The Claude Reckoning, AI, Effort Compression, and the Unfinished Reinvention of India’s IT Services Industry
On a seemingly ordinary day in early 2026, Anthropic released Claude Cowork, an AI “digital coworker” equipped with 11 enterprise automation plug-ins. The announcement, buried in the endless scroll of AI product releases, triggered an immediate and violent reassessment of the Indian IT services industry. Analyst notes that had been cautiously optimistic about the sector’s ability to adapt to generative AI were hastily revised. Predictions of revenue deflation of up to 40 per cent began to circulate. The phrase “agentic AI”—machines that do not merely assist humans but autonomously execute multi-step workflows—entered the lexicon of every technology analyst, portfolio manager, and chief information officer.
This was not the first AI scare the industry had weathered. The launch of ChatGPT in late 2022 had prompted similar, if less intense, soul-searching. Each successive wave of large language model capability had been met with assurances that AI would augment rather than replace, that the industry’s deep domain expertise and client relationships were durable moats, that the complexity of enterprise environments would resist easy automation. These assurances were not wrong, but they were becoming increasingly difficult to sustain.
Claude Cowork was different. Its 11 plug-ins targeted core enterprise functions—customer support, finance, data analytics, marketing, product development, legal compliance—not peripheral tasks. It demonstrated the capacity to execute multi-step workflows autonomously, not merely to respond to prompts. It was designed to integrate with existing enterprise systems, not to operate in isolation. It was, in short, the first credible demonstration of a technology that could systematically displace the routine, repetitive, rules-based work that has long been the foundation of the IT services industry’s business model.
The authors of the accompanying analysis, Nitin Bhatt and Mahesh Makhija of EY India, do not dismiss these fears. They acknowledge that “the next 12-24 months will be disruptive.” They estimate that AI-driven “effort compression” could reduce labour hours by up to 40 per cent in build, support, and analytics work. They warn that pricing resets will follow, with a lag of one contract cycle, as clients benchmark incumbent providers against AI-enabled delivery costs. They predict that firms heavily reliant on time-and-materials billing will “feel the impact first” as routine work is displaced and purchased hours fall.
Yet their analysis is not a eulogy. It is a strategic roadmap for survival and reinvention. It identifies the enduring sources of competitive advantage that technology service providers possess: the capacity to integrate AI tools into complex, fragmented enterprise environments; the expertise to govern and audit autonomous systems; the relationships and trust that enable long-term partnerships. It prescribes a decisive shift in delivery and commercial models: from effort-based billing to outcome-based pricing, from labour arbitrage to AI-assisted productivity, from reactive service provision to proactive capability building. And it issues a stark warning to incumbents: “Those who do not adapt will lose. And lose fast.”
The Complexity Imperative: Why Integration Matters More Than Automation
The first reality that the authors invoke against the “end-of-outsourcing” narrative is the enduring complexity of enterprise environments. Large organisations do not run on pristine, unified, API-accessible data platforms. They run on fragmented, heterogeneous, often antique systems: mainframes running COBOL code written in the 1980s, customer relationship management databases that have been customised beyond recognition, supply chain platforms that communicate with each other through bespoke interfaces and manual workarounds. This is not a bug; it is a feature of organisational evolution. Systems accumulate, layer upon layer, as companies grow, merge, and adapt to changing markets.
AI tools, no matter how sophisticated, cannot simply plug into this environment and begin delivering value. They require integration, orchestration, and governance. They need to be connected to the right data sources, configured to respect organisational policies, monitored for errors and biases, and updated as systems change. These are not one-time tasks but ongoing responsibilities. They require deep knowledge of both the technical architecture and the business context of each client.
This is the capability that technology service providers bring. They are not merely vendors of labour; they are architects of integration. They understand how their clients’ systems work, where the bodies are buried, what the workarounds are. They have invested decades in building the relationships, processes, and intellectual property that enable them to operate effectively in complex environments. This is not a moat that will be breached overnight.
The authors’ observation that “whenever software creation becomes easier, demand accelerates” is supported by decades of evidence. The transition from assembly language to high-level programming languages did not eliminate the need for programmers; it expanded the scope of what could be programmed. The open source movement did not commoditise software development; it enabled the creation of vastly more complex and capable systems. The cloud did not render IT services obsolete; it created new categories of work around migration, optimisation, and security.
AI will follow the same pattern. By lowering the cost and complexity of building certain types of software, it will expand the universe of problems that can be economically addressed with code. Enterprises that could not justify the investment in custom workflow automation will now find it feasible. Processes that were too marginal to warrant dedicated development teams will now be automated. The pie will grow, even as the share of labour in each slice diminishes.
