
By Joshua D. Hiwatig
The event brought together C-suite executives and business leaders from across New Zealand to discuss AI adoption, risks, opportunities, and the future of work through a series of panel discussions and fireside chats.
Much of what I heard at 6DAI Auckland was familiar: AI needs trusted data, strong governance, measurable ROI, and investment in people. These themes are becoming increasingly common across AI conferences and events.
What interested me more was the realisation that AI is evolving at a fast pace, while some organisations seem more focused on keeping up with the hype than building the foundations needed to use it well. That gap between rapid technological change and organisational readiness is where the real challenge, and opportunity, lies for leaders today.
Here are my seven key takeaways from the event:
A major theme throughout the event was that organisations are no longer asking only, “Can we use AI?” They are confronting the harder questions: How do we deploy it safely? How do we put it into production and scale it? And how do we generate measurable business value?
In my experience, moving AI into production exposes problems that pilots can hide: non-operational data sources, weak business cases, and unclear ownership and controls. During the pilot stage, organisations should give equal attention to potential operational challenges to avoid delays and technical debt once a pilot proves successful.
The bottleneck is shifting from technology capability to organisational readiness. This is where Mosaic’s Data by Design service line can be a differentiator: helping organisations establish the data foundations, governance, and operating frameworks needed to turn AI investments into sustainable data capabilities.
One of the strongest technological shifts discussed was the move from LLMs that recommend to agents that can act. This creates new types of risk that may not be fully addressed by existing governance frameworks and controls.
Some emerging risks include:
Governance designed for AI that recommends may not be enough for AI that acts. As agents gain greater autonomy, organisations need to rethink permissions, accountability, monitoring, escalation, and the ability to reverse unintended actions.
Governance was discussed repeatedly across the panels with a positive tone. Rather than slowing innovation, good governance gives organisations the confidence to move faster because leaders understand what AI can and cannot safely do.
For some organisations, governance still feels like a handbrake because it is treated as a checkpoint at the end of delivery rather than something built with the use case from the start. When AI initiatives encounter unclear risk boundaries, generic review processes, and manual compliance checks just before deployment, teams can end up revisiting decisions that should have been addressed much earlier.
The opportunity is to bring governance into the process earlier and make it easier to apply repeatedly. Instead of debating the same questions for every new AI use case, organisations can establish clear guardrails and approval pathways based on different levels of risk. This gives delivery teams greater certainty about the boundaries they can work within and helps successful AI initiatives move from experimentation to production with less friction.
AI is only as useful as the data, context, and permissions behind it. Raw data alone is not enough. AI requires business, system, customer, and process contexts. Without these, agents are just sophisticated guessers.
Organisations therefore need a trusted source that their AI systems can rely on. Building that trust means being able to answer some fundamental questions:
The dilemma is that these are not new questions. Many organisations have known about their data issues for years. What AI changes is the cost of continuing to ignore them. As AI becomes more embedded in business processes and decision-making, unresolved data issues can increasingly affect the quality and reliability of its outputs.
Realistically, organisations cannot fix every data problem at once. The opportunity is to prioritise fixing the data that the most valuable AI use cases depend on. Make ownership, quality, and access clear enough that both humans and AI can trust it. Then expand this foundation as more AI use cases move into production.
One panel introduced the concept of AI tokenomics, connecting what AI costs to run with the value it generates. Many organisations are pushing the frontier of AI innovation, but fewer have mature ways of measuring whether those investments are delivering sufficient value.
Token unit cost may have dropped significantly because of enhanced models but that doesn’t mean organisations will be paying less: organisations may use AI across more processes, agents may perform more reasoning steps, and systems may invoke more tools. Lower unit costs can therefore coexist with higher overall consumption.
The risk is that cheaper AI can create more AI consumption without necessarily creating more business value. Organisations therefore need to measure AI returns at the use-case level, connecting operating costs with outcomes such as productivity gains, revenue uplift, reduced processing time, improved customer outcomes, or avoided costs.
