Google AI specialists
The Google AI stack, mapped honestly.
Google ships a lot of AI products. Most businesses need four of them. Here is what each layer actually does, what we build with it, and where it stops being the right tool.
Fixed scope, fixed price. No licence commission.
The stack, layer by layer
Names in Google’s stack move fast — Vertex AI became the Gemini Enterprise Agent Platform in 2026. This page is reviewed quarterly; the mapping below is how we actually build.
| Layer | Google product | What we build with it | When it is the wrong tool |
|---|---|---|---|
| Staff assistant | Gemini app (Google AI Pro / Ultra) | Day-to-day drafting, research and analysis for individual staff | Regulated data that cannot leave your tenant |
| Inside Workspace | Gemini in Gmail, Docs, Sheets and Meet | Drafting, summarising and reporting inside the tools staff already use | The workflow needs custom logic or access to other systems |
| Company knowledge | Gemini Enterprise | Private assistants over your own documents, with permissions that match Drive | Fewer than about ten users — consumer Gemini is enough |
| Agent platform | Gemini Enterprise Agent Platform (formerly Vertex AI) | Building, deploying and governing agents; retrieval over your data; model choice | You need on-premise or multi-cloud portability |
| Custom build | Gemini API, AI Studio, Antigravity | Prototypes, custom applications and agentic development | The job is a spreadsheet macro |
| Data and grounding | BigQuery, Cloud Storage | Grounding answers in your own numbers instead of the open web | Your data already lives in a SaaS you can query directly |
| Run and monitor | Cloud Run, Cloud Scheduler, Cloud Monitoring | Hosting and operating what we build, with alerting | You already run this in-house — we will still monitor it |
What we implement
Workspace automation
Inbox triage, quote and invoice drafting, reporting packs, document generation, meeting notes into actions.
Gemini in WorkspaceApps Script
Private assistants
An assistant that answers from your own documents with permissions that mirror Drive.
Gemini EnterpriseAgent Platform
Custom agents
Gemini API for the model, AI Studio to prototype, Antigravity and Cloud Run to build and host, BigQuery to ground it.
AI StudioCloud Run
AI-ready infrastructure
The unglamorous half: servers, containers, networks, monitoring and backups that keep the AI layer running when nobody is watching.
Cloud RunMonitoring
How a Google-first engagement runs
Assess (week 1)
Free. We map your systems, data and bottlenecks, then write down which Google products fit and which do not.
Prove (weeks 2–4)
The Google AI Challenge, or a small paid pilot on one workflow. Success criteria agreed in writing.
Build (weeks 4–8)
Fixed-scope delivery: the automation, the assistant, the integrations, the access controls.
Run
Monitoring, alerting, self-healing and a monthly review. Managed from NZ$950 per month.
Where Google is the wrong answer
A specialist knows the edges of the platform. If your job sits in one of these, we tell you before you pay.
- Air-gapped or fully on-premise. Google’s AI is a cloud service. If the data legally cannot leave your building, the answer is a local model, not Gemini.
- Deep in Microsoft 365 or AWS. If your estate is Entra, Exchange and Azure, the integration cost usually beats the licence saving. We say so.
- Specialist media pipelines. Frame-accurate video work, voice cloning at scale or unusual 3D pipelines usually belong with a dedicated tool.
- Narrow, high-volume classification. At millions of calls, a small fine-tuned model can cost a fraction of any frontier API.
Specialists, not resellers
We do not earn commission on Google licences. That means the recommendation we give you is the one we would give a friend: Google where it fits, something else where it does not — and we name the something else.
We are an independent New Zealand consultancy specialising in Google’s AI stack. We do not claim partner or certification status we do not hold.
Questions we get asked
Do you only work with Google?
No. Google is our default because it covers the most ground per dollar for a typical New Zealand business. When another platform is the better fit, we say so — and the Google AI Challenge is how we prove that honestly, on your task, before you commit to a build.
Are you a Google Cloud partner?
We are an independent consultancy that specialises in Google’s AI products. We do not claim partner or certification status we do not hold. What we do have is delivery experience across Google’s AI stack, and a published offer that puts our confidence on the line.
We are on Microsoft 365. Can you still help?
Sometimes. Where Google’s models are the right engine we can call them from your Microsoft environment. But if your estate is genuinely Microsoft-native, staying there is usually cheaper than moving — and we will tell you that rather than sell you a migration.
What does it cost?
The Google AI Challenge is NZ$495, refunded in full if Google’s AI cannot meet the criteria we agree in writing. Builds are fixed-scope, typically NZ$6,000–NZ$30,000. Managed AI operations start at NZ$950 per month. The assessment is free.
How long before something is working?
One week to assess, two to four weeks for a pilot, four to eight weeks for a production build. The challenge is one hour plus a written verdict within two business days.
The Google AI Challenge
Bring us the task you think Google’s AI can’t do
One real business task, one hour live in Google’s stack. Acceptance criteria agreed in writing before you pay — that document is the only test. If Google’s AI cannot meet them, you get a full refund and we tell you which platform will.
NZ$495
Credited in full against any build started within 60 days. Full refund if Google’s AI cannot meet the agreed criteria. Four challenges a month.