Without industry-native operational autonomy, you’ll just keep babysitting copilots

For most COOs, operational autonomy seems out of reach because every attempt to scale AI increases supervision rather than reducing it. This can only be achieved when AI runs multi-step, cross-system workflows using industry-native context (workflow topology, institutional exceptions, regulatory constraints, system dependencies) and embedded trust architecture (explicit risk tolerance, policy-based escalation, and real-time auditability at the outcome level).

Announced in February 2026, the Infosys–Anthropic collaboration, pairing Claude models with Infosys Topaz and the Claude Agent SDK, is a direct bet on addressing this gap.

Two-thirds of enterprises are stuck in copilot sprawl

A recent HFS Research study showed that roughly two-thirds of enterprises remain stuck in low-complexity, assistive copilot deployments, concentrated heavily in software development, customer service, and general knowledge work. That’s not surprising; these are domains where “answers” are accepted quickly, and the cost of being wrong is usually limited or recoverable (see Exhibit 1).

Exhibit 1: Most organizations remain in low-complexity deployments

Source: HFS Research, 2026

But for COOs, the true value lies in core business workflows: order-to-cash, accounting, network operations, field service, demand planning, operations management, and quality management. We’ve been seeing many individual and specific use-case demos in these areas, but nothing that works end-to-end. The Infosys–Anthropic partnership is explicitly designed to target multi-step workflow execution in system-heavy operations where copilots have fallen short.

The jump to operational autonomy is a step change, but with context and trust as bottlenecks

A clean way to separate agent demos from operational autonomy is to ask where the program sits relative to the autonomy threshold (see Exhibit 2).

Exhibit 2: The autonomy threshold separates agent demos from where humans sit in the workflow

Source: HFS Research, 2026

Most enterprises are stuck at level 1 and see level 2 as the best they can get to. The move to level 3 is not a continuum but a step change constrained by two bottlenecks:

1. Industry-native context: No single owner holds the “truth” in industry operations. It resides in institutional process logic, organization-specific exceptions, legacy system behavior, and tacit domain judgements.
2. Embedded trust architecture: Operational autonomy forces a move from deterministic controls to probabilistic execution. This is not a learning curve problem but a design challenge: building a trust architecture that defines acceptable error tolerance, exposure per action (financial, service-level agreement [SLA], compliance), escalation mechanism, and audit trail.

Enterprises trying to address these challenges in their organizational silos often get stuck because they underestimate the difficulties of codifying fragmented institutional knowledge and designing a trust architecture that can safely scale.

The partnership is a strong bet on industry-native operational autonomy, not simply model access

Infosys and Anthropic came together with three goals in mind:

1. Pick telecom as the proving ground: The partnership is being launched in telecommunications with a dedicated Anthropic center of excellence (CoE). Telcom has complex workflows, heavy systems, high SLA sensitivity, and significant regulatory exposure. If customized agentic systems can reduce exception density and supervisory overhead here, it becomes a credible blueprint for any regulated or system-heavy industry.
2. Focus on multi-system agentic autonomy: The companies are positioning their partnership around agentic AI, going beyond Q&A to handle multi-step tasks (for e.g., claims processing and compliance reviews), using the Claude Agent SDK for persistent execution across systems. For COOs, this means that agents can operate across system boundaries without human intervention at every step.
3. Remove legacy ceiling on autonomous operations: The collaboration integrates Claude models with Infosys Topaz and explicitly targets legacy modernization. This matters because aging infrastructure has historically been the hard ceiling on deploying agentic workflows at scale, and abstracting that layer converts isolated pilots into enterprise-wide throughput.

The real test is whether these COEs become more than just another press release and slide in service provider proposals. Proof of success will come from referenceable production deployments and reusable industry IP, contributing to the US$1.5 trillion Services-as-Software™ economy.

The Bottom Line: If you keep deploying generic agents, you’ll continue paying to babysit them.

Enterprises can’t achieve operational autonomy just by identifying more uses case to build agents. The Infosys–Anthropic partnership signals that industry-native design with domain translation and trust architecture lays the foundation for the operational autonomy needed to scale agentic AI across the organization. Without this, enterprises will continue deploying copilots that generate insights but can’t act safely across workflows without constant supervision.