Microsoft’s product chiefs are tearing up the old playbook. Instead of starting with features and user interfaces, they now begin with the questions software should answer—then build AI agents and pipelines to automate entire workflows. Aparna Chennapragada, the company’s global chief product officer, described this “prompt-first” philosophy in a recent interview, pointing to Indian IT services giant Persistent Systems as a prime example. Persistent used Microsoft’s enterprise AI agents to create a contract management system that slashed negotiation email traffic by up to 95% and cut navigation time by 70%.

This is not just a tweak to the development process. It is a fundamental change in how products are conceived, built, and monetized—one that could reshape the $280 billion Indian tech services industry and cement Microsoft’s role as the backbone of enterprise AI.

From feature roadmaps to prompt-first design

Chennapragada, a former 12-year Google veteran who joined Microsoft two years ago, made it clear that the old way of building software is fading. “The very definition of a product today is built around intelligence,” she told Mint. “The next big shift is the change of the primary interface of a product from voice to natural language. The way of building a product itself has changed—rather than being bracketed into hardware or software, we’re starting with what prompts it should be able to answer. Then, we decide the rest.”

This agent-first approach means that instead of designing a UI and then layering on AI, product teams begin with the desired outcomes—faster approvals, fewer manual handoffs, smarter search—and build the agents, APIs, and governance layers around those goals. The model is already bearing fruit in the Indian IT sector, where firms are turning client engagements into repeatable, packaged solutions powered by Microsoft’s cloud and AI stack.

The Persistent Systems case study: ContractAssIst in action

The most concrete example comes from Persistent Systems. In January 2025, the company launched ContractAssIst, an AI-driven contract management solution built on Azure AI, Microsoft 365 Copilot, Teams, and the Azure OpenAI service. According to both Persistent’s own announcements and Microsoft’s customer blog, the results were dramatic: a 95% reduction in email traffic during contract negotiations, a 70% cut in navigation and negotiation time, and an initial Copilot deployment to nearly 2,000 Persistent employees before wider rollout.

ContractAssIst does more than automate simple tasks. It combines a centralized Teams dashboard that aggregates contract status, deadlines, and approvals; a conversational agent that can answer natural-language queries about clauses, flag unusual terms, and prepare approval summaries; and automated approval routing with templated responses. These are not chatbots but orchestrated agents that span document search, knowledge grounding, rule-based workflows, and fine‑tuned generative models—deployed with observability and governance baked in.

“We worked closely with Persistent on how to reduce the time gaps within its contract management system,” Chennapragada said. “Our AI agents shortened manual approval times. This sort of prototyping is where AI is dramatically reducing the time and resources consumed for a business.”

How AI agents are stitching together Microsoft’s ecosystem

Microsoft’s strategy hinges on making Copilot the connective tissue across Windows, Microsoft 365, Teams, and Azure. The company sells Copilot both as a per‑user productivity assistant and as a platform for building domain‑specific agents. Tools like Copilot Studio, Microsoft 365 Copilot Chat, and Azure AI Foundry allow service firms to compose agent templates for common workflows—contracts, HR onboarding, RFP responses—and productize them quickly. These agents live where users already work: inside Teams, Outlook, and Office apps.

“People spend a lot of time in Teams, Outlook and the Office apps. We’re trying to get the best of our AI in those places where people already are,” Chennapragada said. “Our next goal is to get the M365 Copilot AI assistant to everyone for various tasks. Finally, we’re also looking to build a team of agents that can be present with an employee at work from day one.”

On the hardware side, Microsoft has introduced Copilot+ PCs, Windows machines with dedicated neural processing units (NPUs) capable of 40+ trillion operations per second. These devices enable on‑device AI features such as Recall (screen activity snapshotting for later retrieval), Cocreator in Paint, Restyle Image, and Windows Studio Effects—experiences that blend on‑device speed with cloud intelligence. For enterprises, Copilot+ PCs promise latency reductions and data‑privacy advantages, but they also raise new questions about what data is captured and how it is stored.

Market forces and competitive pressures

Microsoft’s AI pivot comes amid a booming market for AI services. India’s IT services sector is expected to cross $300 billion in FY2026, and major players like TCS and Infosys are deepening their ties with Microsoft to embed agentic automation alongside human workforces. Investors have taken notice: Microsoft’s stock price surged on a July 30, 2025, earnings beat, briefly pushing its market capitalization above $4 trillion, joining Nvidia in the ultra‑large‑cap club. As of the interview, the company’s valuation stood at $3.77 trillion, making it the world’s second‑largest company.

