Why boards need AI agents (before agents need them)

Picture a Tuesday morning in a mid-cap Australian boardroom. The chair has read 340 pages of papers. The CFO has briefed her on three risks she did not know existed last week. And the most consequential question on the agenda, whether the agentic system the engineering team deployed last quarter should keep its access to the customer database, will get nine minutes before lunch.

This is not a hypothetical. It is roughly how every board meeting in the country is now structured, and it is the reason almost no Australian board is governing AI well.

In March 2026, ASIC Chair Joe Longo told the AICD’s Australian Governance Summit that agentic AI is “an inflection point in how organisations manage risk.” Quoting Anthropic’s Dario Amodei, he warned that humanity is being handed almost unimaginable power, and it is deeply unclear whether our systems possess the maturity to wield it.

If you chair an Australian board, that sentence should worry you more than it has.

Why your best directors are your biggest problem

Boards are very good at the things boards have always been asked to do. They read carefully. They probe management. They watch the risk register. Five traits show up almost everywhere: a love of governance for its own sake, a strong risk aversion hard-wired by directors’ duties, a heavy reliance on data in the form of board packs running to several hundred pages, a conservatism in temperament reinforced by who ends up around the table, and a habit of looking backwards because most board reporting is a mirror of last quarter rather than a window onto next year.

Learn more: Toby Walsh debunks the top 12 artificial intelligence myths

These qualities have kept Australian companies out of trouble for decades. They are also, almost without exception, the qualities AI rewards least. The technology moves faster than any quarterly cadence, punishes caution as readily as recklessness, and asks directors to imagine futures that look nothing like the recent past.

Agentic AI raises every stake.

What your developers can do this week that they couldn’t do last year

Until 2025, AI mostly produced text and pictures. A model gave you a draft. A human decided what to do with it.

Agents have collapsed that distance. As Mayer Brown put it in February 2026, agentic systems “autonomously plan and execute multi-step tasks to achieve a goal,” calling tools and other agents to act on the company’s behalf. McKinsey’s 2026 AI Trust Maturity Survey put the implication starkly: organisations now have to contend with systems doing the wrong thing, not just saying it. Only about 30% have reached meaningful maturity in agentic AI governance.

Two questions follow. Most directors have not asked either.

First, do you understand what your own developers can do now? The line between specifying software and writing it has dissolved. A junior engineer with the right agent can stand up production code in a morning that would have taken a team a quarter.

Second, do you know what loose agents inside the organisation can already touch? Customer records. Supplier ledgers. Payment systems. Public communications. These are not future questions. They are this-quarter questions.

The shadow boardroom you didn’t know you had

The most uncomfortable finding in the recent Australian research is not about AI strategy. It is about your own directors.

In late 2025, the AICD published Early Insights on AI use in boardrooms. Its central observation: director-level AI use is happening, but on an individual rather than collective basis. The AICD called it “a two-speed dynamic at play, with collective board use of AI lagging ‘shadow’ individual director use.”

Learn more: Why boards are asking the wrong question about AI strategy

Translation: your directors are quietly using ChatGPT to read your board pack in the back of an Uber, while your formal board agenda still treats AI as something the technology committee will report on next quarter.

Longo offered a second number that should sit uncomfortably with every chair. Board time spent on compliance has doubled from 24% to 55% over the past decade. More than half of every meeting is now backward-looking, defensive work, at exactly the moment the technology in front of you rewards forward-looking experimentation.

Yes, but: the question that will break the room

The first time someone proposes putting an AI agent into a board meeting, the senior counsel will raise a hand. The objection will be discoverability and privilege. It will be a fair objection.

A fair objection deserves a fair answer. AI-generated minutes are discoverable in exactly the same way as handwritten ones are. The legal exposure is real, but it is not new. In February 2026, Directors & Boards set out a workable answer: governance-specific tools with contractual data ownership, mandatory human review before any draft becomes the official record, and automatic deletion of raw audio once minutes are approved.

The question is not whether agents create new legal risks. They do. The question is whether those risks outweigh the cost of continuing to govern at the speed of paper.

The two agents that should be at every board meeting

Here is the part most boards have not yet imagined.

Picture the same Tuesday morning, six months from now. Two agents are present, with the chair’s permission and tightly scoped data access. Neither is a director. Neither makes a decision. Australian law leaves no room for doubt on this point, and the AICD has been clear: only natural people can be directors.