Despite the focus on automation, there was still a strong emphasis on the role of humans. Most speakers positioned AI as a tool to augment human capabilities rather than replace them.
But there is another side to this that I feel strongly about. Putting a human in the loop only works if that human is actually applying judgement. As AI-generated content becomes easier and faster to produce, there is always a temptation to accept outputs at face value, make minor edits, and move on. The result is often low-quality work that may look polished but lacks accuracy, context, originality, or critical thinking.
Organisations need to set a higher bar for what “human-in-the-loop” actually means. Start by identifying where AI-generated work requires mandatory human review, particularly where outputs affect customers, financial decisions, or regulatory obligations. Then make the reviewer’s responsibility explicit: validate important facts, use critical thinking, apply business context, and take ownership of the final output.
Teaching employees how to prompt effectively is not enough. They also need to learn how to recognise weak AI outputs, verify the source, and know when not to trust the answer.
Technology capability is advancing faster than many organisations’ ability to absorb it. Several speakers suggested that the limiting factor is increasingly not what AI can do, but whether people, processes, and organisational structures can change quickly enough to take advantage of its benefits.
AI enablement training therefore cannot be optional. Roles will evolve as AI takes on more manual work. Technology, security, data, risk, and business teams need to move and upskill together rather than at different speeds to move projects forward.
AI transformation therefore requires two roadmaps working in parallel: a technical roadmap for the technology and an adoption roadmap for the people and organisational change needed to make it work.
Perhaps the biggest takeaway for me is that AI readiness is becoming less about access to technology and more about an organisation’s ability to adapt around it. The technology will continue to improve, costs will change, and new capabilities will emerge. But many of the harder challenges remain organisational: getting the data in order, establishing clear accountability, putting the right guardrails in place, measuring value, and preparing people to work differently.
The organisations that get this right will not necessarily be those that adopt every new AI capability first. They will be the ones that know where AI can genuinely create value, have the foundations to support it, and can scale it safely and economically. Perhaps that is what moving beyond the AI hype really means.
Move beyond the AI hype. Let’s chat today.

By Joshua D. Hiwatig
The event brought together C-suite executives and business leaders from across New Zealand to discuss AI adoption, risks, opportunities, and the future of work through a series of panel discussions and fireside chats.
Much of what I heard at 6DAI Auckland was familiar: AI needs trusted data, strong governance, measurable ROI, and investment in people. These themes are becoming increasingly common across AI conferences and events.
What interested me more was the realisation that AI is evolving at a fast pace, while some organisations seem more focused on keeping up with the hype than building the foundations needed to use it well. That gap between rapid technological change and organisational readiness is where the real challenge, and opportunity, lies for leaders today.
Here are my seven key takeaways from the event:
A major theme throughout the event was that organisations are no longer asking only, “Can we use AI?” They are confronting the harder questions: How do we deploy it safely? How do we put it into production and scale it? And how do we generate measurable business value?
In my experience, moving AI into production exposes problems that pilots can hide: non-operational data sources, weak business cases, and unclear ownership and controls. During the pilot stage, organisations should give equal attention to potential operational challenges to avoid delays and technical debt once a pilot proves successful.
The bottleneck is shifting from technology capability to organisational readiness. This is where Mosaic’s Data by Design service line can be a differentiator: helping organisations establish the data foundations, governance, and operating frameworks needed to turn AI investments into sustainable data capabilities.
One of the strongest technological shifts discussed was the move from LLMs that recommend to agents that can act. This creates new types of risk that may not be fully addressed by existing governance frameworks and controls.
Some emerging risks include:
Governance designed for AI that recommends may not be enough for AI that acts. As agents gain greater autonomy, organisations need to rethink permissions, accountability, monitoring, escalation, and the ability to reverse unintended actions.
Governance was discussed repeatedly across the panels with a positive tone. Rather than slowing innovation, good governance gives organisations the confidence to move faster because leaders understand what AI can and cannot safely do.
