Is a public sector this bad at big complicated things really going to thrive with AI?

Wednesday, 17 June 2026

Henry Cooke is The Post’s deputy political editor.

OPINION: Some things in life really are complicated.

Building one system to hold and check the biometric details of everyone who comes into contact with our immigration system is a complex task. It is not something you could readily expect existing bureaucrats to design from scratch or an IT firm to hack together over a few months.

But the state is supposed to be able to do complicated things.

And yet reading the review of Immigration New Zealand’s (INZ) attempt to build that very biometric system feels like reading the detailed account of a nightmare.

The 170 changes to the initial contract. The 12 different project managers at the tip of a battalion worth of staff turnover, with some people reportedly leaving because they raised concerns about the project. The briefing to the minister asking for more money that didn’t mention her predecessor had rejected this bid and said a Treasury review had approved of the project when it had actually lashed it.

Above all, the $33 million and seven years that resulted in zilch.

Read more:

• Watchdog to investigate after $33m immigration tech project collapses, delivering nothing

• Government backs transport funding review in response to national infrastructure plan

• Probe into $1.3b national ticketing project finds delayed timeline still ‘ambitious’

It is a slow spiral into bureaucratic hell.

One of the reasons it reads so clearly as a nightmare is because it is not that far from the experience anyone involved in a failed project has had. A slipped deadline, a crucial staff-member leaving, a sense that those who are the most invested in the project are the ones who understand its difficulty the least - this is not alien stuff. Yet where some of us may have experience in slowly putting down a bad idea over a few meetings, this failed project just never ended.

It got bigger, it got more expensive, it got messier, and it never dealt with the initial problems in how it was scoped and designed. No one involved seemed able to wake up - except by just quitting, and passing the buck to someone else.

Unfortunately, it is far from the first report of its kind.

New Zealand’s public sector has a way of bungling big IT projects. The “National Ticketing Solution” for public transport has been delayed so often that Wellington and Auckland have moved on without it. The word “Novopay” still sends a shiver down the backs of Ministry of Education lifers. And the $150m Department of Internal Affairs (DIA) project to put births, deaths and marriages information in the cloud had to be halted and eventually terminated altogether as it was so off track.

At a smaller scale there is the inexplicable $600k website for Wellington library that isn’t even really a website for Wellington library.

It is worth remembering that parts of the public sector are okay at procuring big complicated things. IRD’s gigantic change project came in under-budget and over-delivered, with a substantial reduction in call centre staff and an ability to deliver many people automated tax returns they trust, and complex cost of living upgrades to Working For Families. RealMe is a successful and useful project.

Infrastructure Minister Chris Bishop sees the “disaster” of Dunedin Hospital as a prime example of how big complicated projects go wrong. He says politicians committed to the project without deciding how to build it, no real business case was ever really developed, and then officials failed to escalate multiple red flags to ministers.

In other words - leadership makes decisions that they are five pay-grades above having to actually implement, and several levels of experience below really understanding. There is then too much political buy-in and momentum behind the project to even consider cancelling it or re-scoping it to be anything other than bigger. Those close to the project try to either leave without being tainted, deliver on a timeline that is impossible, or raise concerns that are ignored.

Given all that, it’s hard not to wonder just how ready big parts of the public sector are to cut a cumulative 10% of their spending over the next two years, as the Government’s budget is reliant on them doing, with AI replacing and exceeding currently-available services. Sir Brian Roche says the public service is buying into this and has taken on the role of heading up the Government Digital Delivery Agency.

Roche is an impressive and experienced man. But is he really close enough to frontier AI systems to know what they are good at and what they are bad at and how they might interact with the Privacy Act? Has he really considered the capability gap between the CoPilot most of the public sector can access and the tools like Claude Code that most of them can’t? Is he ready to let some 25-year-old burn through $150,000 in tokens working out whether a public sector pain point can be solved easily?

Perhaps he is. Perhaps a gigantic revamp of the public service is possible in two years to meet an election-year budget commitment. But it will definitely be complicated.