Larsen & Toubro Group (L&T) has extended Microsoft 365 Copilot to roughly 140,000 of its employees, turning the rollout into India’s largest corporate workplace AI deployment. But the real story is not the license count—it’s the two purpose-built agents that are now handling everyday HR, IT, and sales tasks with measurable speed.

What’s Actually Happening on the Ground

The expansion is being driven by LTM, formerly LTIMindtree, which already had Copilot deployed among its own 71,000 employees. According to a company announcement reported by ScanX on July 15, the new phase brings the larger L&T Group—spanning technology services, engineering, construction, manufacturing, energy, and financial services—onto the platform.

Two homegrown Copilot agents are doing the heavy lifting:

  • RAIma is an HR and employee-support assistant built with Microsoft Copilot Studio. It fields Leave requests, attendance corrections, office seat bookings, service tickets, and even insurance info searches—all through natural-language chats in Teams, the Copilot interface, or LTM’s enterprise portal via both text and voice. Microsoft’s own April 2026 case study pegged RAIma’s active user base at over 78,000 monthly. LTM now reports a 70% improvement in IT and HR query resolution time, plus a 15% increase in employee engagement and HR productivity. Those are company-supplied figures; no independent audit methodology has been published.
  • A.S.K. (Agent for Stories and Knowledge) grounds sales materials in reality. Tied into more than 40 internal SharePoint repositories, it lets sales and presales teams retrieve relevant case studies, pitch decks, and proposal templates, and even generate first drafts based on approved content. Over 1,000 employees currently have access, and the agent respects existing file permissions when returning results—a crucial distinction from an ungoverned internal chatbot.

More than 23,000 developers now use AI coding assistants daily, backed by 1,300 AI experts and enablement leads. And the foundation for all this: LTM says it has already trained over 90% of its own workforce in generative AI.

What This Means for You

For the average employee at a large firm, RAIma’s model points to a future where you never navigate three different portals to book leave or check a policy. The convenience is real—but so is the need to understand what systems the agent touches and whether it can screw up. If your leave balance is wrong because the backend integration missed an update, that’s still your problem.

For IT administrators and decision-makers, the lessons are starker. A.S.K. shows that the most valuable Copilot agents are the ones that turn existing SharePoint libraries into searchable, reusable knowledge bases. But that also means every past permission mistake, every “Everyone except external users” link, and every duplicate document suddenly becomes discoverable via a simple chat query. Before enabling semantic search, Microsoft 365 admins must:
- Audit SharePoint permissions and remove overexposed content.
- Enforce version control and retention labels so employees don’t pull outdated presentations.
- Clean up abandoned Teams workspaces that might contain sensitive files.

Agent governance needs equal attention. RAIma and A.S.K. wouldn’t fly if nobody knew who owned them, what connectors they used, or how to update them when backend systems changed. In a 140,000-user deployment, agent sprawl—unvetted Copilot Studio creations popping up in business units—could lead to overlapping tools, stale data access, and compliance headaches. Administrators need a living inventory of production agents, each with a named business and technical owner, usage monitoring, and a sunset clause.

For developers, the fact that 23,000 coders are now daily Copilot users amplifies a different set of risks. Generated code can introduce insecure dependencies, replicate flawed patterns, or pass superficial testing while failing in production. Organizations need clear policies on which source code can be submitted as AI context, mandatory peer reviews for AI-generated pull requests, automated security scans, and unambiguous accountability—the developer, not the assistant, owns the finished code.

How We Got Here

LTM’s journey didn’t start with a 140,000-seat megadeal. The company had already achieved 71,000 Copilot users internally by fiscal 2026 and, according to a Microsoft customer story, took a systematic approach: train first, then build agents for specific workflows.

RAIma emerged from a recognition that routine queries—password resets, policy lookups, leave applications—were swallowing support staff’s time. By connecting Copilot to existing HR and IT systems, LTM turned the bot into a conversational front-end that could authenticate employees and act on their behalf within the rules of each backend system. A.S.K. tackled a classic enterprise pain point: valuable sales collateral sitting in dozens of SharePoint sites that nobody can find when a proposal deadline looms.

Both agents reflect a broader industry shift away from generic AI chat toward “grounded” agents that operate on a company’s own data with fine-grained access controls. Microsoft has been pushing this model hard with Copilot Studio, and L&T’s deployment is one of the largest real-world validations yet.

What to Do Now

If your organization is contemplating a similar Copilot expansion, here’s a practical checklist distilled from L&T’s experience:

  1. Clean your house before inviting guests. Run a SharePoint permissions audit. Identify and remediate links shared with “Everyone,” outdated confidential documents, and abandoned project sites.
  2. Define agent ownership. Every production agent needs a business sponsor (who decides what it does) and a technical owner (who maintains connectors and monitors performance). No exceptions.
  3. Set developer guardrails. Document which codebases can be used with AI, require manual review of AI-generated commits for safety-critical applications, and run static analysis on all new code regardless of origin.
  4. Train users, don’t just hand out licenses. LTM reached 90% workforce AI training before scaling agents. Users need to understand what the tool can and can’t do, especially when answers impact HR decisions or customer commitments.
  5. Plan for agent lifecycle. Define when an agent graduates from experiment to production, how usage is tracked, and when it gets retired. A bot that once booked seats might linger after the office layout changes, giving employees wrong information.

Outlook

The next test for L&T Group will be scaling these agents beyond LTM’s technology-focused workforce into engineering, manufacturing, and financial services subsidiaries. In those contexts, data classification rules vary widely, and a misconfigured agent could expose engineering drawings or customer financials. Will the same permission model that works for internal sales decks suffice for infrastructure project documents? That’s the question.

Microsoft will be watching, too. If L&T can demonstrate sustained value across such a diverse conglomerate, it will become a powerful reference for Copilot’s enterprise appeal. But the real story for the rest of us isn’t the 140,000-user number—it’s that the most successful AI deployments might just be the ones that make a boring HR query 70% faster.