The Autonomy Gap: Why Agents Are Not Ready for Prime Time
The second reality is that fully autonomous agents at enterprise scale are still years away. Claude Cowork is impressive, but it is not ready to be turned loose on a production ERP system without human supervision. Agentic workflows still “loop, stall or misfire” when confronted with ambiguous inputs, unexpected edge cases, or conflicting instructions. They require robust guardrails—constraints on their scope of action, approvals for high-risk operations, fallback procedures when they fail.
More fundamentally, the security architectures required to support autonomous agents do not yet exist. To be useful, agents need access to enterprise systems—customer databases, financial records, source code repositories, legal documents. Granting this access creates enormous risk. A compromised agent could exfiltrate sensitive data, execute unauthorised transactions, or introduce vulnerabilities into production systems. Current security models, built around human users with granular permissions and multi-factor authentication, are not designed to accommodate machine actors with broad, programmatic access.
Enterprise leaders, particularly in regulated industries such as banking, insurance, and healthcare, are deeply reluctant to approve such access. The potential benefits of automation are outweighed, in their calculus, by the potential costs of a security breach or compliance violation. This reluctance will not be overcome by better marketing or more persuasive demos. It will require the development of new security architectures, new regulatory frameworks, and new trust mechanisms. This is a multi-year, not multi-month, timeline.
The authors’ assessment—that “we are still years away from replacing developers in complex environments”—is not complacency; it is realism. It recognises that the transition to agentic AI will be gradual, contested, and uneven. It will proceed at different paces in different industries, organisations, and functions. It will require not only technological advancement but also organisational learning, regulatory adaptation, and cultural change.
This timeline creates a window of opportunity for incumbent service providers. They have time to retool their delivery models, upskill their workforces, and develop the governance and integration capabilities that will be essential in the agentic era. They have time to experiment with new commercial models, to pilot AI-enabled delivery approaches with willing clients, to learn what works and what does not. They have time to become the trusted partners that enterprises will turn to when they are ready to deploy autonomous agents at scale.
The Commercial Reckoning: From Effort to Outcome
The most painful and unavoidable dimension of the AI transition is the restructuring of the commercial model. The Indian IT services industry has grown wealthy on time-and-materials billing. Clients pay for hours worked; providers capture the margin between the billing rate and the cost of employment. This model aligns incentives poorly—providers profit from inefficiency—but it has been remarkably durable. It survives because clients lack the information and bargaining power to enforce strict linkage between payments and value delivered.
AI will shatter this equilibrium. When a task that required 100 hours of human labour can be completed in 60 hours with AI assistance, clients will demand to share the productivity gain. When a competitor offers to perform the same work for 40 per cent less, citing their own AI-enabled efficiencies, incumbent providers will be forced to match their pricing or lose the business.
The authors’ estimate of up to 40 per cent effort compression in build, support, and analytics work is not a prediction; it is already being realised in early pilots. This is not a future scenario; it is a present reality. The only question is how quickly pricing will adjust to reflect the new cost structure.
The impact will vary by contract type. Firms with high exposure to time-and-materials billing will feel the pressure first and most acutely. As routine work is displaced by AI tools, the hours purchased by clients will decline. Revenue will fall, and margins will be squeezed as fixed costs are spread over a smaller base.
Firms with a more balanced contract mix—a portfolio of time-and-materials, fixed-price, and outcome-based engagements—will see a more gradual adjustment. Volumes will soften; price resets will occur at renewal rather than immediately. But the direction of travel is unmistakable. Even firms with minimal time-and-materials exposure will face tougher renegotiations as clients benchmark their fixed-price bids against the cost of AI-enabled delivery.
The authors’ prescription is clear: move from effort-based billing to outcome-based pricing. This is not merely a change in invoicing methodology; it is a fundamental reorientation of the provider-client relationship. Instead of selling hours, providers must sell results: reduced customer churn, faster time-to-market, lower operational costs, higher code quality. Instead of being compensated for inputs, they must be compensated for impact.
This transition is extraordinarily difficult. It requires providers to assume risk that was previously borne by clients. It requires them to develop the analytical capabilities to measure and attribute outcomes. It requires them to redesign their delivery processes around efficiency rather than effort. It requires them to cannibalise their own revenue streams before competitors do it for them.
But it is also the only viable path to long-term survival. Providers that cling to the time-and-materials model will be systematically undercut by more efficient competitors. Their revenue will erode; their margins will collapse; their talent will depart for more dynamic organisations. They will not fail suddenly, in a dramatic bankruptcy or acquisition. They will fail slowly, over years, as each renewal brings another round of price concessions and each quarter brings another round of revenue decline.
