Legacy systems without APIs are becoming an agent-execution problem, not a rewrite prerequisite_

A surprising amount of enterprise work still ends at a screen. The system may be business-critical, but it has no useful API, the integration surface is incomplete, or replacing it would turn a workflow improvement into a multi-year modernization programme. That makes legacy UI execution a practical architecture problem for AI automation.

The market signal became concrete this week. HCLSoftware announced an agreement to acquire Robotiq.ai, an enterprise RPA provider, with the stated aim of extending HCL UnO Agentic from decision support into execution across legacy systems that lack APIs. The deal is expected to close in November 2026, subject to customary conditions. Banking, insurance and telecom are among the sectors highlighted for the technology.

Treat the UI as an adapter, not the source of truth_

The wrong abstraction is 'let the agent click through the application'. A safer one is: business intent → policy check → deterministic UI adapter → state observation → post-condition verification → authoritative workflow update. The agent can decide what needs to happen; the UI layer should expose a narrow, testable action contract.

That contract should specify the target application and version, permitted screens, required preconditions, exact action, expected post-condition, timeout, retry policy and evidence captured. If the application layout changes or the expected state is not visible, execution should stop rather than improvise.

API-first still matters_

UI automation is useful when an API does not exist or cannot cover the required transaction. It should not become the default integration layer. APIs usually provide stronger contracts, clearer error semantics and better observability. A production design should therefore route each action through the strongest available interface: API where possible, constrained UI execution where necessary, and human handoff where neither path is reliable enough.

The difficult failures are operational rather than cinematic: stale sessions, modal dialogs, changed labels, duplicate submissions, slow page loads, partial saves and a successful click that did not produce the intended business state. This is why post-condition verification matters more than a screenshot showing that automation reached the final screen.

Measure completed business state, not successful clicks_

Useful metrics include verified completion rate, UI-change failure rate, duplicate-prevention events, average recovery time, human-handoff rate, actions blocked by policy and the share of executions that can be migrated from UI automation to a stable API. These measures expose whether the automation is reducing operational work or simply moving fragility into a bot.

This pattern maps to Parallaxis capabilities around durable event-driven workflows, multi-system integrations, browser and application automation, CRM operations and controlled AI execution. Those capabilities are relevant engineering evidence; they are not a claim that Parallaxis has implemented HCL UnO Agentic, Robotiq.ai, or delivered the outcomes described by those vendors.

Legacy software does not always need to be rewritten before useful automation can begin. But when the only execution surface is a UI, that surface should be treated as a constrained integration adapter with explicit permissions, deterministic actions and verified post-conditions. The goal is not to make an agent good at clicking. It is to make the resulting business state dependable.