One in five enterprises can't stop runaway AI spending in real time
Trust and control issues persist in enterprise AI orchestration platforms
1Multi-platform usage becomes standard
Enterprise AI teams no longer rely on single orchestration platforms. According to VB Pulse data, the median company now runs three platforms simultaneously. While Microsoft leads in current adoption, Anthropic's Claude Platform emerges as the most considered alternative. This trend reflects not just vendor lock-in concerns but also doubts about platforms' security and permission management capabilities.
2Diversification of platforms accelerates
Among 107 surveyed enterprises, 85% use two or more orchestration tools, with 64% running three different platforms concurrently. Microsoft AI Foundry/Copilot Studio leads with 70% adoption, followed by OpenAI's Agents SDK (68%) and Anthropic's Claude Platform (47%). Notably, 22% of companies have developed custom in-house orchestration solutions.
3Hybrid control planes shape the future
53% of respondents predict hybrid will become the primary control plane by 2026. While 14% plan to use provider-managed services, 13% prefer custom in-house solutions. 28% of enterprises intend to switch platforms within 6-12 months, demonstrating lessons learned from early cloud vendor lock-in experiences.
4Investment priorities shift
Companies are prioritizing agent monitoring (31%) and security/permission enforcement (30%), with workflow tooling accounting for 19% of spending. Top optimization goals include task completion reliability (30%), multi-step workflow management (27%), and developer productivity (23%). End-user experience currently ranks as a priority for only 7% of respondents.
5Visibility and control concerns dominate
Security limitations (37%), vendor lock-in (23%), and limited observability (22%) emerge as top platform selection concerns. 16% cite inflexibility around models and tools. In these early agent days, many enterprises struggle with spending control - one in five can't stop runaway agent spending in real time.
6The runaway spending challenge
20% of enterprises lack real-time controls to stop excessive AI agent spending. To manage costs, 30% rely on native platform controls, while 25% have built custom gateway middleware. 25% use dynamic routing to cheaper models, and 21% depend solely on reactive monitoring without real-time intervention capabilities.
7Cost control strategies evolve
Enterprises are deploying various strategies to control AI spending, from built-in budget caps to custom proxy solutions. Dynamic routing to low-cost models gains traction, while some organizations still rely on post-hoc log analysis without proactive controls. The approaches reflect growing maturity in AI cost management.
8Future expectations
As orchestration platforms mature, enterprise priorities are expected to shift. Current satisfaction scores average 4.17/5, with implementation ease (3.91) and value (3.63) trailing. The ability to manage multi-step workflows reliably will become increasingly critical for platform success in enterprise environments.
Original article: VentureBeat — Şirketlerin beşte biri kontrolsüz AI harcamalarını durduramıyor