Summary

• HCLSoftware intends to acquire Croatian RPA provider Robotiq.ai, with completion expected in November 2026.
• Robotiq.ai adds the ability for automated workflows to operate software interfaces where APIs are unavailable or inadequate.
• The deal exposes a practical constraint on agentic AI adoption: many business processes still depend on old applications that were never designed for autonomous software.

The ambitions surrounding autonomous AI agents are colliding with an older piece of enterprise reality: a substantial share of business work still runs through applications that cannot be controlled cleanly through modern APIs, leaving software agents with decisions they can make but no reliable way to execute them.

HCLSoftware plans to acquire Zagreb-based Robotiq.ai and fold its robotic process automation technology into HCL UnO Agentic, adding an execution layer capable of interacting with enterprise applications where direct programmatic integration is unavailable or inadequate.

The acquisition is expected to close in November, although HCLSoftware has not disclosed the price in its announcement. Robotiq.ai’s platform is already used in banking, insurance, and telecommunications, according to the companies, with deployment options, audit logging, and security controls intended for regulated enterprise environments.

The two companies are not starting from scratch. HCLSoftware and Robotiq.ai have worked together since at least 2025, combining HCL’s orchestration technology with Robotiq.ai’s RPA software, so the acquisition converts an existing technical partnership into ownership.

Agentic AI still meets old software

Agentic platforms are generally designed to interpret a goal, decide which steps are needed, and invoke applications or tools to complete them. The model works most neatly when each business system exposes a well-documented API that allows software to retrieve information or perform an action in a controlled and predictable way.

Many enterprise environments look nothing like that architecture. Banks, insurers, telecoms companies, government bodies, and industrial organisations can have important processes running through desktop applications, old web interfaces, virtual desktops, proprietary software, mainframe front ends, and systems that were purchased long before API-driven integration became normal.

RPA grew partly to bridge that gap by allowing software robots to interact with screens and applications in ways that resemble human users. Bots can enter data into forms, copy information between systems, trigger transactions, and move through fixed interface workflows without the underlying application exposing a modern integration layer.

HCLSoftware wants to use that capability underneath its agentic orchestration. An AI-driven workflow might decide which action is required, while Robotiq.ai provides the mechanism for carrying out that action inside an application that the agent cannot reach through an API.

The execution layer creates its own fragility

The combination is useful precisely because RPA can reach awkward systems, although that flexibility comes with operational trade-offs. Automations that depend on screens, buttons, field positions, or application behaviour can break when interfaces change, login processes are redesigned, security controls intervene, or an application responds differently from expected.

Adding an AI decision layer above those automations also raises the cost of poor governance. A conventional RPA bot typically follows tightly defined rules; an agentic system may have more latitude to choose a path or decide which tool to call, making permissions, approval boundaries, audit trails, credential handling, and exception processes more important.

HCLSoftware is consequently emphasising governed execution rather than the idea of AI agents operating without controls. Robotiq.ai provides audit logging and flexible deployment options, while HCL UnO is intended to coordinate processes across agents, applications, data sources, and existing automation.

A practical example already exists inside the companies’ earlier partnership material. HCLSoftware describes a European bank using UnO and Robotiq.ai to automate onboarding for small-business customers, with orchestration triggering bots for front-office entry and back-office compliance checks across systems that do not all expose straightforward interfaces.

RPA is being pulled into the agent era

The acquisition also reflects how an older automation category is being repackaged around generative and agentic AI rather than disappearing beneath it. RPA was frequently sold as a way to automate repetitive office processes, but its ability to manipulate legacy systems gives it another role when AI software needs to perform actions beyond writing text or querying a database.

That does not mean organisations should preserve weak legacy architecture indefinitely simply because an agent can now click through it. Screen-based automation can become another layer that has to be maintained, tested, secured, and understood, particularly when hundreds of workflows depend on software whose original limitations remain unchanged underneath the automation.

Where modernisation is possible, APIs and well-governed system integration can still offer a more robust route. Yet replacing a core banking platform, insurance system, enterprise resource-planning application, or bespoke operational tool may take years, which is why businesses continue buying technology that allows old and new systems to coexist.

HCLSoftware’s acquisition therefore points to a less theatrical version of agentic AI adoption than autonomous assistants replacing whole departments. Production automation has to work with the software estate organisations already own, including the applications that nobody particularly likes but nobody can switch off.

Robotiq.ai gives HCLSoftware another way across that gap. If the acquisition closes as planned, the test will be whether combining AI orchestration with RPA produces reliable execution across messy enterprise environments without turning the automation layer itself into another difficult legacy system.