One in five enterprises can't stop runaway AI agent spending in real time
Companies struggle to control AI agents' real-time expenditures
1Multi-platform usage on the rise
85% of enterprises use multiple AI orchestration platforms, averaging three different ones. Microsoft leads with 70% adoption rate. OpenAI's Agents SDK has 68% usage, while Anthropic's Claude Platform stands at 47%. Companies adopt this strategy to avoid vendor lock-in and maintain flexibility in their AI operations.
2Hybrid control planes gaining traction
53% of respondents predict hybrid will be the primary control plane by 2026. 14% prefer provider-managed services, while 13% opt for custom in-house control planes. 28% plan to switch platforms within 6-12 months, highlighting the importance placed on operational flexibility and avoiding single-vendor dependencies.
3Security concerns take priority
Security and permission limitations constitute 37% of top platform selection concerns. 23% worry about vendor lock-in, while 22% cite visibility limitations. Only 2% prioritize latency and memory performance. In these early stages of AI agent adoption, control mechanisms remain immature across the board.
4Real-time control gaps persist
21% of enterprises cannot stop runaway AI agent spending in real time. 30% rely on native platform controls like budget caps, while 25% have built custom gateway solutions. Another 25% use dynamic routing to cheaper models, and 21% still depend solely on reactive monitoring without real-time intervention capabilities.
5Investment focus areas
Enterprises allocate 31% of spending to agent monitoring/debugging and 30% to security/permissions enforcement. Workflow tooling accounts for 19% of budgets. Top priorities include task completion reliability (30%) and multi-step workflow management (27%). End-user experience currently ranks as priority for only 7% of respondents.
6Platform satisfaction levels
Respondents rate their platforms 4.17/5 for overall satisfaction, but implementation ease scores 3.91 and value for money 3.63. As platforms mature, these ratings may shift. Anthropic's Claude Agent SDK is under consideration by 43% of builders, indicating strong market interest in alternative solutions.
7Purchase decision drivers
Flexibility leads purchase considerations at 29%, followed by security/permissions (17%), production reliability (15%), and agent execution control (15%). Model gravity matters to only 10%, while development ease (8%) and total cost of ownership (4%) rank lower. These metrics underscore the premium placed on control.
8Preparing for the future
Enterprises are architecting for a future where multiple orchestration platforms, models, and agents interoperate across hybrid control planes. Learning from early cloud lock-in experiences, they're avoiding single-vendor bets. As AI roadmaps mature, developer and end-user experience considerations will likely gain importance in deployment strategies.