Change: Flexible architecture prevents AI lock-in

This is chapter 5 of Box's State of AI in the Enterprise report 2026. Read more:

-----

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).

There will be multiple agent platforms — there's no one platform to rule them all. Enterprises will run several, which means those platforms need to be able to interoperate.

Rao Surapaneni, Google

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.

Adoption rises with maturity for nearly every tool the survey measured. From early stage to leading edge, ChatGPT use doubles (34% to 68%), Gemini climbs by roughly three-quarters (28% to 48%), and Salesforce Agentforce and GitHub Copilot more than double (9% to 28% and 11% to 28%).

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 cannot 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.

A clear majority — 80% — say it′s important or critical that agents can operate ˮheadlesslyˮ: autonomously, without a human interface, connecting directly to systems, APIs, and data sources.

The gap between leading edge and early stage is wide. 54% of leading-edge organizations call headless operation critical, against 3% at the early stage. 94% of leading edge organizations call headless operation important or critical, against 39% at the early stage.

Headless operation is a near-universal view at the front of the maturity curve and a minority position at the back.

This is chapter 5 of Box's State of AI in the Enterprise report 2026. Read more:

-----