Wolters Kluwer's March article correctly identifies the shift from task-level automation to agentic orchestration of complete engagement workflows. For UK accountancy and legal firms, this matters because it exposes a uncomfortable gap between current capability and what regulators now expect. The FRC's audit standards and FCA Consumer Duty PS22/9 both assume firms have defensible, auditable decision-making. A basic invoice classification bot is auditable. An agent that autonomously decides which expense goes to which cost centre across 200 different client rule sets is not—unless the entire workflow has been redesigned and the technology stack integrated to create an auditable trail. Most mid-market firms have neither.

This story reflects a wider truth: the AI industry has sold automation, but clients actually need orchestration. Firms using point solutions like disconnected OCR engines, chatbots built on consumer LLMs, or rules-based RPA scripts are discovering they cannot scale beyond 60–70% without manual intervention. Harvey and Legora have built legal-native agents, but accounting has fewer such options. The reason is technical: accountancy workflows are deeply embedded in legacy ERP systems, tax software, and billing platforms that were never designed to expose their logic to AI. Without API-level integration and process mapping, agentic AI stays theoretical.

Trovix's position is that agentic AI requires three things in sequence, not parallel. First, map the actual workflow—not the one in the procedures manual, but the one people actually follow. Second, build integration fabric that connects your ERP, practice management, and knowledge systems so an agent can see the full picture. Third, instrument it with governance from day one using Trovix Audit to log what the agent decided and why. Firms that skip to 'buy an agent' without doing steps one and two will waste money and create compliance risk. The EU AI Act and ISO 42001 both require documented algorithmic impact assessment—you cannot retrofit that to a deployed agent.

If you run a mid-market accountancy, legal, or insurance firm, here is what to do immediately. Audit your technology stack: can your systems talk to each other via APIs or are they islands? If they are islands, plan integration work before you buy agentic AI. Second, identify one high-volume, standardized workflow—not a complex client relationship, something like expense categorization or invoice matching—and use that as your pilot. Third, ensure you have governance infrastructure in place (FCA-regulated firms should reference PRA SS1/23 on third-party AI risk). Do not let vendor roadmaps drive your timeline. A 2027 implementation done properly beats a 2026 implementation that fails audit.

Source: Wolters Kluwer