The first agent is your governance partner. It opens the meeting by reporting which decisions from last month have been actioned, which have slipped, and which need a fresh decision. As the discussion runs, it drafts minutes against your house style and builds the action register in real time. Between sessions, it chases unresolved items so the next agenda is not a graveyard of last month’s good intentions.

The second is your research partner. Briefed before the meeting, it reads every regulator update, peer announcement and relevant news item from the previous month. It prepares devil’s-advocate briefs on the major proposals and conducts scenario analyses of the strategic options. When a director asks “what are our peers doing on this,” the answer is in the room, not in next quarter’s pack.

Learn more: Spending big on AI: So where’s the proof it’s working?

Tools that do parts of this exist today: OnBoard, Diligent, V7, Sherpany, Boardable. Most boards using them treat them as glorified dictation. That is selling them short.

The 55% of board time now spent on compliance is exactly the kind of work agents are good at carrying. The 45% left for strategy is exactly the work that benefits from a tireless research partner.

From yes but to yes and: what to do on Monday morning

Boards have spent two years asking “yes but” questions about AI. Yes, the productivity gains look real, but what about the bias risk? Yes, the customer service scores improve, but what about hallucinations? These are good questions. They are also insufficient on their own.

The AICD recently observed that boardroom conversations are shifting from “how do we govern AI” to “how does AI help us govern.” That is the shift. What follows is not a framework. It is three conversations that need to happen before the next board cycle, one for each role that can make them happen.

If you are the chair: put two items on the next agenda. The first is a standing report from management on every agentic system currently operating inside the business, what data it can access, what actions it can take, and who approved it. Not a strategy paper. An inventory. The second is a 90-day pilot of an AI governance tool in the boardroom itself, scoped to minutes, action tracking and pre-meeting research. Run it alongside your existing company secretary process, not instead of it. Give the board a sense of what agents can do before you ask them for an opinion.

If you are a director: stop using AI privately and start using it collectively. The shadow use the AICD identified is understandable but dangerous: it creates an invisible gap between what individual directors know and what the board can act on. Bring your AI use into the open. Ask three questions at the next meeting: what agentic systems are live in the business today? Who has the authority to deploy new ones? Does our D&O insurance cover decisions made or influenced by autonomous systems? If nobody at the table can answer, that is the answer.

If you are management presenting to the board: change the frame. Do not walk in with a risk paper about AI. Walk in with a one-page map of every AI agent currently running, who owns it, what permissions it has, and what guardrails are in place. Then show the board what an AI-assisted governance cycle looks like: a pre-read prepared by a research agent, a meeting supported by a governance agent, and a follow-up cycle that actually closes the loop on last month’s decisions. Give the board something to react to, not something to fear.

The traits that make boards cautious are not the problem. Caution is precisely the quality needed to deploy this technology thoughtfully. The problem is when caution becomes the only setting.

Agents do not punish caution. They punish the kind of caution that mistakes itself for control. Nine minutes before lunch is not enough time to govern that. It never was.

Learn more: Why boards should be nervous about the rise of “shadow AI”

FAQ: Agentic AI for board directors

What is agentic AI governance for boards?
Agentic AI governance is board oversight of AI systems that can plan tasks, use tools, access data and take actions for an organisation.

Why does agentic AI create board risk?
Agentic AI creates board risk because systems can act across customer data, payment systems, supplier ledgers and communications without traditional approval steps.

What should Australian directors ask about agentic AI?
Directors should ask which agentic systems are live, what data they can access, what actions they can take, who approved them and what controls apply.

How can boards reduce shadow AI use?
Boards can reduce shadow AI use by creating approved tools, agreed processes and collective rules for how directors use AI in board work.

Can AI agents make board decisions?
AI agents can support research, minutes and action tracking, but board decisions remain the responsibility of directors.

What should management provide to the board on AI?
Management should provide an AI inventory listing each agent, its owner, permissions, data access, actions, guardrails, and approval status.

Jeffrey Tobias is an accomplished and prominent innovation thought leader and strategist, drawing expertise from the academic, government, entrepreneurial and corporate worlds. He serves as an Adjunct Professor and Fellow at AGSM @ UNSW Business School, and holds a B.Sc (Hons), University Medal and Ph.D from UNSW Sydney. In 2003, he founded The Strategy Group, where he currently serves as Managing Director.