AI-Wiki

AI Agents and the Business of Complexity | Kelly Bayer Rosmarin (CEO & Group MD, Australian Unity)

5 min read

Bayer Rosmarin (Australian Unity) with BCG — AI agents and the business of complexity

▶ Watch on YouTube · Boston Consulting Group · 26:24

Kelly Bayer Rosmarin is drawn to complex problems – from scaling AI agents to reimagining how technology can support Australians as they age.

In this episode, Kelly sits down with Sanjay Chari to explore the challenges occupying her thinking today, drawing on a career spanning banking, telecommunications, technology, health, wealth and care. Kelly shares her perspective on what it will take to manage AI agents at scale, why the hardest problems are often the most energizing, and the principles that shape how she approaches teams and personal growth.

BCG maintains a professional relationship with the company represented by this episode’s guest.

TL;DR

A 26-minute interview on the Boston Consulting Group channel, published 29 September 2026. BCG’s Sanjay Chari talks with Kelly Bayer Rosmarin, CEO and group managing director of the Australian mutual Australian Unity for five months, formerly CEO of Optus and an executive at Commonwealth Bank. About a third is on AI; the rest is leadership and career. Four AI claims:

  1. Agents need what people already have: HR and risk systems. Between roles she co-founded, and now chairs, a start-up for the governance and management of AI agents. Organisations manage “non-deterministic intelligence” today, people, “with HR systems and risk management systems”; her company builds “the equivalent … for AI instances and AI agents,” a “system of record for your AI agents.”
  2. Pilots do not scale because nothing watches them. Firms “are still in the phase of building experiments, and they’re struggling to take things to scale,” a “chicken and egg”: without systems to monitor and manage agents for the long run, it is hard to get them into production.
  3. Making agents deterministic removes their value. “If you try and make it deterministic rather than non-deterministic and button it down for every use case, you’re actually removing a lot of the power of AI.” It would be like telling people “they can’t do a whole bunch of human things because it’s too risky.” The alternative: monitor, and alert people to review when an agent goes “outside of risk appetite or policy.”
  4. Not every agent is IT’s problem. “Right now all roads lead to an IT engineering or development team,” which cannot be accountable for every agent’s business outcome. The system should let the business see whether an agent meets its objectives and let risk and compliance confirm it operates within appetite.

Key claims

Prevention in aged care

Australian Unity covers health insurance, home health (“the largest home health provider in the country”), retirement villages, residential aged care and, through infrastructure funds, hospitals. Australia will see three times as many people turning 80 in the next 20 years as in the last 20. Her AI opportunity is prevention: combining insurance data, in-home care observations and monitoring from phones, watches and bespoke devices “to anticipate and prevent situations from deteriorating.” “We’re just at the beginning of experimenting.”

Culture and what gets attention

Trust comes from culture aligned to purpose (at Optus, the purpose of being “the most loved everyday brand”). Resilience comes from making it “easy for people to do the right thing”; complicated systems teach people to go around them. People are not driven by KPIs, which most junior staff cannot name, but “by what their manager notices, calls out in a meeting, praises them for.” Her career advice: understand how the core business succeeds, take roles that are 50% giving and 50% learning, and treat your team as your leverage (“I’m the kind of person who cancels one-on-ones last”).

Dynamic-capabilities reading

  • contextual/internal-barriers: pilots that cannot reach production for lack of monitoring, and over-restriction of agents out of risk aversion.
  • digital-transforming/redesigning-internal-structures: a system of record that brings business owners, risk and compliance and IT together around each agent, instead of routing everything to IT.
  • strategic-renewal/organizational-culture: purpose-led culture, and management attention as what actually steers behaviour.

How it connects

  • Smith (Okta) & Cohen (NanoCo) at TechCrunch Disrupt — “we personify agents in the same way that we do a human worker” but give them none of the onboarding. Rosmarin’s company is a product built on the same analogy, extended from identity to performance and risk.
  • Chatterjee on the agentic governance gap — governance built for documents, not transactions. Rosmarin agrees on the gap and adds a design rule: do not close it by removing what makes agents useful.
  • McKinsey’s CEO agenda — 500,000 agents per 50,000 employees as a scenario, which is the scale her system of record assumes.

What was actually ingested

The full interview (26:24, length_seconds: 1584) through the yt-dlp fallback from the auto-generated track, 111 lines. The channel’s chapter timestamps run ahead of the long caption cues, so a chapter heading in the raw file can sit a little before its content. The start-up she chairs is not named in the interview. No stills: a two-person studio interview.

Linked entities and concepts

Dangling (single-source mention, deferred): Kelly Bayer Rosmarin, Sanjay Chari, Australian Unity, Optus, Commonwealth Bank.

Scope and reliability

A CEO interviewed by a consultancy with a disclosed client relationship, and a founder describing the market her own start-up sells into: her claims that pilots stall for lack of agent governance support her product. Nothing is measured. The interview is useful for the framing (agents as a workforce needing HR-like systems) and for one design rule, not as evidence of how often pilots fail.

Relationships

Supports

Graph View

Backlinks