This is chapter 5 of Box's State of AI in the Enterprise report 2026. Read more:
1: Executive Summary | 2: The Maturity Gap | 3: Context | 4: Control | 5: Change | 6: Capability | 7: Conclusion
-----
TL;DR: The tools will keep changing, so the leading edge refuses to bet on which one wins. It builds headless, runs multiple models, and keeps every part swappable.
Infrastructure flexibility is the third foundation of the agentic enterprise. In 2025, we found that leading AI-first organizations weren′t limiting themselves to single solutions: respondents reported using an average of 2.3 model providers. In 2026, that trend has accelerated, with respondents now using about 3.3 AI tools on average (summed selection shares).
What AI agents enable is that they have the ability to plan and reason. What used to take a week to plan is now done in a matter of hours at AWS. In terms of ROI, we've saved thousands of developer hours as well. It's a game changer in productivity for us.
Multi-model is becoming deployed reality
A year ago, the multi-model pattern was visible in the data: organizations with the highest AI ROI used roughly three model providers; those with the lowest used two.
In 2026, it′s a stated strategy. 44% now say multi-model is the best way to scale AI; 23% say they standardize on one; 32% say it depends on the use case; 2% aren′t sure.
Concern about lock-in cuts across the population
There′s broad agreement that model lock-in is a risk: 68% are somewhat or very concerned about being locked into a single AI model or provider. The share barely varies by company size, industry, or region.
By maturity, it runs from 62% at early stage to 65% at leading edge, peaking at 72% in the advanced tier.
Lock-in isn′t a worry organizations grow into or out of; it′s a near-constant.
Headless operation lets agents do real work
Agents can't just sit inside chat windows. To do real work, they need to operate directly across systems, APIs, workflows, and trusted content, with the right permissions and controls.
The survey found widespread acceptance and approval of this idea. 80% call it "important" or "critical" that agents be able to access and operate software "headlessly" -- without a human interface -- instead connecting directly to systems, APIs, and data sources.
As elsewhere, this question reveals a wide gap based on the maturity curve. 54% of leading-edge organizations call headless operation critical, against just 3% at the early stage. A near-universal 94% of leading edge organizations call headless operation important or critical, against 39% at the early stage.
This is chapter 5 of Box's State of AI in the Enterprise report 2026. Read more:
1: Executive Summary | 2: The Maturity Gap | 3: Context | 4: Control | 5: Change | 6: Capability | 7: Conclusion
-----
