Industry Trends
-
Google Cloud and Oracle let AI agents work directly with Oracle databases
Google Cloud made the remote MCP server for Oracle AI Database@Google Cloud generally available, so AI agents can run SQL against Oracle databases under IAM controls, and put a Gemini connector into preview. My take: whether an agent is useful usually depends less on the model and more on whether it can reach the right data. Sorting out data permissions first is worth more than chasing the newest model.
-
Oracle opens an AI consulting practice and plans to certify a million people
Oracle launched an AI consulting practice built on OpenAI technology and plans to train and certify one million developers and professionals through Oracle University. My take: even the giants admit the bottleneck isn't the model, it's people who know how to use it. Teaching your own team first beats buying yet another tool.
-
Salesforce adds evaluations and spending caps for AI agents
Salesforce added Agent Evals, API Evals and cost controls to Agent Fabric, so AI agents are tested continuously, can be stopped automatically when quality slips, and can be held to hard spending limits. Bottom line: an AI agent is like a new hire. It needs performance reviews and an allowance. Decide how you'll check its work and what it can spend before rollout, not when the bill arrives.
-
Zendesk introduces AI agents for employee service, from request to resolution
Zendesk launched Employee Service AI Agents and Advanced Action Flows to handle multi-step internal requests such as access approvals, laptop requests and employment letters. They are in open early access, with general availability expected within weeks. Bottom line: the first place AI saves time is often the paperwork inside your own company. Start with the internal request everyone complains about most.
-
Goodwill expands AI and digital skills training in California with Google.org support
Goodwill announced $4 million in funding from Google.org to give about 51,000 Californians access to AI and digital skills training over four years. My take: when nonprofits teach AI as a basic job skill, learning it is no longer a bonus for individuals. It's catching up on the basics.
-
Google Cloud introduces Agent Compute Engine and broader agent governance
Google Cloud announced Agent Compute Engine, a compute service built for AI agents that runs them in hardware-isolated sandboxes, with a preview expected in the coming months, and expanded agent governance in Gemini for business. My read: once agents start doing real work, the question shifts from "is it smart?" to "can I trust it?" Decide what data each agent may touch, and who is accountable when it slips.
-
Oracle announces AI Agent Studio for Industry Applications
Oracle announced AI Agent Studio for Industry Applications, a no-code and low-code environment for building, testing and managing AI agents around industry-specific processes, planned for the U.S. within 12 months. My read: building agents keeps getting easier, so the real gap will be who understands their own processes best. Write your workflows down now, so the tools have something to plug into.
-
NVIDIA launches the OpenShell Reference Architecture for safe AI agents
At GTC Berlin, NVIDIA launched the OpenShell Reference Architecture, which combines a secure runtime with BlueField DPUs so companies can isolate, monitor and restrict what AI agents do. Bottom line: when even chipmakers sell seatbelts for agents, everyone agrees agents will make mistakes. Planning permissions and logging up front is far cheaper than cleaning up later.
-
Google introduces Gemini for government services
Google launched Gemini for government services, with skills built around regulations, connectors to existing agency systems and central governance, in preview for U.S. government agencies. My take: if the most regulated sector is adopting AI agents, "our industry is special" is becoming a weak excuse. Repetitive work like document review is the practical place to start.
-
Microsoft treats AI agents as the newest insiders and adds new controls
Microsoft extended Project Perception with Guardian Agents that watch enterprise AI agents, plus controls that adjust an agent's permissions as its risk changes and require human approval for risky actions. My read: manage AI agents like new hires, with just enough access and a clear record. The easiest first step is simply listing which AI tools are already running in your company.