Wellington, New Zealand
AI development for Wellington teams, handed over in writing
Magic Ship is an AI development agency in Vancouver, British Columbia. Wellington and Vancouver share no working hours at all, and that is stated up front because it changes how the work runs: written handoff at the end of each side's day, a loop that turns once every twenty-four hours, and nobody waiting on a meeting.
- base
- vancouver, bc, canada (UTC-08:00)
- market
- Wellington, New Zealand
- your clock
- UTC+12:00 (Pacific/Auckland)
- overlap
- none on the same date, written handoff instead
- presence
- remote only, no office in Wellington
No shared working day, so the loop is written
Wellington is 20 hours ahead of Vancouver. When we start at 09:00 it is 05:00 on the next day for you, and when we stop at 17:00 it is 13:00 on the next day. On the same calendar date there is no hour where both sides are at work.
It also produces a better record than a colocated project would. Everything of consequence exists as written text with a date on it, because writing is the only channel available.
- A handoff note goes out at the end of every Vancouver working day: what shipped, what broke, and what needs a decision from you.
- You read it at 13:00, or whenever your day starts, and review the pull requests it points at.
- Decisions are logged as they are made, in the repository, so neither side has to reconstruct why something was done.
- A blocking question arrives with a proposed answer attached, so the default outcome is progress rather than a stalled day.
One live call a week, and it is scheduled honestly: 13:00 in Vancouver is 09:00 in Wellington on the next day. Neither side takes a bad hour for that one. The call simply lands on two different calendar dates, which is a scheduling detail rather than a hardship.
What Wellington buyers usually bring us
The record for Wellington lists two industries. Here is the part of each one that turns into an engagement, and the service line it lands under.
public administration
Legislation, guidance and internal procedure answer most questions, provided retrieval quotes the passage and shows which version it came from.
Hybrid search beats embeddings alone in nearly every corpus we have opened, and metadata filters behave as hard constraints or they are not filters.
Closest service line: Knowledge bases and retrieval
film production
Scripts, schedules and rights records are the working data. A tool that reads them is useful; a model that guesses at them is a liability.
The eval harness gets built beside the first feature, so week three can prove that week two did not quietly make the output worse.
Closest service line: AI products, built end to end
Contracts, currency and where the data sits
Invoices are in New Zealand or US dollars. The date offset is large enough that every deadline in the contract carries the timezone it is measured in.
The commercial part stays deliberately boring. Fixed acceptance criteria per increment, weekly shipping, and a written record of every decision. Nothing about working from Vancouver instead of Wellington changes what you are buying.
The five lines we sell
Same five lines everywhere, Wellington included. Most engagements are one of them, sometimes two.
- AI products, built end to end
- Knowledge bases and retrieval
- Codebase and team optimization
- Training and enablement
- Forward-deployed engineering
The languages, models and infrastructure underneath all five are listed on the Magic Ship stack page, with what each one is good for and where it is the wrong choice.
Other markets in Oceania
Where else the same work happens in Oceania, with the numbers redone per market.
Start a project
Describe what is broken or what you want built. The reply comes from the engineer who would build it, and it covers scope, timeline and fit before anything else.