Answer-First Lead
Google launched Gemini 3.5 Flash at I/O 2026 on May 19, claiming it beats Gemini 3.1 Pro on coding and agentic benchmarks at 4× the output speed. The bigger news: Gemini Spark, a 24/7 personal AI agent that works autonomously even when your devices are off. And NZ’s own Xero is named as an enterprise partner, deploying 3.5 Flash agents to automate multi-week tax compliance workflows.
🔍 THE BOTTOM LINE
Google is betting the next AI era isn’t about better chat — it’s about agents that do things for you. Xero’s involvement means Kiwi small businesses will be among the first to test whether that bet pays off.
What Gemini 3.5 Flash Actually Does
The headline number: 76.2% on Terminal-Bench 2.1 and 83.6% on MCP Atlas — both agentic coding benchmarks where Gemini 3.1 Pro scored lower. Google claims 3.5 Flash outputs tokens at 4× the speed of other frontier models.
What is Gemini 3.5 Flash? It’s Google’s latest AI model optimised for agentic tasks — meaning it can plan, execute, and iterate through multi-step workflows autonomously, rather than just answering questions. It’s available immediately in the Gemini app, AI Mode in Google Search, and via the Gemini API in Google AI Studio and Android Studio.
The real shift isn’t the benchmark scores — it’s the cost. Google says 3.5 Flash runs at half to one-third the price of comparable frontier models. That’s the kind of pricing that makes “always-on agents” economically viable for the first time.
Gemini 3.5 Pro is coming next month, currently being used internally at Google.
Gemini Spark: Your 24/7 AI Co-Worker
The flashier announcement was Gemini Spark — a personal AI agent that lives inside the Gemini app and handles tasks across your connected apps, even when you’re not watching.
According to Google’s product page, Spark “runs 24/7, helping you navigate your digital life, taking action on your behalf while under your direction.” That’s carefully worded — you’re still the boss, but the agent works in the background, handling multi-step tasks that could take days or weeks.
Gemini Spark is rolling out to trusted testers first, with a beta coming to Google AI Ultra subscribers in the US next week. That $100/month Ultra plan is starting to look like Google’s answer to the “pro tier” of AI access — and Spark is the headline feature.
Xero: The NZ Angle That Matters
Here’s where it gets interesting for Aotearoa. Google named Xero as one of its enterprise partners deploying 3.5 Flash agents:
“Xero is deploying agents to autonomously manage complex, multi-week workflows, such as identifying suppliers and gathering information for 1099 tax forms, enabling small businesses to automate tedious admin tasks.”
This isn’t casual integration. Xero is building agents that work for weeks — not minutes. Think about what that means: an AI agent that identifies your suppliers, gathers their tax information, and prepares your 1099 forms without you touching it. That’s a task that currently takes small businesses hours of tedious admin.
And here’s the context that makes this more than a press release: Xero just launched XeroForce on May 14 — five days before Google’s I/O. XeroForce is a natural language AI agent builder that lets accountants and small businesses create custom agents for financial workflows. It’s powered by Xero OS, the company’s AI-native financial operating system.
The timing isn’t coincidental. Xero has been building toward agentic AI for months with JAX (its financial superagent) and now XeroForce. The Google partnership gives them Gemini 3.5 Flash as the engine underneath — and it gives Google a credible enterprise partner who can show agents doing real work, not just demo tricks.
Other named enterprise partners include: Shopify (subagents for merchant growth forecasts), Macquarie Bank (customer onboarding), Salesforce (Agentforce integration), Ramp (OCR for invoices), and Databricks (monitoring and diagnostics).
Gemini Omni: The World Model
Google also announced Gemini Omni, a world model that simulates physical environments and predicts what happens next based on user actions. It’ll work across Flash, the Gemini app, Google Flow, and YouTube Shorts.
The demo showed Omni editing videos — changing what’s happening in a clip, adding characters or objects. That’s fun for YouTube Shorts, but the real play is robotics and gaming, where world models have been a DeepMind research focus for years.
Why This Matters
Three things make this I/O different from the usual Google feature dump:
-
Agents, not chat. Every major announcement was about AI doing things, not just answering things. Spark, Xero’s multi-week workflows, Shopify’s subagents — the industry is pivoting hard from “AI that talks” to “AI that works.”
-
The NZ connection is real. Xero isn’t a demo prop. They’re shipping an agent builder (XeroForce) and have a working financial superagent (JAX). When Google says Xero is “deploying agents,” that’s not future tense — it’s happening now.
-
Price makes it possible. Half to a third the cost of frontier models. That’s the difference between an agent that’s cool in a demo and an agent that makes economic sense to run 24/7.
The Skeptical Take
Let’s be honest: Google has announced a lot of AI features at I/O over the years. Not all of them have stuck. Gemini Spark is “rolling out to trusted testers” — which is code for “not fully baked yet.” The $100/month Ultra plan is steep for a personal agent that might not work as advertised.
And Xero’s XeroForce is still in alpha, invite-only. “Plans to bring to general release later this year” is the kind of timeline that has slipped before.
The agentic AI era is coming. But “coming soon” and “available today” are not the same thing. Watch what ships, not what’s announced.
❓ Frequently Asked Questions
Q: What does this mean for NZ? Xero is headquartered in Wellington and employs thousands of Kiwis. If Xero’s AI agent strategy works, NZ becomes a proving ground for agentic finance — and the skills and infrastructure that grow around that could make Aotearoa a hub for AI-native business tools.
Q: How is Gemini Spark different from ChatGPT or Claude? Spark is an agent, not a chatbot. It works 24/7 in the background, connecting across your apps, handling multi-step tasks that unfold over days. ChatGPT and Claude are primarily conversational — you ask, they answer. Spark asks less and does more.
Q: Should I sign up for Google AI Ultra? At $100/month, only if you have a concrete use case for an always-on agent. The model (3.5 Flash) is available free in the Gemini app and AI Mode in Search. Spark is the Ultra differentiator, and it’s still in early testing.
Q: What’s XeroForce and how does it relate to Gemini? XeroForce (launched May 14) is Xero’s own agent builder for financial workflows. It runs on Xero OS. The Google partnership means Xero can use Gemini 3.5 Flash as the underlying model for some of these agents — giving them frontier intelligence at a fraction of the cost.
🔍 THE BOTTOM LINE
Google I/O 2026 wasn’t about smarter chatbots. It was about agents that work while you sleep — and Xero’s involvement means NZ small businesses will be among the first to find out if that’s a revolution or a sales pitch. The tech is real. The economics are getting there. The question is whether “trusted testers” becomes “everyone” before the next I/O.
