Twelve years in AI and automation · Auckland, New Zealand 021 365 082 Taking new projects Book a free call
Home What I build Case studies FAQ Newsletter Contact Book a free call

Service · AI agents NZ

AI Agents for New Zealand Businesses

Custom AI agents that take a goal, work on it using your own systems, and stop for a person before anything that matters. Built with tight limits and full logging, at a fixed price, from Auckland.

DiscoveryFree, then fixed price
Typical build3 to 6 weeks
BasedAuckland, working NZ-wide

What an AI agent is

People say "AI agent" for three different things, and the difference decides what you should pay for.

  • Automation. Fixed rules: when X happens, do Y. No AI involved. Eg: a booking lands and a confirmation text goes out.
  • AI workflow. A path you design, with AI doing the reading and judging at certain steps. Eg: an email is read, the customer's history is pulled from the CRM, and a reply is drafted for one-click approval.
  • AI agent. A goal plus a set of tools, and the AI picks its own steps. Eg: "research this new lead and update the CRM", where it searches, reads and writes on its own.

What separates them is who controls the path. In automation and AI workflows, you do. With an agent, the AI does, and that drives reliability, cost and what it is safe to use for. The longer comparison is in automation vs AI workflow vs AI agent.

What I build

Everything below is in production or client testing for a New Zealand organisation. Some are agents in the strict sense and some are AI workflows doing an agent's job. Names stay private.

A company brain

Staff ask questions in plain English and get answers from a construction group's own project and tender files, each person seeing only the documents they can already open. Read-only by design, running in the client's own Azure tenant.

An AI operating layer

A company brain across email, calendar and documents for a music company, shipped in weekly releases with an in-app changelog.

Guest and customer assistants

A holiday park's assistant answers website chat and the main inbox from the park's own knowledge base, live availability and prices, and hands the rest to reception. It never changes a booking.

Voice assistants

A hands-free desktop assistant for a safety consultancy's field team: dictation into any window, and email read out and sent by voice with a read-back check.

Inbox agents

Supplier emails read as they arrive for a three-person distributor, purchase orders and shipping milestones logged, and carrier tracking pulled in automatically.

Document agents

Clearance report PDFs read by vision, billing fields extracted with a confidence score for each, matched to the job and written into simPRO. Doubtful fields are flagged, never guessed.

Agent or workflow: which do you need?

Most businesses that come to me asking for an AI agent actually need an AI workflow. They want the reliability of a process with the intelligence of a model inside it. Three questions sort it out:

  • Could I write the rule on a whiteboard? Then it's plain automation. Don't pay for AI to do an if-statement's job.
  • Does a person read something and make a routine call? That's an AI workflow. Eg: invoice processing, inbox triage, quote drafting, document review.
  • Do I want it to work out the approach itself? That's an agent. Ask whether you'd hand the task to a new hire with no supervision. If not, don't hand it to an agent unsupervised either.

Real systems mix all three. Mine usually use automation for the plumbing, AI for the judgement steps, and a tightly scoped agent for research-style jobs with a person checking the result.

Keeping an agent safe

An agent with broad access to live systems, acting unsupervised on high-stakes work, is the risky version. These are the limits I build in:

  • Scoped tools. It can only reach the systems and actions the job needs. Eg: it can read the CRM and draft a reply, but it can't send one.
  • Spending and step limits. A cap on how many steps it takes and what a task can cost, so a confused run stops instead of looping.
  • Every action logged. You can see what it read, what it did and why, and correct the source when an answer needs fixing.
  • A person signs off. Anything customer-facing or financial waits for a human before it goes out.
  • Read-only where possible. Eg: the construction company brain answers from the files but can't change them.
  • Tested on your real data first. In a test mode, so nothing touches live records until you've seen it work.

What it costs

AI assistants, agents and company brains run NZ$3,000 to $12,000 over 3 to 6 weeks on my pricing page, fixed before anything is built. Discovery is always free.

Agents cost more to run than workflows, because each task takes many AI calls instead of a few. The running costs guide works through the monthly bill in NZ dollars. Hosting, monitoring and support start at NZ$99 a month.

Frequently asked questions

What is an AI agent?

Software that is given a goal and a set of tools, and decides for itself which steps to take, in what order, and when it has finished. That is different from an AI workflow, where a person designs the path and the AI handles the judgement inside it.

How much does a custom AI agent cost in New Zealand?

Most of the AI assistants, agents and company brains I build cost NZ$3,000 to $12,000, fixed before anything is built, plus monthly hosting and AI usage. Discovery is always free. Prices exclude GST.

Are AI agents safe to use in a business?

Yes, for the right jobs with the right limits: scoped tools, spending and step limits, full logging, and a person checking before anything customer-facing or financial happens. What's risky is giving an agent broad access to live systems and letting it act unsupervised on high-stakes work.

Do I need an AI agent or an AI workflow?

Most businesses need an AI workflow: a fixed path with AI doing the reading and judging at certain steps. An agent suits open-ended jobs, like research, where an unexpected result is annoying, not expensive, and a person reviews the output.

Is ChatGPT an AI agent?

Out of the box, no. ChatGPT is a model you talk to, and it acts only when you prompt it. It becomes agent-like when it is given tools and a goal to pursue across several steps. The same models power automations, workflows and agents; the difference is what is built around them.

Who owns the agent and where does it run?

You do. The code, the accounts and the data are yours. By default it runs on AWS in Sydney in an account you control, or in your own Microsoft Azure tenant if that's where your business already lives.

Got a job you'd like to hand to an agent?

Book a free 30-minute call and describe it. I'll tell you whether it needs an agent, a workflow or a plain rule, and what it would cost.

Book a free 30-minute call