In an opinion column published by Zeus Kerravala on SiliconANGLE, findings are analyzed from a new study by Cisco Systems Inc. titled "Permission to Scale: The Critical Role of Trust in Successful AI Adoption." The research, authored by independent researcher Brad Howarth and presented at the Cisco Live APJC conference in Melbourne, is based on data from 25 Australian organizations. According to the column, while for several years the conversation around enterprise artificial intelligence focused on technological capabilities — such as which model is smartest and which graphics processing unit (GPU) is fastest — the current limit on the pace of expansion is not rooted in technology. Today, AI systems are capable of doing far more than most organizations are willing to permit them to do, and the primary gap that has emerged stems from the issue of trust. As part of the conference, Cisco hosted a press panel moderated by Howarth, featuring executives from organizations that participated in the study, including Carl Solder, Chief Technology Officer at Cisco Australia and New Zealand; Anna Tosky, Chief People Officer at Elmo Software; Chris Locke, CTO at Flight Centre Travel Group; Naz Wilson, Group Culture and Enablement Manager at Funlab; and Lucio Ribeiro, Chief AI Officer at TBWA Australia.
The Case of Australia as a Test for Global Adoption
Kerravala notes that although a sample size of 25 organizations is small, the findings align with conversations he has held with corporate leadership, emphasizing that the largest barrier to scaling AI use is not technical but people-related. Kerravala notes that Australia makes an interesting test case. According to research by the University of Melbourne and the firm KPMG cited in the report, half of Australians use artificial intelligence regularly, but only 36% trust it, and just 30% believe its benefits outweigh its risks. This rate is the lowest among 47 surveyed countries. Kerravala adds that if you can build trust in AI in Australia, that work will be able to scale globally. In a press release, Stefan Leitl, Vice President and General Manager of Cisco Australia and New Zealand, addressed the advantages of winning over this region, stating: "It’s a playbook that travels well beyond Australia."
The Cultural Pillar and the Trust Adoption Flywheel
In his remarks on the panel, Carl Solder noted that out of the six pillars in Cisco's AI Readiness Index, Chief Information Officers (CIOs) and Chief Technology Officers (CTOs) tend to gravitate toward areas familiar to them, primarily infrastructure and data, leaving the topic of culture to others. Solder described the case of a customer that invested in infrastructure, governance, data, and information technology skills, but did not focus on culture, and found itself facing strong pushback from employees and customers regarding the use of the toolsets it had invested in. Kerravala added that this is a phenomenon he has witnessed repeatedly, where companies spend millions on AI platforms only to see utilization remain tepid.
Howarth described an adoption "flywheel" mechanism in the discussion, in which successful use creates evidence, evidence builds trust, and trust enables broader use. Kerravala emphasized that many companies sit around waiting for trust to appear before deploying solutions, whereas the proof of trust's existence must be something measurable. Furthermore, it was stressed that trust in AI must be established with realistic parameters. Many people reject AI systems after a single mistake, even though in most cases the systems' error rate is much lower than that of humans. Kerravala compared this to the automotive industry's reaction when a crash involving a Tesla vehicle triggers a media storm, even though in reality Tesla's Autopilot system experiences one crash for roughly every 7 million miles, compared to human drivers who experience one crash every 1 million miles on average. The expectation of perfection is not a realistic goal and can erode trust, whereas the proper goal for users is being faster and more accurate relative to alternatives.
Leadership Approaches and Field Examples
Several examples of leadership practices were presented during the discussion. Chris Locke from Flight Centre offered the example of the company's 77-year-old founder, Graham "Skroo" Turner, who uses AI every day and is considered a regular user of Anthropic's Claude model. Locke noted that Turner, who said in 2000 that the internet would not catch on, is now betting that AI will catch on, thereby granting the organization permission to experiment. Lucio Ribeiro from TBWA Australia described a more structured approach, defining trust as an equation based on four elements: ownership, competence, consistency, and transparency. Ribeiro argued that asking anxious employees whether they want AI does not yield an accurate answer, just as one does not ask a barber if a haircut is needed. As an alternative, he suggested honesty: companies cannot promise to preserve every job, but they can promise to make employees the most prepared, skilled, and trusted if AI disrupts their roles.
