Daily AI brief · May 2, 2026
Agent 365 Goes GA as Open-Source Models Hit Laptop Scale
Microsoft's Agent 365 governance layer went live May 1, open-source models like Gemma 4 now run on consumer hardware, GPT-5.5 rolls out to business tiers, and the Pentagon's AI vendor shakeup puts Anthropic's future in question.
1. Microsoft Agent 365 Goes GA — and It Costs $15/User/Month *(Microsoft / Redmond Magazine)*
Microsoft announced the general availability of Agent 365, its control plane for governing and securing AI agents across enterprise environments, alongside the release of Microsoft 365 E7.
The new Microsoft 365 E7 "Frontier Suite" bundles Microsoft 365 E5, Microsoft 365 Copilot, and Agent 365 — now generally available — into a single solution powered by Work IQ.
Separately,
the latest Copilot updates let it take multi-step actions directly inside documents, spreadsheets, and presentations instead of simply responding to prompts or suggesting next steps.
What it means for SMBs / MSPs:
Agent 365 addresses the oversight gap for enterprise buyers who've been cautious about autonomous agents — but at $15/user/month it's targeted at enterprise IT and security teams, not individual users or small businesses.
Evaluate whether the E7 bundle math works before pitching it to clients in the 50–250 seat range.
2. AI Builder Credits Sunset Clock Is Ticking *(Microsoft Learn)*
This is the licensing change your Power Automate clients may not know is coming.
New customers must purchase Copilot Credits to run AI Builder features; AI Builder credits seeded in Power Platform or Dynamics licenses will be removed in November 2026, and AI Builder trials are discontinued.
There is no automatic conversion of old AI Builder Credits to Copilot Credits.
What it means for SMBs / MSPs: Any client running invoice processing, document extraction, or prompt-based flows via Power Automate Premium is affected.
November 1, 2026 is the hard deadline; organizations that fail to plan will experience service disruption.
Run a credit audit now — not in October.
3. GPT-5.5 Rolls Out to Business and Enterprise Tiers *(OpenAI / Releasebot)*
GPT-5.5 is rolling out to Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, and GPT-5.5 Pro is rolling out to Pro, Business, and Enterprise users in ChatGPT.
The API rollout is coming "very soon."
GPT-5.5 shipped with major gains in agentic coding, computer use, and knowledge work.
What it means for SMBs / MSPs: If your clients are on ChatGPT Business or Enterprise plans, they're getting meaningfully upgraded models without doing anything — but it's worth validating that existing prompts and workflows still behave as expected after the bump.
4. Open-Source Models Now Run on Laptop Hardware — Frontier-Level *(Multiple sources)*
Twelve months ago, picking an open-source LLM meant choosing between "good enough for prototyping" and "not ready for production." That era is over. In April 2026, six major labs ship open-weight models that match or beat proprietary alternatives on key benchmarks — and you can run several of them on a single consumer GPU.
Google's Gemma 4 packs 26B parameters into just 14 GB and hits 85 tokens per second on consumer hardware.
The biggest shift in 2026 isn't just model quality — it's that open-source models now support the agentic patterns (function calling, tool use, MCP integration) that were previously exclusive to proprietary APIs.
What it means for SMBs / MSPs: Clients with data-residency concerns, healthcare privacy requirements, or tight monthly API budgets now have a credible local-first option. This is a real conversation to have — not a hobbyist recommendation.
5. Pentagon Picks 8 AI Vendors, Excludes Anthropic *(CNN Business / Breaking Defense)*
The Defense Department made agreements with Amazon Web Services, Google, Microsoft, NVIDIA, OpenAI, SpaceX, and Reflection — with Oracle added hours later.
Until recently, Anthropic's Claude was the only AI model available in the Pentagon's classified network, but the Trump administration severed ties after Anthropic refused to allow the military to use Claude for "all lawful purposes," including autonomous weapons and mass surveillance. The Pentagon declared Anthropic a "supply chain risk."
What it means for SMBs / MSPs: Anthropic's commercial Claude products are unaffected for now, but the dispute is a reminder that AI vendor relationships are subject to political and regulatory disruption. If a client is betting heavily on a single provider, build a fallback.
Gemma 4 26B via Ollama — Local LLM in Under 10 Minutes
Running powerful AI language models locally has become increasingly accessible in 2026, offering privacy, cost savings, and full control over data — with groundbreaking releases like Google's Gemma 4 now rivaling cloud-based services.
- What it is: Google's Gemma 4 26B, an Apache 2.0-licensed open-weight model that runs on a single consumer or workstation GPU via Ollama.
- Who it's for: MSP techs and internal AI champions evaluating private, on-prem inference for clients with HIPAA, legal, or data-residency constraints.
- How long a first test takes: 30–45 minutes including download and a few trial prompts.
- First step: Install Ollama (ollama.com), then run `ollama run gemma4:26b` in your terminal. Test against a document summarization or classification task your client already does manually today.
If you have Power Automate Premium, check what uses your AI Builder credits before November 1, 2026. The credits seeded in those licenses disappear on that date, with no automatic conversion. Look at the credit usage report in the Power Platform Admin Center and note anything running on them, like invoice processing or form extraction. Plan the move to Copilot Credits before something stops working.
*For MSPs:* A credit consumption report plus a Copilot Credits migration plan is a hard-dated reason to review each client's Power Platform setup.
Don't lead client pitches with Agent 365 pricing math for small tenants.
Agent 365 is not relevant for individual users or small businesses — it's priced at $15/user/month on top of existing Microsoft 365 subscriptions.
Positioning it to a 12-person professional services firm as a must-have will erode trust. Save the governance layer conversation for clients already running Copilot Studio agents at scale.
Governance is becoming the new AI product.
Three things define the AI landscape entering May 2026: agents are the product now, not the feature. Microsoft Agent 365, Cursor's Agents Window, Claude Code's multi-agent coordination — the shift from AI as a chat interface to AI as an autonomous executor is complete at the tooling level. The question for May is adoption and governance, not capability.
Watch for SMB clients starting to ask not just "can AI do this?" but "who's watching it when it does?" — that's your opening to build audit trails, approval workflows, and human-in-the-loop checkpoints as a billable service layer, not just a deployment footnote.