Enterprises Rush AI Agents Before Controls Are Ready
Survey shows enterprises deploy AI agents fast, but lag in governance, security, cost tracking, and evaluation, risking incidents and wasted spend.
Key Takeaways
Enterprises are adopting AI agents faster than they can secure, evaluate, and monitor them, creating governance gaps.
Survey data from 573 technical leaders reveal that 54% reported a security incident or near‑miss in the past year, while 27% only track costs after invoicing.
About six in ten firms intend to switch or add vendors across the five control layers within the next 12 months, with some planning moves within a single quarter.
The control layers include identity for agents, output evaluation, cost telemetry, contextual data, and orchestration.
GPU utilization is low — 86% of enterprises with owned GPUs report less than 50% usage — and only 44% rigorously measure compute expenses.
Most agents are single‑prompt chatbots; only roughly 10% can complete multi‑step tasks autonomously.
Two‑thirds allow agents to push code to production based on automated evaluations, but only 5% fully trust those evaluations.
Credential sharing across agents is common, increasing security risk, and 57% of wrong answers stem from missing or inconsistent business context.
Few organizations have deployed governed semantic layers or scoped identities, highlighting unfinished governance needs.
Potential Impact Areas
- Higher risk of security breaches as credential sharing goes unchecked.
- Uncontrolled spend leads to budget overruns and wasted GPU resources.
- Inaccurate AI outputs can damage customer trust and brand reputation.
- Delayed governance slows innovation and creates technical debt.
- Enterprises must invest in scoped identities, semantic layers, and real‑time quality monitoring to mitigate these risks.
Our Insight
Enterprises are moving fast with AI agents, but governance, security, and cost controls lag behind, increasing exposure to incidents and inefficiency.
Key risks include credential sharing that raises breach likelihood and missing business context that produces wrong answers.
Opportunities arise from adopting scoped identities and semantic layers, which can improve accuracy and trust.
Investing in real‑time evaluation testing and granular cost telemetry can prevent waste and enable sustainable scaling.
Overall, closing the governance gap is essential before full autonomous deployment can deliver reliable value.
External Credit
Original source: venturebeat.com
Full credit goes to the original publisher. We link to this content for informational and commentary purposes only.