At Funlab, where 79% of the workforce is age 21 or younger, they rely on peer learning. A team named the "AI Super Squad" convenes representatives from every department once every four weeks. Naz Wilson explained that the first sessions were devoted to clarifying which tools are defined as safe to use, as the true limitation for many employees stemmed from not knowing whether they were allowed to perform certain actions.
Calibrated Trust and Addressing "AI Slop"
The central concept in the report is calibrated trust, according to which employees need to know when to rely on AI, when to verify its outputs, and when a decision must remain in human hands. Blind trust in these systems represents a significant risk. During the discussion, a question was raised regarding how companies protect against over-trust, particularly in high-intensity roles where employees tend to take the answer and move on. Anna Tosky from Elmo described the phenomenon of "AI slop," where an employee is asked a question, inputs a prompt into the AI system, and sends out a completely erroneous 10-page report without looking at it. At Elmo, they make sure to keep a human in the loop on nearly everything, and Tosky argued that the solution starts with redesigning the work itself, instead of just throwing agents at the process.
The report recommends that IT leaders adopt a dedicated practice: stop treating log-ins as a proxy for trust, and track how often employees challenge or override AI outputs. Log-ins show that employees are clicking, whereas the override rate shows they are thinking and checking. Additionally, a concern was raised regarding contact centers: AI automates mundane tasks that previously served as cognitive breaks for agents, leaving them with nonstop, high-intensity work that leads to increased churn — a condition Tosky termed "AI brain fry." Wilson added that when AI absorbs routine tasks, employees lose the repetition that builds context and expertise, which creates a talent pipeline problem in the organization.
AI Agents and the Need for Securing Nonhuman Identities
Kerravala notes that AI that drafts content is one thing, but agents that act and take actions are another. Chris Locke stated that at Flight Centre, agents now manage a small subset of airline ticketing that was previously outsourced offshore, with security aspects integrated from day one. However, while he used AI to book an Uber ride to the conference, he noted that he would not yet let the system book a return flight to Europe.
According to the study's findings, fewer than half of respondents are very or extremely confident that their foundations are ready for more autonomous AI, and more than half indicated that their infrastructure needs upgrading. In addition, Cisco's 2026 Relentless Defense report found that only 36% of organizations in Australia apply consistent controls to AI agents and nonhuman identities. In the report, Devesh Maheshwari, Chief Technology Officer of Lendi Group, described the goal as "bounded autonomy," a state where the agent can act on the user's behalf but cannot exceed the boundaries established for it. Solder argued that trust must be enforced at the network and security levels, so that built-in control turns from a brake into an enabler of faster movement. Cisco is betting that its networking, security, and Splunk product portfolio will serve as the trust layer for agentic AI systems.
Five Lessons Presented by Kerravala for Technology Leaders
In concluding the column, Zeus Kerravala presents five practical lessons for IT leaders, based on the research findings and his conversations with executives:
- Make AI a leadership team project: Include the Chief People Officer and Chief Financial Officer alongside the CTO, as this is a cultural and business transformation and not just a technology project. Leaving the matter solely to the IT department leads to well-built platforms that employees do not use.
- Start small and show results: It is recommended to begin with visible, low-risk wins and publish the results, including error rates. The report's data showed that working three times faster with no increase in errors convinced skeptics more than any managerial mandate.
- Establish clear boundaries: Employees must know exactly what they can and cannot do with AI without needing to search through policy documents. Clear guardrails give employees confidence to experiment.
- Measure the right metrics: Stop relying on system log-ins as a sole metric, and track how often employees check, challenge, or override AI outputs, as a workforce that never pushes back against outputs is a warning sign.
- Fix infrastructure before scaling autonomy: Upgrade identity management, monitoring, and controls for AI agents prior to expanding their autonomy, as agents operate at machine speed and most organizations still do not apply consistent controls to nonhuman identities.
Kerravala concludes that the winners in AI will not necessarily be those with the best models, but those whose employees, leadership, and customers trust the systems enough to allow them to do more. As Solder put it: the leaders in the research are not scaling AI first; they are scaling trust first.