AI development companies in New Zealand.

An AI development company designs and builds software in which a model does part of the work: an agent that handles routine admin, a search tool that answers from your documents, or a pipeline that posts supplier invoices to Xero. Providers range from national IT firms to small studios. This guide sets what each says about itself side by side.

If you are still deciding what to build, start with our guide to AI consultants in New Zealand, or check whether a product would do the job with custom AI versus off-the-shelf tools.

New Zealand AI development companies compared

“Published proof” means named clients, case studies or products the firm publishes itself. Prices are as listed on 20 October 2026.

Provider profiles

Levy Street

Levy Street is a Wellington AI studio and engineering company that designs, builds and runs production AI systems. Our custom AI development starts with real examples and acceptance criteria you sign off, proves value with a narrow production-grade slice, then adds monitoring and a runbook. Published work includes an internal agent for APL Property and PainPal, on which Levy Street worked with BetterWorx. We publish no prices.

Avtrix

Avtrix is a Taupō company that builds AI agents, automation and custom software at fixed prices. Its pricing page lists AI agents and document processing at NZ$8,000 to NZ$35,000 excluding GST, indicative until scoped, and its case studies do not name clients.

Catalyst IT

Catalyst IT is an open-source technology company headquartered in Wellington. Its AI page offers bespoke AI development, consulting and managed services on NZ-owned Catalyst Cloud, so data can stay in Aotearoa. A case study describes the platform it built for NZCER’s PAT Tuhituhi writing assessment, whose automated marking NZCER developed with Vantage Labs.

Datacom

Datacom is a technology company with offices across New Zealand. Its AI page says its AI consultants develop governance frameworks, proofs of concept and AI tools for enterprise organisations and government departments, and names a payroll AI assistant for Datapay among its examples.

Do AI

Do AI is a Wellington AI product studio for founders and businesses. Its build service runs fixed-scope milestone projects typically from $20K, with the client owning the code, and its Easy Read case study describes an app that converts documents into Easy Read formats.

ElementX

ElementX is an Auckland applied AI engineering consultancy, founded in 2013, with a Wellington base. It builds enterprise AI for healthcare, insurance, higher education and government, names the University of Auckland, Mott MacDonald and Southern Cross as customers, and publishes a case study on the university’s AI Assistant.

Enlighten Designs

Enlighten Designs is a Hamilton software, web and data company whose AI development page covers Azure OpenAI applications, computer vision, agentic AI and document processing. In an October 2026 press release it announced it had been named an OpenAI Select Partner. Its published work is mostly web and software.

Harkness AI

Harkness AI is a Christchurch company founded in 2025 that offers AI training, consulting, custom builds and an embedded AI engineer for businesses of roughly 20 to 100 staff. Its custom AI page names Paagaa, an open banking budgeting app, among its builds.

Optimation

Optimation is a New Zealand technology consultancy with offices in Wellington and Auckland that focuses on regulated industries. Its AI page offers enterprise AI, IT consulting and custom software, starting each engagement with discovery.

Theta

Theta is a New Zealand-owned technology consultancy founded in 1995. Its generative AI page says it designs and builds AI agents and offers Theta Assist, its own assistant platform. Its GenAI Rapid Start tests one prototype in four to six weeks for $25,000 plus GST, which Theta says is not a production implementation.

How to test a firm’s build capability

Choose the firm that can show you a system it built that is still running, and the tests that prove it works. Send every shortlisted provider these questions, us included, and compare written answers. Our AI vendor evaluation checklist helps score the answers, and you can send the questions to us too.

• Show me an AI system you built that is in production, and who uses it.
• What number did it move, and how was that measured?
• Show me a past evaluation set: real inputs, agreed correct outputs, a pass threshold.
• Will the people in this meeting write the code?
• Where will our data be processed and stored, and under whose terms?
• What do we own at the end: code, prompts, evaluation set, cloud accounts?
• What happens when the model is wrong: review queue, logging, rollback?
• Who runs it after launch, and what will it cost each month?

Weight the answers by project type.

Where AI development projects go wrong

• The demo never meets real data. A prototype tuned on ten clean examples fails on the awkward ones. Test a paid thin slice on your own inputs.
• No acceptance test. Without an agreed evaluation set, “done” becomes an argument. Put the pass threshold in the contract.
• The vendor keeps the keys. If the provider holds the code, prompts and model accounts, you are renting. Get ownership and a runbook in writing.
• Write access without approvals. An agent that sends emails or posts to Xero needs a person approving its actions until its error rate is known.
• Website ranges treated as quotes. Published prices are often indicative and exclude GST. Compare scoped written quotes that include monthly running costs.

New Zealand rules and funding that affect a build

• Privacy Act 2020. The information privacy principles apply to personal information your system collects, uses or sends offshore. Since 1 May 2026, IPP 3A requires reasonable steps to tell people when you collect their information from someone else, which catches enrichment, scraping and bought-in data.
• MBIE guidance. MBIE’s voluntary Responsible AI Guidance for Businesses calls for a person in the loop on important decisions, and notes that New Zealand servers may better support data sovereignty.
• The AI Advisory Pilot. The government pilot co-funds up to 50% of eligible costs through an approved provider: up to $2,500 for an AI roadmap and $12,500 for implementation, excluding GST. It does not fund developing AI products to sell or general software development, and runs until 30 June 2027.
• GST and currency. Check whether prices include GST and whether model fees are billed in US dollars.

Written by the Levy Street team in Wellington. Last updated 19 October 2026.