Yet the competitive landscape is intense. Google, under Sundar Pichai, went all‑in on AI for consumers and enterprises, and its stock has surged nearly 35% in the past year—far outpacing Microsoft’s year‑on‑year growth on Nasdaq. Analysts at Morgan Stanley remain bullish, however, with Keith Weiss noting in July 2025 that Microsoft maintains the lead in enterprise AI spending intentions. “Microsoft has been able to maintain its leadership position in terms of core spending intentions and specifically related to capturing share of generative AI spend,” Weiss wrote. “CIOs expect Microsoft to see the highest growth in AI spending from companies among all the Big Tech firms.”

Chennapragada also addressed the company’s relationship with OpenAI, which has faced scrutiny and reports of strategic differences. “We continue to retain a very strong partnership with OpenAI,” she said. “We’re also working on our own foundational models for consumer‑facing Copilot operations. Azure remains a key infrastructure provider for all AI applications.”

Critical look: the promises and pitfalls

For all the optimism, several caveats demand attention. The efficiency gains reported by Persistent—95% fewer emails, 70% faster navigation—come from vendor case studies and pilot programs. These numbers are impressive but require independent validation. Short‑term metrics in controlled settings may not translate equally across different contract types, legal environments, or user populations. Enterprises should treat such claims as indicative rather than universally reproducible.

Generative AI still hallucinates, and in legal or contractual contexts, a fabricated clause summary can be disastrous. Human‑in‑the‑loop gates, provenance tracking, and rigorous test suites are non‑negotiable. Privacy is another flashpoint: features like Recall, which captures screen snapshots for later retrieval, accelerate productivity but raise genuine data‑leakage concerns. Security teams must scrutinize what is captured, how it is encrypted, and who can access it—particularly in regulated sectors. Microsoft documents these safeguards, but independent security reviews are advisable.

Vendor lock‑in is a real risk. Bundling Copilot and agent capabilities into Office, Teams, and Azure—and coupling on‑device features to Copilot+ hardware—can increase switching costs. CIOs must analyze total cost of ownership, data egress fees, and licensing terms before committing workloads. Regulatory uncertainty around the Microsoft–OpenAI partnership adds another layer: enterprises that process regulated data need contractual clarity on data usage, model training, and hosting boundaries.

Finally, agentic automation will reshape workforces. While productivity soars, roles will be redesigned, and bench rationalization is already visible in parts of India’s services sector. NASSCOM and industry reports note hiring slowdowns; firms must pair deployment with reskilling and redeployment strategies to avoid social backlash and talent attrition.

What Indian IT services leaders should do now

Enterprises and CIOs evaluating agentic products should treat the shift as strategic change management, not a simple procurement decision. A pragmatic approach includes:

  • Pilot narrowly: Select a single high‑volume, well‑scoped workflow (contracts, procurement approvals, RFP responses) and measure baseline KPIs—time‑to‑approve, number of handoffs, email volume—before and after.
  • Define governance: Require agents to produce source citations and change logs, implement approval gates for legal/regulatory outputs, and lock down data access with tenant‑only models or on‑premise/edge inference where feasible.
  • Plan for humans: Ensure every high‑risk output is reviewed by a named human owner until the model is demonstrably reliable.
  • Negotiate for portability: Seek contractual rights to export agent definitions and to move workloads between clouds. Model cost per transaction and worst‑case egress scenarios.
  • Invest in people: Reassign staff toward oversight, prompt engineering, and product stewardship. Create internal certification for agents and governance roles.
  • Audit and iterate: Record service metrics and model drift indicators, and maintain a regular audit cadence for privacy, security, and accuracy.

These steps create a defensible adoption path that balances rapid productization with the controls that regulated enterprises need.

The road ahead

Agent‑first development is more than a new toolset; it is a new operating model for product creation. Microsoft’s stack—Copilot, Copilot Studio, Azure AI Foundry and Copilot+ PCs—gives Indian services firms both a fast runway and a set of constraints. Persistent Systems’ ContractAssIst shows the upside: tangible efficiency gains packaged into a product that scales. But every firm that embraces agentic workflows must pair speed with rigorous validation, security controls, and a concrete reskilling plan.

For IT leaders, the message is clear: treat early wins as prototypes, insist on independent measurement of vendor claims, build governance from day one, and negotiate contracts that preserve data provenance and portability. The competitive prize is real—productization of services into AI‑native offerings will reshape margins and client relationships—but so are the operational and regulatory challenges that come with this new class of software.