Daily AI brief · May 21, 2026
Google I/O Drops Gemini 3.5 Flash + Spark Agent
Google I/O 2026 shipped Gemini 3.5 Flash (faster and cheaper than 3.1 Pro), launched Gemini Spark as a 24/7 personal agent for Workspace, cut AI Ultra pricing, and Microsoft quietly armed M365 Copilot with Claude Opus 4.7 — while a fresh IDC study finds 81% of SMBs are unprepared for AI-driven threats.
1. Google Gemini 3.5 Flash Goes Live — Speed Up, Price Down *(Google I/O / Google Cloud Blog)*
Gemini 3.5 Flash combines frontier intelligence with agentic task execution, surpassing Gemini 3.1 Pro across coding, agentic, and multimodal benchmarks — at 4× the output-token speed of comparable frontier models.
Gemini 3.5 Flash is generally available today, priced at $1.50/$9 per million tokens, with 1M context and strong scores on agentic benchmarks.
What it means for SMBs / MSPs: If you're building RAG pipelines or client-facing automations on Google's stack, this is a real upgrade path — faster throughput at a price point where production volume deployments are actually affordable.
2. Google Launches Gemini Spark — A 24/7 Workspace Agent *(Google I/O / 9to5Google)*
Gemini Spark is described as a "personal agent" that takes actions on your behalf, transforming Gemini "from an assistant that can answer your questions into an active partner that does real work on your behalf and under your direction,"
integrating with Gmail, Docs, and Workspace before expanding to third-party tools via MCP.
Gemini Spark will be available next week to Google AI Ultra subscribers in the US.
What it means for SMBs / MSPs: Workspace-first businesses get a direct rival to Microsoft's Copilot agent story — worth a head-to-head test if clients are on Google Workspace rather than M365.
3. Claude Opus 4.7 Lands in Microsoft 365 Copilot *(Microsoft 365 Roadmap)*
Microsoft is bringing Claude Opus 4.7 to Microsoft 365 Copilot, promising stronger instruction following, improved tool selection, and higher-resolution visual understanding for complex, multi-step work tasks.
Separately,
GPT-5.5 Instant is now available in M365 Copilot, delivering low-latency responses optimized for common work questions, image-based inputs, and STEM tasks.
What it means for SMBs / MSPs: Copilot licensees now get model choice inside M365 without any integration work — use this as a concrete ROI conversation starter with clients still sitting on the fence.
4. Microsoft Extends Insider Risk Controls to AI Agents *(MSFTNewsNow / Microsoft 365 Roadmap)*
Microsoft is extending Insider Risk Management to AI agents across Copilot Studio, Azure AI Foundry, and Agent 365, so admins can define policies that detect, flag, or block risky agent activity — acknowledging that AI agents can exfiltrate or misuse data just as quickly as human users.
What it means for SMBs / MSPs: If you've already deployed agents in Copilot Studio, now is the time to review your DLP and insider risk policies before auditors do it for you.
5. IDC Study: 81% of SMBs Unprepared for AI-Driven Threats *(Help Net Security / IDC — May 21, 2026)*
A survey of over 2,000 SMBs finds 52% rank cybersecurity as a top priority for the next 12 months, second only to growth, and 60% expect to increase security spending.
But
AI adoption is intensifying cybersecurity pressure, and 81% of SMBs say they are not prepared or are in the early stages of preparedness for AI-related threats, with 22% having yet to implement any dedicated protections for AI applications.
What it means for SMBs / MSPs: The willingness-to-spend signal is strong — this data is tailor-made for a managed security conversation framed around AI threat readiness rather than generic compliance checkboxes.
Gemini 3.5 Flash via the Gemini API
Google kicked off the Gemini 3.5 series with 3.5 Flash, which delivers frontier performance for agents and coding and excels at complex long-horizon tasks.
It's live in the Gemini API right now.
- Who it's for: MSP developers, internal AI champions, and anyone already prototyping with LangChain, n8n, or direct API calls.
- How long a first test takes: 20–30 minutes to swap a Gemini 3.1 Pro call to 3.5 Flash in an existing project and compare outputs.
- First step: Head to Google AI Studio (aistudio.google.com), select Gemini 3.5 Flash from the model picker, paste your current prompt, and run a cost/quality comparison against your existing baseline.
The model is priced at $1.50 per 1M input tokens and $9 per 1M output tokens.
List three repetitive tasks, like inbox triage, meeting prep and document routing, and time how long they take. Google's Gemini Spark is a Workspace agent that takes actions on your behalf, coming next week to Google AI Ultra subscribers in the US. Before paying for any agent, know which chores cost you the most hours. That number tells you whether a new plan is worth it.
*For MSPs:* A Workspace AI audit that maps a client's manual workflows against what Spark can handle gives Google shops their answer to the Copilot story.
Don't pitch Gemini 3.5 Flash as a straight cost-reduction play without checking actual usage patterns first.
Independent analysis indicates that token consumption associated with operational costs has surged nearly fivefold alongside the new model, meaning enterprises with high token usage may see total costs rise rather than fall.
Benchmark the throughput of your specific workload before promising clients a lower bill.
AI agents are becoming first-class security actors — and governance policy is racing to catch up. Microsoft's move to bring AI agents under Insider Risk Management policies is not a one-off feature drop; it's a signal that every major platform is starting to treat agents the same as human users from a compliance and audit perspective.
Organizations need risk analytics and enforcement that treat agents as first-class actors — and security teams should start including AI agents in insider risk tabletop exercises and DLP testing workflows now.
For MSPs, this opens a durable service opportunity: agent governance reviews, policy templating, and quarterly agent-activity audits are becoming as essential as endpoint management was five years ago. The SMBs building AI workflows today without governance scaffolding are tomorrow's breach case studies.