The Reinvention Imperative: Productising AI Governance
The authors identify a significant opportunity that lies beyond the defensive imperative of commercial restructuring: the productisation of AI governance.
As enterprises embed AI deeper into sensitive workflows, they will confront a host of new risks and challenges. How do they ensure that autonomous agents comply with regulatory requirements? How do they audit decisions made by opaque machine learning models? How do they maintain security when systems are granted broad, programmatic access to enterprise data? How do they manage the costs of AI inference, which can scale unpredictably as usage grows?
These are not problems that enterprises can solve on their own. They require specialised expertise, developed over years of experience across multiple clients and industries. They require repeatable methodologies, standardised tooling, and scalable delivery models. They require, in short, the kind of capabilities that technology service providers have historically brought to complex enterprise challenges.
The firms that succeed in packaging AI governance as a recurring service offering will create durable, defensible value pools. They will not be competing on price against low-cost providers or point-solution vendors. They will be selling outcomes—compliance, security, cost control—that are directly valued by enterprise clients. They will be building relationships that survive individual contract cycles. They will be shaping the market rather than reacting to it.
This is the path from survival to reinvention. It requires not merely adapting to AI but leading with AI—investing in the development of proprietary platforms, methodologies, and intellectual property. It requires not merely responding to client demands but anticipating them. It requires not merely defending existing revenue streams but creating new ones.
Conclusion: The Courage to Cannibalise
The authors’ concluding invocation of “trailblazers with the courage to cannibalise outdated paradigms before someone else does” is not rhetorical flourish. It is the central strategic imperative of the AI era.
Cannibalisation is painful. It requires accepting short-term revenue losses for long-term competitive advantage. It requires investing in new capabilities before the old ones have fully depreciated. It requires telling clients that the way you have worked together for decades is no longer sustainable. It requires betraying the legacy that made you successful.
But the alternative is worse. The firms that refuse to cannibalise their own business models will be cannibalised by competitors who have no such loyalty to the past. The revenue that they protect today will be stripped away tomorrow, piece by piece, contract by contract. The talent that they retain today will depart for more dynamic organisations, taking with it the institutional knowledge and client relationships that are the firm’s only durable assets.
The Indian IT services industry has navigated multiple technological transitions over its half-century history. It survived the shift from mainframes to client-server, from on-premise to cloud, from waterfall to agile. It adapted to the rise of global delivery, the proliferation of open source, the commoditisation of infrastructure. It has demonstrated remarkable resilience, ingenuity, and entrepreneurial energy.
The transition to agentic AI is different. It is not merely a change in the tools that developers use or the platforms that applications run on. It is a change in the fundamental economics of the industry. It compresses the effort that has been the primary source of revenue and profit. It disrupts the commercial models that have been the foundation of client relationships. It challenges the very identity of the industry, which has historically defined itself in terms of the labour it supplies rather than the value it creates.
This transition will not be easy. It will require difficult choices, painful adjustments, and genuine strategic courage. Many firms will fail to make the transition; they will be acquired, downsized, or simply faded into irrelevance. But others will succeed. They will emerge from the disruption as different kinds of organisations: more efficient, more valuable, more essential to their clients’ success. They will have traded labour arbitrage for intellectual capital, time sheets for outcome dashboards, reactive service for proactive partnership.
The Claude Cowork release is not the end of the Indian IT services industry. It is the beginning of its next chapter. The authors have provided a roadmap. The rest is execution.
Q&A Section
Q1: What is “agentic AI,” and why does the release of Claude Cowork represent a more significant threat to the IT services industry than previous AI advances?
A1: Agentic AI refers to systems that autonomously execute multi-step workflows rather than merely assisting human users. Unlike earlier generative AI tools that responded to prompts or automated isolated tasks, agentic AI can plan, execute, and adjust sequences of actions across multiple systems without continuous human supervision. Claude Cowork represents a more significant threat because its 11 plug-ins target core enterprise functions—customer support, finance, data analytics, marketing, product development, legal compliance—not peripheral tasks. It is designed to integrate with existing enterprise systems, not operate in isolation. Previous AI advances could be framed as productivity tools that augmented human workers; agentic AI directly displaces labour by automating workflows that previously required teams of developers, analysts, and support staff. The authors estimate potential effort compression of up to 40 per cent in build, support, and analytics work. This is not speculative; it is already being realised in early enterprise pilots. The threat is not that AI will eliminate all IT services work but that it will fundamentally alter the economics of the industry, compressing the labour hours that have been the primary source of revenue and profit.
