Utilising AI is one of the most important skills for business owners, employees and new graduates, and universities are struggling to keep up. I set up Young Kiwis in AI so young people could share what they are learning. For this session, two young AI engineers working at large NZ companies talked through their journeys, what they got wrong and their best tips. Here is what they said, rewritten as an article.
How did these young Kiwis get into AI engineering?
Neither took a straight path. The first panellist, started a Bachelor of Science in computer science at Canterbury in 2020, in the middle of COVID. A university adviser told her she was not smart enough to work in AI, so she switched to a Bachelor of Commerce in information systems with a maths minor. She then landed a cloud engineering internship at a large NZ company.
She had always been interested in AI. Back in 2020 a friend used AI-assisted code completion and they both thought it was amazing. When a job opened up in a new internal AI engineering team, she applied and got one of the roles. Her advice to her past self: you are capable of anything you put your mind to, even if someone tells you otherwise.
The second panellist, learned to code in high school, then studied engineering science, which is more maths and algorithms. After several data internships he tried working as a stockbroker, did not enjoy it, and got drawn to AI safety. Learning how models are trained and starting to vibe code led him to a junior AI integration engineer role. He had been in it about six months, three of them as an intern.
Do you need to know everything about AI before you apply?
No. The second panellist had done very little in AI automations or workflows before he applied. He did not expect to get the job. His interest came across in the interview, and a bit of vibe coding was enough to get him through the practical part. His takeaway: never let not knowing everything about AI be a blocker.
The first panellist said the same about skills. Soft skills and work ethic came from university, but she learned the actual technical skills on the job because tertiary education has not quite caught up. To see what employers look for, read what employers hire for in Auckland.
What does an AI engineer actually do day to day?
The first panellist works a 40-hour week, in the office at least two days. A normal week is heads-down focus time, a morning stand-up for blockers, reviewing and shipping code, and working with developers, product managers, product owners and designers. Her company gives staff 10% of their time for learning. Most of her time is now spent architecting solutions with Claude Code rather than writing code line by line.
The second panellist describes his job as roughly 50% building and 50% people. The building side is vibe coding, planning apps for teams and building MCP servers so people can connect to their data. The people side is working with non-technical teams, such as marketing and commercial, to understand their pain points and where a human should stay in the loop.
His team runs on the motto “fire bullets first, then cannonballs.” Ideas come in from across the business, and the team spends 3 to 5 days proving an idea has value or killing it. Good ones go to a full development team. Another big part of his job is automating himself: when he does something more than three times, he asks whether it can be automated. Driving adoption of Claude across the business is part of it too. For the wider picture, read what I do as an AI engineer in New Zealand.
What tools do young AI engineers use?
The second panellist named Cursor and Claude for coding and planning, Airtable for structured data (its API and connectors work well with AI), Fly.io for hosting simple sites with more control than a tool like Lovable, and Terraform for spinning up cloud infrastructure quickly.
The first panellist shared a favourite hack: a Claude skill that runs a diagnostic on her existing Python knowledge, including the code she has written and reviewed. It then feeds a scheduled task that sends her a Slack message each day to quiz her, and her replies tell her whether she got it right. Her point is that people forget how much of their own learning they can automate.
The second panellist also built a skill that turns a submitted idea into a ready-to-build project. It pulls in the meeting transcript, any documentation and his preferred way of setting up a code base, so he can start building a proof of concept straight away. Projects he mentioned include a copy-checking tool for retailer listings, an ingredients compliance checker, and a knowledge assistant that searches company documents. If you are choosing a Claude plan to start with, see Claude pricing in NZ.
What mistakes do new AI engineers make?
- Trying to keep up with everything. the first panellist admits that as a perfectionist she tried to watch every video and read every article. It is not realistic. Give yourself time to absorb what you learn and do not beat yourself up for missing a model or trend.
- Ignoring cost. the second panellist accidentally spent around $4,000 on a data platform in his first few months. AI tools make it easy to run something many times without thinking about cost. Build dashboards and cost reporting into your tools.
- Not measuring value. His manager kept asking for numbers: how many people used a tool and what value it created. Managers need figures, not just a working tool.
- Explaining instead of exciting. People believing a tool works is different from being excited enough to use it. A visual, relevant demo beats a long technical explanation.
The second panellist also keeps a running list of hundreds of mistakes and learnings, because AI is new to everyone.
What are the best tips for getting into AI?
The first panellist’s three tips:
- Pick one thing at a time. Do a shallow bit of research, choose one tool and learn it well. The skills transfer, and many tools are the same thing in a different font.
- Start from a problem you actually have. It does not need to be work. It could be fitting a physio plan around your gym goals. Use that as a passion project.
- Do not wait for the perfect time. Many people told her starting was their biggest barrier, and once they started it all followed.
The second panellist’s tips, framed as what would impress an interviewer beyond good grades:
- Start by coding. Pick a project you care about, open Claude Code or Cursor and build something from scratch. You learn more from a project you are interested in.
- Keep up with AI news. He impressed his boss by knowing recent models and where the industry is heading. His sources: the AI Daily Brief podcast, the Epoch newsletter, the Dwarkesh podcast and the AI Frontiers newsletter.
- Think in systems, not code. Anyone who can draw a good systems diagram or flowchart can now get a coding tool to build it. Not everyone who can code can think about how a solution fits with people and where the human in the loop should be.
Is AI engineering becoming more about architecture than coding?
The first panellist thinks so. In a small team she is making architecture decisions that a year or two ago would have come at senior or tech lead level, because AI tooling like Claude has moved her key work from developing to architecting. She says other engineers are starting to see the same shift. It is a good reason to build judgement about systems and people early, not just syntax.
