AI agents: the promise, the hype and the risks nobody mentions

An "AI agent" does more than answer. It acts: books, sends, updates, chains multiple steps together. The potential is real, especially for repetitive workflows. But in 2026 almost everything is sold as "agentic", whether or not it's true. Analysts call it agent washing.

Gartner predicts that over 40% of all agentic-AI projects will be cancelled before the end of 2027, due to escalating costs, unclear business value and inadequate risk controls. That's not tech pessimism; it's what happens when autonomy is introduced faster than governance.

The more an agent can do, the bigger the damage when it gets it wrong. Prompt injection, where hidden text in a document or email hijacks the agent's instructions, isn't theoretical; EchoLeak showed it in practice. Add over-permissioned agents and weak traceability, and you've built something that can act wrongly at scale, fast.

• Scope it down. Give the agent least privilege and a clear, narrow job.
• Human in the loop for anything hard to undo.
• Log everything. If you can't see what the agent did, you can't trust it.

That's how we introduce automation: controlled, vendor-neutral and measurable. Read more about AI & Automation or book a consultation.