Enterprise AI Adoption and Agentic AI

From AI pilots to measurable, governed production value

Enterprise AI adoption is not just a matter of giving people access to new tools. Organisations need to choose valuable use cases, change how work gets done, establish clear ownership, and build the data, engineering and governance foundations to operate AI reliably.

As AI systems gain the ability to use tools and take actions, the challenge grows. Leaders must decide what an AI system may do, how its behaviour will be monitored, and when a person must intervene. This hub brings together Enterprise Tech Talk conversations and articles on moving from experimentation to responsible, measurable AI adoption, including the emerging use of agentic AI.

What's Changing Now?

Agent adoption is advancing at different speeds

McKinsey’s August 2026 global survey found that 40% of respondents from organisations with more than US$1 billion in annual revenue reported scaling AI agents in at least one function, compared with 27% a year earlier. The figure for smaller organisations remained broadly unchanged at 22%.

This suggests a widening gap in organisations’ ability to absorb agentic AI, rather than a uniform transition to autonomous enterprises. These are self-reported survey results, not audited deployment measures.

Source: McKinsey — The state of AI in 2026

Delegated work is spreading beyond software development

OpenAI’s Enterprise Signals, updated in August 2026, reports growth in enterprise Codex use across legal, sales, recruiting and marketing, alongside engineering. It also reports that Codex accounted for 64% of combined Codex and ChatGPT output tokens among its enterprise customers in June.

The signal is growing use of AI to execute work across functions. Token share measures activity within one provider’s customer base; it does not establish the proportion of enterprise work automated or the value delivered.

Source: OpenAI — Enterprise Signals

Operating costs are becoming a constraint

McKinsey’s survey found that about one in five respondents said AI operating costs were constraining their organisation’s AI use. This finding covers AI broadly, rather than agents alone.

For agentic workloads, the relevant emerging issue is the cost of completing useful work: repeated model calls, tool use and human checking can change the economics behind an impressive demonstration.

Source: McKinsey — The state of AI in 2026

Interoperability is attracting broader industry participation

In August 2026, the Linux Foundation announced 57 new members of the Agentic AI Foundation, including Visa, Wells Fargo and Alibaba. The foundation supports open agent infrastructure and projects including Model Context Protocol, which connects AI applications with tools and data.

This is a signal of industry investment in shared foundations. Participation in a standards ecosystem does not, by itself, establish compatibility, security or production maturity across every implementation.

Source: Linux Foundation — Agentic AI Foundation membership announcement

Guidance is becoming more specific to agent behaviour

The Australian Government’s Agentic AI addendum, updated in June 2026, extends its AI technical standard with guidance covering agentic systems across their lifecycle, including governance, evaluation, integration and monitoring.

The distinction matters because agents introduce actions and interactions that require attention beyond the quality of an individual response. The guidance is written for Australian Government agencies; it provides useful context for private enterprises without creating a universal private-sector mandate.

Source: Australian Government — Agentic AI addendum

Why The Change Matters?

Three developments reinforce one another: AI systems can use more tools, organisations can supply more business context, and emerging protocols offer common ways to connect components. Together, they broaden the work that can be delegated to AI.

ETT interpretation: as agents become more connected to live workflows, the constraints increasingly span the whole organisation: information access, integration, accountability, user adoption and operating economics. This helps explain why technical progress and enterprise adoption can move at different speeds.

Enterprise implications

Area	What the trend brings into focus
Business operations	Improvements to individual tasks may affect downstream hand-offs, exception handling and service outcomes.
Architecture and data	Agent effectiveness depends on usable context, reliable integration and access that follows business permissions.
People and operating models	Delegation changes the balance between doing work, specifying it, reviewing results and managing exceptions.
Risk and assurance	Visibility must extend to the actions agents take and the systems they affect.
Investment and economics	Greater activity can increase both potential benefit and operating cost; the relationship needs to be visible.

These are ETT’s interpretation of the developments above. Their significance varies with the workflow, its consequences and the organisation’s existing capabilities.

Signals to watch

Over the next six to twelve months, the most informative signals will include:

Sustained production outcomes: evidence of better service, quality or economics after initial deployment.
Cost per completed workflow: whether improved models and tooling outweigh additional execution and review effort.
Human intervention: how frequently people must correct, restart or take over agentic work.
Interoperability in practice: whether organisations can change models, tools or providers without substantial rework.
Operational assurance: clearer evidence of agent permissions, traceable actions and effective recovery.
Adoption beyond early users: repeat use across business functions, rather than reliance on small specialist teams.

Related Podcast Conversations

Watch

Sources and updates

Evidence reviewed on 18 October 2026. Source dates range from June to August 2026; they describe the latest editions verified for this overview. Survey findings, provider usage data, industry announcements and government guidance represent different forms of evidence and should be read within their stated scope.