Q2: What are the two “realities” that the authors invoke against the “end-of-outsourcing” narrative, and why do they provide a window of opportunity for incumbent service providers?
A2: The first reality is enterprise complexity. Large organisations run on fragmented data, legacy systems, and interconnected workflows that cannot be simply plugged into standalone AI tools. Real value comes from integration, orchestration, and governance—capabilities that technology service providers have developed over decades. This complexity cannot be resolved overnight; it requires deep knowledge of each client’s technical architecture and business context. The second reality is the autonomy gap. Fully autonomous agents at enterprise scale are still years away. Current agentic workflows “loop, stall or misfire” without robust guardrails. Security architectures required to grant agents broad system access do not yet exist, and enterprise leaders in regulated industries are deeply reluctant to approve such access. These realities provide a window of opportunity because they create time for incumbent providers to retool delivery models, upskill workforces, develop governance capabilities, experiment with new commercial approaches, and build the trust relationships that will be essential when enterprises are ready to deploy agentic AI at scale. The window is not infinite—the authors estimate 12-24 months before disruption intensifies—but it is sufficient for decisive action.
Q3: What is the “commercial reckoning” facing IT services firms, and how will the impact vary across different contract types?
A3: The commercial reckoning is the inevitable adjustment of pricing to reflect AI-driven reductions in labour hours. When a task that required 100 hours can be completed in 60 hours with AI assistance, clients will demand to share the productivity gain. When competitors offer AI-enabled delivery at lower prices, incumbents must match them or lose business. The authors estimate up to 40 per cent effort compression in build, support, and analytics work, with pricing resets lagging by one contract cycle. The impact varies by contract mix. High time-and-materials exposure: firms will feel impact first and most acutely, as routine work is displaced and purchased hours decline. Revenue falls; margins are squeezed. Balanced contract mix: volumes soften gradually; price resets occur at renewal rather than immediately. Low time-and-materials exposure: firms hold revenue slightly longer but face tougher fixed-price renegotiations as clients benchmark against AI-enabled delivery costs. Across all scenarios, the pattern is consistent: effort compresses first, pricing follows. Even when headline revenue remains stable initially, margin pressure increases as renewals replace against AI-driven delivery baselines and competitor output.
Q4: What is the prescribed solution to the commercial reckoning, and why is it described as extraordinarily difficult to implement?
A4: The prescribed solution is to move from effort-based billing to outcome-based pricing. Providers must stop selling hours and start selling results: reduced customer churn, faster time-to-market, lower operational costs, higher code quality. Compensation must be tied to impact rather than inputs. This is extraordinarily difficult for four reasons. First, risk transfer: providers must assume risk previously borne by clients, accepting payment contingent on achieving specified outcomes. Second, measurement capability: providers must develop analytical systems to credibly measure and attribute outcomes, which requires investment in data infrastructure and methodology. Third, delivery redesign: providers must reengineer processes around efficiency rather than effort, which requires retooling tools, retraining people, and rethinking management systems. Fourth, revenue cannibalisation: providers must accept short-term revenue losses from reduced billable hours to achieve long-term competitiveness. This requires genuine strategic courage—the willingness to betray the business model that made the firm successful before competitors do it for them. The authors note that this transition is “not merely a change in invoicing methodology; it is a fundamental reorientation of the provider-client relationship.”
Q5: What is the “productisation of AI governance,” and why do the authors identify it as a major opportunity beyond defensive adaptation?
A5: The productisation of AI governance is the development of repeatable, scalable service offerings that address the new risks and challenges enterprises face as they embed AI into sensitive workflows. These include: ensuring regulatory compliance of autonomous agents; auditing decisions made by opaque machine learning models; maintaining security when systems have broad, programmatic access; managing unpredictable AI inference costs; and establishing guardrails, change control, and cost discipline. These are not problems enterprises can solve independently; they require specialised expertise developed across multiple clients and industries. This is a major opportunity because it creates durable, defensible value pools that are not subject to the same effort-compression pressures as traditional services. Firms that successfully productise AI governance will be selling outcomes—compliance, security, cost control—that are directly valued by enterprise clients. They will not be competing on price against low-cost providers or point-solution vendors. They will be building recurring revenue streams and long-term client relationships. This is the path from survival to reinvention: not merely adapting to AI but leading with AI; not merely responding to client demands but anticipating them; not merely defending existing revenue but creating new sources of value. It requires investment in proprietary platforms, methodologies, and intellectual property.
