"Agentic" Is the New Cloud

Microsoft has Copilot Cowork. Salesforce is touring the world with “agentic enterprise.” Gartner named multi-agent systems a top 2026 strategic trend, G2 released its first “Best Agentic AI Software” list, and Forbes declared agent workflow design the hottest skill in the economy. Every enterprise vendor has retrofitted their marketing with the word.

What I keep seeing in the market: “agentic” applied to anything from chatbots with slightly better prompts to genuinely autonomous multi-step systems, with no distinction between the two.

This is the cloud playbook: first the word means something specific, then every vendor needs it in their deck, then it means nothing at all.

If you’re evaluating something that claims to be agentic, there’s a simple three-part test. Does it maintain context and state across multiple steps — not just pass variables, but actually remember what happened earlier and why? Does it make decisions when information is incomplete or ambiguous, rather than halting for human input at every junction? And when something breaks mid-workflow, does it recover, reroute, and continue, or does it just fail and wait?

Most of what I see in the market fails at least two of those. Real agent workflows involve careful permission boundaries, failure handling, and context management that vendors rarely show in demos because it’s messy and situational.

That complexity is exactly what makes agents actually useful. The rebranded automation with better copy? That’s just a workflow with a marketing budget.