FutureIsNow Intelligence | Signals over noise
- Adoption is rising, but adoption is not the same as value
- What makes an AI agent different?
- Where the opportunity is most credible
- 1. High-volume, bounded workflows
- 2. Exception handling rather than blanket automation
- 3. Enterprise IT and cybersecurity operations
- 4. Customer operations
- The current enterprise signal: deployment partners are building for control
- Why the promised ROI often fails to materialise
- A practical deployment scorecard for Indian enterprises and GCCs
- What this means for India’s enterprises and GCCs
- The FutureIsNow view
Enterprise AI is moving beyond copilots that draft text or answer questions. The next phase is about systems that can plan a sequence of steps, use approved tools, retrieve business context and act within a defined workflow. These systems are commonly described as agentic AI. The opportunity is significant—but the business case is not automatic, and claims of effortless autonomy or guaranteed savings deserve scrutiny.
The more important question for Indian enterprises is not whether to deploy an AI agent. It is which workflow is suitable, what level of autonomy is safe, how success will be measured, and who remains accountable when the system gets something wrong.
Adoption is rising, but adoption is not the same as value
EY’s November 2025 GCC Pulse Survey reported that 58% of India-based global capability centres were investing in agentic AI, with another 29% planning to invest within a year. The same survey reported that 83% were investing in generative AI. These are survey findings about investment activity—not proof that those organisations have deployed autonomous agents at scale or achieved a particular return.
More Read
EY’s India AI survey also points to the implementation gap: while many organisations reported multiple generative AI use cases in production, only a small share reported enterprise-wide deployment. Moving from a promising pilot to a dependable operating capability requires integration, data quality, security, process redesign and sustained oversight.
Sources: EY GCC Pulse Survey 2025; EY, The AIdea of India: Outlook 2026.
What makes an AI agent different?
A conventional chatbot primarily responds to a prompt. A workflow agent may break a goal into steps, retrieve information, call approved tools, update a system and check whether the task succeeded. In practice, many products combine language models with rules, APIs, retrieval systems and conventional software. The label “agentic” alone does not tell a buyer how autonomous, reliable or safe a system is.
For high-stakes processes—credit decisions, payments, insurance claims, healthcare or compliance—the right design is rarely unrestricted autonomy. Enterprises should define permitted actions, approval thresholds, audit trails, fallback paths and a way to stop or reverse actions. Human review should be built into the workflow according to risk, not added after an incident.
Where the opportunity is most credible
1. High-volume, bounded workflows
Document classification, information extraction, case summarisation, ticket routing and routine internal service requests can be useful starting points when the inputs, permitted actions and success criteria are clear. Begin with assistance or recommendations, then increase autonomy only after the system demonstrates reliable performance.
2. Exception handling rather than blanket automation
Many business processes contain a long tail of unusual cases. An agent may help gather evidence, explain why a case needs attention and prepare a recommended next step, while a person handles exceptions or consequential decisions. This can improve throughput without pretending that every case can be safely automated end to end.
3. Enterprise IT and cybersecurity operations
AI can help investigate alerts, summarise incidents, recommend remediations and automate approved low-risk actions. But security agents themselves can become a new attack surface if they receive excessive permissions, act on untrusted instructions or lack traceable controls. Least-privilege access, sandboxing, logging and human approval for sensitive changes are essential.
4. Customer operations
Voice and service agents can retrieve account context, answer routine questions and complete selected tasks across channels. Their value should be measured through verified resolution, customer satisfaction, repeat-contact rates, escalation quality and total cost per resolved case—not how human-like a conversation sounds.
The current enterprise signal: deployment partners are building for control
On 12 October 2026, HCLTech announced a strategic partnership with Anthropic aimed at enterprise adoption of agentic AI, including a dedicated business unit and a focus on safety, cybersecurity and governance. The announcement is a useful market signal: large-scale adoption requires implementation capability and operational controls alongside model access. A partnership announcement, however, is not evidence that every proposed use case has already delivered measurable results.
Source: Reuters, 12 October 2026.
Why the promised ROI often fails to materialise
Claims such as 200–300% first-year returns, 10x faster onboarding or automatic 60% task automation should not be treated as industry benchmarks without transparent evidence. Actual results depend on workflow complexity, error costs, integration work, model and infrastructure costs, human review, change management and the volume of exceptions.
Before scaling, calculate the full cost of the workflow: model usage, orchestration, integration, monitoring, security, human oversight and remediation. Compare that with a credible baseline. Measure quality and risk as well as speed; an agent that processes cases faster but increases errors or customer complaints may destroy value.
A practical deployment scorecard for Indian enterprises and GCCs
- Choose the workflow: Start with a frequent, measurable task with clear boundaries and accessible data.
- Set a baseline: Record current cycle time, cost per case, error rate, backlog, customer outcomes and exception rate.
- Define the autonomy level: Decide what the system may read, recommend, execute, approve or never do.
- Test failure modes: Evaluate hallucinations, prompt injection, unauthorised access, tool errors, inconsistent outputs and unusual cases.
- Keep actions observable: Log decisions, sources, tool calls, approvals and outcomes; make rollback or escalation possible.
- Prove net value: Include ongoing technical and human costs, then compare results with the baseline over a meaningful period.
- Scale only with ownership: Assign a business owner, technical owner and risk owner, with a process for incident response and continuous evaluation.
What this means for India’s enterprises and GCCs
India’s GCC ecosystem has the engineering talent and process depth to help global organisations design, test and operate AI-enabled workflows. The opportunity is to move from being a location where automation is implemented to a centre that owns the architecture, evaluation, governance and continuous improvement of those systems.
That shift will require more than prompt engineering. Teams need process expertise, data engineering, software integration, security, model evaluation, risk management and the ability to redesign work with employees. Leaders should also plan for role changes and reskilling rather than assuming automation has no workforce impact.
The FutureIsNow view
Agentic AI is not a magic layer that makes enterprise work disappear. It is a new way to coordinate software, data and decisions—and it creates value only when the workflow is well chosen, the system is controlled and the result is measurable.
The signal to watch is not the number of agents announced or the size of a vendor’s promise. It is the number of production workflows that reliably improve business outcomes while remaining secure, auditable and accountable. For Indian enterprises and GCCs, disciplined execution—not autonomy for its own sake—will determine who turns experimentation into advantage.
FutureIsNow Intelligence — Intelligence for what comes next.
