“We want AI doing real work for the business, safely.”
Agents that do the work, and ask before it matters.
Hand the routine work to agents that live inside the tools you already use. Your data stays put, every action is on record, and a person approves anything you could not undo. We build them, and we keep them running.
- For
- Owners, GMs and operations leads in NZ businesses
- Starts with
- Agent Readiness Review
- Then
- Secure Agent Builds, Managed Agents
- Priced
- Fixed scope, then monthly
- Your data
- Stays where you agree, NZ Privacy Act in view
We measured it before we sold it.
- the work of a team that size, from a small one
- 6 to 8×
- of what went live never had to be undone
- 99%+
- returned on every $1 spent on AI tools
- 20:1
- improvements live every week. Fewer than 1 in 5 companies ship that often.
- 10+
- of conventional work, delivered in ten months
- 6 years
Before we offered any of this, we measured it on a real engagement where we ran the technology side for a client. A small team took a product from a blank page to launch and got the work of a team six to eight times the size.
The gain did not come from the tools. It came from setting the rules first: what the AI was allowed to touch, how its work was checked, and who signed off. Then the speed was switched on. Every agent we build runs under the same discipline. The client, their industry and the product stay unnamed.
How we help
Three offers, in the order people take them. Every one is fixed scope, agreed before we start. Begin with the Review. It stands on its own, and you can stop there with a plan in hand.
- F1fixed scope
Agent Readiness Review
We look at how work actually moves through your business, where the hours go, and where the risk sits. You get a short list of the first agents worth having, and a straight answer about the ones that are not.
- F2fixed scope
Secure Agent Builds
An agent doing a real job in your CRM, accounting, inbox or documents. It only sees what it needs, every action is on record, and anything you could not undo waits for a person. Handed over working.
- F3monthly
Managed Agents
We keep it working as your business changes. Someone is watching, updating, and adding the next workflow when you are ready for it.
Product co-development
For partners who know their market and want a build partner with skin in the game. We build the product with you, from first architecture through launch and the months after, and share the outcome.
How the work runs
Four steps that close into a loop. The first stands on its own, so you can stop with a plan in hand and no obligation to go on. Once an agent is live, the loop turns monthly.
- Step 1Agent Readiness Review
Measure
Before we build anything, we find out where you actually are: where the hours go, what they cost, what is at risk. You get a plan you can act on with or without us.
- Step 2Secure Agent Builds
Build
Fixed scope, agreed before we start. No surprises on the invoice, and nothing goes live without the safeguards in place.
- Step 3Managed Agents
Run
We stay accountable after launch. Someone is watching, updating and keeping it working as the business changes, so you do not have to.
- Step 4Managed Agents
Learn
Each month we look at what the agent did, what it cost and what it saved, and pick the next job worth handing over. That goes straight back into the Review.
The rules every agent runs under
You do not have to take our word for it. Every agent we build works under these six rules, and you can see the log.
- Sees only what it needschecked
Access to one system at a time, scoped to the job. Nothing else is reachable.
- Every action on recordchecked
What it read, what it did, and why. Kept where you can read it.
- Irreversible actions wait for a personchecked
Paying money, deleting records, emailing customers. A human approves each one.
- Treats what it reads as untrustedchecked
A crafted email or document cannot talk it into doing something it should not.
- Data stays where you agreedchecked
Decided in writing before we build, with the NZ Privacy Act in view.
- Switched off in one stepchecked
If anything looks wrong, you or we stop it, and the business carries on.
Questions people ask us
Where does our data go?
Wherever you agree it can, and nowhere else. We decide this with you before anything is built. Options range from major model providers on terms that suit NZ businesses, through region-pinned deployments, to models running where your data already lives. Nothing is sent anywhere you have not agreed to in writing.
How do you keep agents safe?
Each agent sees only what it needs. Every action is on record. Anything you could not undo, such as paying money, deleting records or emailing customers, waits for a person to approve it. Anything the agent reads is treated as untrusted, so a crafted email cannot talk it into doing something it should not.
Which models do you use?
The one that fits the job and the risk. For most work that is a current model from a major provider. Where data cannot leave, or the task is narrow, we run private or smaller models. We are not tied to a vendor, and we change models when a better fit appears.
What if AI is not the right fit for us?
We tell you. Some problems are better solved with a report, a form or a well-placed rule. The Readiness Review exists to find that out before you spend on a build.
Why a small Dunedin studio and not a big firm or an AI agency?
Because you deal with the people doing the work, and they have spent about twenty years accountable for systems that handle money and people's data. That bar did not move when AI arrived. We stay after launch, we price by fixed scope, and if AI will not pay for itself in your business we say so.