The first day of AI & Big Data Expo Global and the Intelligent Automation Conference placed autonomous digital co-workers at centre stage. Yet the sessions repeatedly returned to less theatrical foundations: governed data, observable systems, reliable networks, physical safety, and workforce trust.
Speakers described agents as a step beyond robotic process automation. Instead of following fixed scripts, they can reason, plan, and execute across workflows. That flexibility may close the gap between a person’s intent and operational action.
It also makes weak data and vague accountability more dangerous.
Agents Cannot Outrun Their Data Foundation
SAP’s Andreas Krause argued that AI fails without trusted, connected enterprise data. An agent needs information that is accurate, timely, and relevant to the specific business context.
Gigaspaces proposed enhanced retrieval-augmented generation and semantic layers to ground language models in live enterprise facts. Representatives from Equifax, British Gas, and Centrica emphasised cloud-native, real-time analytics that reduces the delay between an event and a useful response.
Data access must remain controlled. Speakers from Informatica, MuleSoft, and Salesforce described governance layers that determine how agents use information and prevent an autonomous workflow from expanding beyond its authority.
The practical sequence is clear: define trusted sources and business meaning, control permissions, retrieve the smallest relevant context, and record what the agent used. A model cannot repair ambiguity it cannot see.
This groundwork should be reusable across projects. A governed semantic definition or approved retrieval pattern becomes infrastructure for the next agent instead of a one-off control rebuilt under deadline.
Physical and Software Autonomy Need Observability
Embodied AI introduces consequences beyond a wrong answer. Robots operating in factories, offices, and public spaces need safety protocols before they interact with people.
Oxford Robotics Institute research on time-of-flight sensors and electronic skin seeks to give machines awareness of themselves and their environment. Better perception can prevent collisions, but hardware, control software, operating rules, and emergency intervention must work together.
Software agents require a parallel form of visibility. Datadog’s Yulia Samoylova highlighted the need to observe internal state and reasoning as autonomous systems change how applications are built and repaired.
Logs should include plans, tool calls, data, timing, errors, approvals, and outcomes. Observability must support real intervention, not merely produce a record after the damage. High-impact actions need gates and safe fallback behaviour.
Physical systems additionally need tested emergency stops, safe operating zones, and responsibility that remains clear when perception confidence falls.
Readiness Includes Networks and People
Expereo’s Julian Skeels argued that AI workloads require sovereign, secure, always-on network fabrics. Distributed retrieval and inference degrade when latency becomes unpredictable, making connectivity part of application quality.
IBM Automation warned of an “illusion of AI readiness,” where technical enthusiasm understates process and organisational complexity. Jena Miller stressed human-centred strategy: employees who do not trust a system will not create a return from it.
Sanofi’s Ravi Jay urged leaders to ask operational and ethical questions early and decide where proprietary development creates value versus where buying a platform is more sensible. That choice should follow data, risk, capability, and differentiation—not fashion.
The conference’s first day suggested that standard automation maturity remains useful preparation. Organisations accustomed to process ownership, exceptions, monitoring, and controlled releases are better placed to introduce non-deterministic agents.
The agentic enterprise will not be enabled by autonomy alone. It needs data that carries business meaning, networks that behave predictably, machines that perceive safely, software that exposes its actions, and employees who understand the new division of work.
Those foundations may look conventional beside a digital co-worker. They are what turn one from a demonstration into a colleague the organisation can actually supervise.
