size of that bet relative to the company’s overall investment plans.
Workforce and competitive pressure
The workforce dimension extends beyond Cognizant. Rivals are making similar moves. Tata Consultancy Services plans to retrain up to 8,900 staff as AI deployment engineers. That figure is smaller than Cognizant’s 18,000 dedicated staff, but it points in the same direction: services companies are reshaping training and career paths around AI delivery. The question for employees is whether these programmes create new roles or simply change the skills required for existing ones. The question for clients is whether large-scale training translates into consistent quality across engagements.
There are also risks in the claims surrounding productivity. AI coding assistants can speed up some tasks, but gains vary by codebase, team experience and the amount of review required. Automation can reduce handling time, but exceptions often consume more effort. Cyber defence tools can generate alerts faster than humans can investigate them. A dedicated unit can help clients navigate those trade-offs, but it cannot remove them. Enterprises will need clear metrics, pilot controls and a realistic view of what agents can do safely.
Cognizant’s move also reflects the growing importance of partnerships in enterprise AI. Google Cloud gains a large implementation partner with global reach. Cognizant gains closer access to Gemini engineering resources, training and co-selling opportunities. Clients gain a single point of contact for strategy, integration and operations, though they should still ask how vendor incentives are managed. The arrangement is not exclusive, and Cognizant’s earlier work with OpenAI suggests it intends to keep options open.
For now, the Gemini Frontier Capability Group is focused on getting AI agents into production across four practical domains. The company has not said which industries will be first or how it will measure success. It has not published client names beyond the unnamed retailer example. Those details will matter as the unit moves from announcement to delivery. The initial signal is one of scale: more than 18,000 people, more than 55,000 trained, and a plan to put Gemini in the hands of 100,000 employees this year, with 200,000 in sight. The broader race among IT services firms is already under way, with TCS planning to retrain up to 8,900 staff as AI deployment engineers.