This week Google launched Chrome Skills, a Gemini-powered feature that automates multi-step browser workflows and executes them persistently across sessions. Translation: you can now set up automation tasks in Chrome that run without you watching. The robot revolution finally figured out how to be useful without supervision.
Meanwhile, Deloitte and ServiceNow dropped their 2026 Workflow Automation Outlook, and the theme is "stop treating AI as magic and start treating it as maintenance." The key takeaway? Successful automation is not about replacing humans—it is about building living systems that improve over time. Groundbreaking, I know. Some consultant is probably printing this as a whitepaper as we speak.
On the security front, things got a touch more concerning. A Dark Reading poll found that 48% of cybersecurity professionals now rank AI agents and autonomous systems as the single most dangerous attack vector—outpacing ransomware, phishing, and whatever else is currently haunting your inbox. If your business runs AI agents, you are now officially in "handle with extreme care" territory.
And speaking of extreme care: prompt injection attacks—the digital equivalent of tricking someone into giving up secrets—are climbing the threat charts. IBM, Red Hat, and Microsoft all published guidance this week emphasizing the same point: AI systems that ingest external content need defense-in-depth protections. One misleading prompt could have your AI assistant accidentally sharing customer data with whoever asked nicely.
Practical Takeaway for Small Businesses: Before you connect another AI tool to your workflows, audit what data it can access. Limit permissions. Use short-lived authentication tokens. Treat your AI assistant like you would any new employee—least privilege principle applies here too. The good news: you do not need an enterprise security budget. You need discipline.
Sources: Google Chrome AI Skills Launches Workflow Automation, Deloitte/ServiceNow 2026 Workflow Automation Outlook, Bessemer: Securing AI Agents 2026, IBM: What is a Prompt Injection Attack, AI Agent Security for SMBs 2026 Playbook