For some organisations, governance still feels like a handbrake because it is treated as a checkpoint at the end of delivery rather than something built with the use case from the start. When AI initiatives encounter unclear risk boundaries, generic review processes, and manual compliance checks just before deployment, teams can end up revisiting decisions that should have been addressed much earlier.
The opportunity is to bring governance into the process earlier and make it easier to apply repeatedly. Instead of debating the same questions for every new AI use case, organisations can establish clear guardrails and approval pathways based on different levels of risk. This gives delivery teams greater certainty about the boundaries they can work within and helps successful AI initiatives move from experimentation to production with less friction.
AI is only as useful as the data, context, and permissions behind it. Raw data alone is not enough. AI requires business, system, customer, and process contexts. Without these, agents are just sophisticated guessers.
Organisations therefore need a trusted source that their AI systems can rely on. Building that trust means being able to answer some fundamental questions:
The dilemma is that these are not new questions. Many organisations have known about their data issues for years. What AI changes is the cost of continuing to ignore them. As AI becomes more embedded in business processes and decision-making, unresolved data issues can increasingly affect the quality and reliability of its outputs.
Realistically, organisations cannot fix every data problem at once. The opportunity is to prioritise fixing the data that the most valuable AI use cases depend on. Make ownership, quality, and access clear enough that both humans and AI can trust it. Then expand this foundation as more AI use cases move into production.
One panel introduced the concept of AI tokenomics, connecting what AI costs to run with the value it generates. Many organisations are pushing the frontier of AI innovation, but fewer have mature ways of measuring whether those investments are delivering sufficient value.
Token unit cost may have dropped significantly because of enhanced models but that doesn’t mean organisations will be paying less: organisations may use AI across more processes, agents may perform more reasoning steps, and systems may invoke more tools. Lower unit costs can therefore coexist with higher overall consumption.
The risk is that cheaper AI can create more AI consumption without necessarily creating more business value. Organisations therefore need to measure AI returns at the use-case level, connecting operating costs with outcomes such as productivity gains, revenue uplift, reduced processing time, improved customer outcomes, or avoided costs.
Despite the focus on automation, there was still a strong emphasis on the role of humans. Most speakers positioned AI as a tool to augment human capabilities rather than replace them.
But there is another side to this that I feel strongly about. Putting a human in the loop only works if that human is actually applying judgement. As AI-generated content becomes easier and faster to produce, there is always a temptation to accept outputs at face value, make minor edits, and move on. The result is often low-quality work that may look polished but lacks accuracy, context, originality, or critical thinking.
Organisations need to set a higher bar for what “human-in-the-loop” actually means. Start by identifying where AI-generated work requires mandatory human review, particularly where outputs affect customers, financial decisions, or regulatory obligations. Then make the reviewer’s responsibility explicit: validate important facts, use critical thinking, apply business context, and take ownership of the final output.
Teaching employees how to prompt effectively is not enough. They also need to learn how to recognise weak AI outputs, verify the source, and know when not to trust the answer.
Technology capability is advancing faster than many organisations’ ability to absorb it. Several speakers suggested that the limiting factor is increasingly not what AI can do, but whether people, processes, and organisational structures can change quickly enough to take advantage of its benefits.
AI enablement training therefore cannot be optional. Roles will evolve as AI takes on more manual work. Technology, security, data, risk, and business teams need to move and upskill together rather than at different speeds to move projects forward.
AI transformation therefore requires two roadmaps working in parallel: a technical roadmap for the technology and an adoption roadmap for the people and organisational change needed to make it work.
Perhaps the biggest takeaway for me is that AI readiness is becoming less about access to technology and more about an organisation’s ability to adapt around it. The technology will continue to improve, costs will change, and new capabilities will emerge. But many of the harder challenges remain organisational: getting the data in order, establishing clear accountability, putting the right guardrails in place, measuring value, and preparing people to work differently.
The organisations that get this right will not necessarily be those that adopt every new AI capability first. They will be the ones that know where AI can genuinely create value, have the foundations to support it, and can scale it safely and economically. Perhaps that is what moving beyond the AI hype really means.
Move beyond the AI hype. Let’s chat today.