Aotearoa AI Summit 2026: Who Owns Your AI, and Is It Working?

Someone in your business built an AI agent; it is talking to HR systems, customer records and finance. But who is watering and feeding it? Who is looking after that agent after it starts exploring? Who keeps it current when systems around it are changing? And most importantly, who is responsible for the decisions it makes?

A CTO asked those same questions at the Aotearoa AI Summit in Wellington, and the room went quiet. Nobody had a clear answer over the two days at the summit.

Why we were in the room

Enlighten Designs sponsored the Best Implementation of AI in Aotearoa award category, where the four finalists were announced at the summit. Enlighten Designs Business Development Manager, Clayton Knox spent both days across keynotes, panels and roundtables on government policy, infrastructure, workforce change and New Zealand’s export opportunity.

We have been using AI across our own development work for the last few years, so we wanted to understand what was stopping people from using AI in their processes. Clayton’s answer after attending the summit wasn't what we expected. The licenses were bought. The policy has been drafted. Someone in the business has already built an agent. But the key question was understanding who owns the AI’s decisions and who is responsible for it when it is wrong.

Our Key Takeaways

New Zealand is falling behind

Adopting generative AI is estimated to add $76 billion to the New Zealand economy by 2038, which is roughly 15 percent of national GDP. That estimate comes from New Zealand's Strategy for Artificial Intelligence. One keynote speaker went further by reframing AI as “New Zealand’s next export economy” built onshore and earned offshore. New Zealand is not moving fast enough to get there. The pace of Kiwi businesses using AI drew the most criticism across both days. Organisations have started using AI, but comparable countries are well ahead, and that gap is widening.

Most organisations are at the Copilot stage, using AI at chat level to draft emails and summarise documents. A smaller group has moved up a level and is building agents to run whole processes.

Understanding how to ‘catch up’ split the room. One political voice argued that the light-touch regulation might not be the right approach for a country moving this slowly. This regulation relies on existing laws, voluntary guidance, and minimal government oversight rather than strict rules. He argued that AI oversight should exist within the government. Suggesting that the government should provide leadership and coordination across the development of AI adoption and governance.

Industry voices pushed the other way. Their argument was that New Zealand has spent years debating governance and ethics. A keynote speaker explained how we are three and a half years in, New Zealand is still having conversations it was having in 2023. In their view, existing laws such as the Privacy Act already cover how businesses use AI, and waiting for new rules only adds delay.

Either way, the tools are no longer the problem. Whether the push comes from government or industry, someone still has to take responsibility for the AI a business puts to work.

Every agent needs an owner

Who is responsible for the decisions an agent makes? A CTO asked this question, but no one could answer who takes ownership of an AI's decisions. Across businesses, people are building their own agents, and IT often doesn’t know they exist. That’s shadow AI, where no one keeps agents updated as the systems around them change and hold them to architectural or security standards. Speakers were clear that the technology can’t take the blame. One panellist touched on accountability and stated how you are the adult in the room; you should be accountable.

Adding ownership to an agent after it's built is harder than designing it in. Before an agent goes live, someone needs to decide where it can act, which decisions stay with a person, and who answers for it. An agent that drafts an email for review is a very different situation from the one that can send it. Deciding where to draw that boundary is where most organisations get stuck.

Most people don’t know which guardrail they need

The stalling point is consistent. Teams can't work out which admin-heavy tasks to hand over, or how to hand them over with guardrails they trust.

Agents can be built with guardrails, but only once someone works out what they need to protect against. That might be legal exposure, policy breaches, and the wrong data ending up where it shouldn't be. Those risks get identified in the planning stage, or they get discovered in production. Leaders need to know what their agents can reach before anything runs.

Part of the problem is where teams start. Multiple sessions rejected the “AI on it” approach of starting with technology in favour of starting with the problem and people doing the work. When teams start there, it becomes clearer what an agent should and shouldn't be interacting with.

Human in the loop

Underneath the hesitation about digging deeper in AI sits a fear: AI means automation, automation means fewer jobs. The message from the summit was consistent. AI is an enhancer, not a replacement. This is where the judgement gap surfaces. A person making a decision brings empathy and context to someone's situation, and AI can’t do that on its own. A speaker shared how they keep a person in the loop of all its AI work for this reason, using AI to support people's judgement rather than replace it.

The Human Rights Commission and the Kāhui Māori table, pushed this point further. People, identity and rights need to be at the centre of AI adoption, not added in as an afterthought. They noted that only a third of New Zealanders trust AI, arguing that human rights aren't a barrier to innovation, they’re what make it trustworthy. This means bringing people in as designers and collaborators from the start, not as consumers. A broadcaster showed what this looks like in practice. Its AI system for tagging video content opens by prompting the user to ask the questions a tool can't answer, such as who is missing from a clip and how a whānau is being represented.

Businesses can't prove AI is bringing them value

Putting humans in the centre of AI plays a big role in getting AI right, but proving it delivers is an important part. Many organisations have tried AI in a small way by getting one team or one process to use it as a trial run. That trial period is called a pilot. The problem is that organsiations can’t measure whether it actually worked. They can point to the trail but not to a business result.

It was stated at a panel that around 60% of New Zealand organisations were cited as AI ready in principle but were stuck rather than progressing.

Speakers were honest that getting it wrong is part of the process. Cost blowouts and false starts are lessons to learn from, not reasons to stop, as long as you're measuring what happened.

In many organisations, governance, permissions and ownership haven't caught up with what staff have already built. That gap is why so many capable businesses are sitting still.

The conversations from the two days at the summit came back to the same points. New Zealand organisations have the AI tools, but what is missing is the structure around them. Understanding those security measures, taking responsibility for the decisions it makes, and proving that it is actually making a positive impact on business ROI.

None of that is theoretical for us. We have been running AI in our delivery team for years. Which means a person reviews everything it produces, and we measured our own delivery data to check if it was making a difference.

AI usage at Enlighten Designs

How we keep a person accountable

Agentic development is part of our baseline delivery model; every developer works with coding agents, several run multiple in parallel, and the team is trained continuously. What does not change is who is accountable. The agent drafts and scaffolds. A person reviews, tests and owns what ships. Our QA team wrote about the groundwork that makes that possible, most of which happens before a single test case exists

We also checked whether it was working. Rather than assuming the AI was working, our Principal Technologist in Data and AI, Verne Roberts, runs a structured study across our own delivery data and publishes what he finds in our blogs.

Enlighten Designs makes it a priority to track and measure our performance with AI in the way we work. We collect data and run our own regular reports for AI productivity in our team to understand the real impact and influence AI has on our work. We do this by having our developer's flag whenever they use AI on a task in our time-sheeting system, so we can compare AI-assisted work with non-AI work against real projects.

Check out our recent findings in our three-part series blog, covering our AI productivity report from February. More of our findings will be coming soon.

Where to start

The hardest part isn't building the agent. It's deciding which process to hand over, proving it works, and being able to say who owns the decisions it makes.

That's the work our AI team does. We help you find the process worth going first and build it with the permissions your organisation needs.