AI Agents Cut Legal Review Time, Most Leaders Unprepared

Deloitte’s August 2026 report puts a sharp number on enterprise AI’s credibility problem: only one in five leaders feel their organisations are ready to redesign core processes around autonomous agents, yet nearly three-quarters of U.S. executives expect roughly half their processes to be rebuilt around agents within four years. Gartner sharpens the point further, predicting more than 40% of agentic AI initiatives will be canceled by 2027, largely due to escalating costs, unclear business value or inadequate risk controls.

Finance and HR: Early Gains, Uneven Readiness

In finance, agentic AI is taking over invoice matching and accounts payable workflows. Agents reconcile invoices against purchase orders and receipts across disparate systems, parse PDF line items from incoming email and log bills directly into accounting software. Genpact has focused heavily on these deployments, extending agents into financial reporting, procurement and compliance workflows.

HR is following a similar path. Agents can sync name changes across payroll, benefits and tax records in a single orchestrated workflow rather than requiring manual updates across each system. Oracle notes that agents can update employee records, manage benefits documentation and generate performance review summaries. Self-service portals, onboarding document verification and personalised training programmes are becoming standard deployments in the space.

Supply Chain Leads on Adoption

BCG’s June 2026 analysis found 44% of companies deploying AI in supply chain management, ahead of both finance and HR. The use cases are concrete: agents monitor IoT sensors, ERP systems and logistics platforms to predict demand, optimise inventory and trigger automatic reorders. Supplier communication, discrepancy flagging and exception escalation are also being automated, removing the manual follow-up burden from planners.

Forsys offers a sharper example of measurable impact in an adjacent function: a healthcare client used its contract management solution to significantly reduce legal review time per contract, according to the company. The pattern is consistent with what AI is doing to contract review across professional services more broadly.

Multi-Agent Orchestration

Microsoft Copilot Studio and Salesforce Agentforce provide native integrations within their respective platforms, letting agents operate across Microsoft 365 and CRM environments without custom connectors.

At the framework level, LangGraph handles durable, stateful agents with human-in-the-loop checkpoints; CrewAI supports role-based multi-agent workflows where agents can delegate tasks to one another; LlamaIndex handles retrieval-augmented generation workloads, ingesting and indexing proprietary data as a retrieval layer other frameworks call into. The governance questions these multi-agent architectures raise are not primarily technical, they concern where human approval must sit in the decision chain and how to produce audit trails regulators will accept. No framework solves that for you.

Where Initiatives Stall

Only 15% of organisations have scaled orchestrated multi-agent deployment, and most of those remain in low-risk, low-return applications. Leaders consistently cite fragmented, inaccessible data as the primary impediment, followed by integration cost and complexity. The readiness gap Deloitte identified is not a perception problem, it maps directly onto the technical debt that makes connecting agents to enterprise systems expensive and slow. Until data architecture catches up, the gap between executive expectation and operational reality is likely to widen before it narrows.