Robotic process automation used to be the quiet workhorse of enterprise IT — bots clicking through screens so people didn't have to. This week's news shows the category is being pulled in two directions at once: upward into conversational, decision-making AI, and downward into stricter governance for when those decisions go wrong.

Automation Anywhere buys its way into the front office

The biggest move came on October 7, when Automation Anywhere announced a definitive agreement to acquire Boost.ai, an enterprise conversational and voice AI company, from Nordic Capital. Automation Anywhere entered a definitive agreement to acquire Boost.ai, an enterprise conversational AI company, from Nordic Capital, with the transaction expected to close in Q4 2026. It's the company's second AI acquisition in roughly a year, following its purchase of Aisera. The agreement follows the late-2025 acquisition of Aisera, which added enterprise knowledge and automated employee service to the platform.

The strategic logic is to close what the company calls the "conversation-to-outcome loop." By integrating Boost.ai's high-resolution conversational engine with its agentic automation platform, Automation Anywhere aims to let AI interpret intent, apply company rules, coordinate agents and humans, and execute work across ERP, CRM, and other systems. Boost.ai brings a mature, regulated-industry customer base to the table — a voice and conversational AI stack supporting more than 36 languages and operating under GDPR, HIPAA, and SOC 2 standards — which lines up neatly with Automation Anywhere's existing enterprise footprint in banking and telecom.

HCLSoftware snaps up a Croatian RPA specialist

A quieter but equally telling deal: HCLSoftware's agreement to acquire Robotiq.ai, a Zagreb-based enterprise RPA provider. HCLSoftware plans to acquire Croatia-based Robotiq.ai, adding enterprise robotic process automation capabilities to its HCL UnO Agentic platform, with the deal expected to close in November 2026. The rationale is strikingly plain: AI agents can reason all they want, but they still need hands to operate legacy software. The acquisition addresses the practical challenge of connecting AI systems capable of making decisions with applications where APIs are unavailable or insufficient, letting automated workflows interact directly with those applications. It's a reminder that classic RPA — the unglamorous business of automating click-paths — remains indispensable plumbing for agentic AI, not a technology it's replacing.

UiPath tackles the "approval theatre" problem

On the governance side, UiPath published a detailed technical post this week describing a failure mode it calls "approval theatre," where a human approves one version of an AI agent's action but the agent executes a different, unverified version. The company's proposed fix is a design pattern it calls an approval object. An approval object is an immutable record of a specific proposed action that carries the action type, exact parameters, risk level, approver identity, and timestamp, which an agent cannot bypass without the tool re-fetching and validating the record against the parameters it's about to send. If an agent replans after approval, the system is designed to catch it: replanning the parameters after a human has said yes causes the check to fail, and the tool refuses to act.

Why it matters

Taken together, these three stories sketch where RPA is heading: it's no longer a standalone category but the execution and compliance substrate for agentic AI. Analysts already expect the broader market to balloon as AI integration deepens — one forecast puts the RPA market growing from roughly $8 billion in 2026 to $28.6 billion by 2031, driven largely by generative AI integration enabling automation of unstructured processes — and this week's deals suggest vendors are racing to own both ends of that stack: the conversational front door and the audited, rule-bound execution layer behind it.

Sources

Agentic AIAutomation AnywhereDaily RPA UpdateEnterprise AIRPAUiPath