# SimScale Survey Finds Most Engineering AI Programs Remain in Pilot Stage > Source: > Published: 2026-07-21 15:13:42+00:00 TL;DR — Key Takeaways - Engineering companies are launching AI pilots quickly, but only 9% have reached mature deployment. - Data readiness is the biggest early obstacle, while software integration becomes the main challenge as programs scale. - Mature AI programs depend on strong governance, clear ownership, executive backing and cloud-native infrastructure. Engineering companies are launching more AI pilots, but few are moving them into large-scale deployment, according to new research from the cloud simulation platform SimScale. The State of Engineering AI 2026 report found 80% of respondents are running limited deployment pilots, up from 42% in 2025. But only 9% have mature programs, up from just 7% last year. The widening distance between experimentation and production suggests engineering teams are finding it much easier to start AI pilots than to integrate them into routine operations across the organization. Turning those pilots into working systems will require companies to identify barriers to scaling and apply lessons from those who have found success, SimScale said. The most common barriers to scaling AI programs reported by respondents include data preparation and availability at 74%, governance and compliance concerns at 48%, and software interoperability challenges at 42%. Those barriers shift with maturity. Data preparation was cited by 80% of organizations running pilots but only 47% of those with mature programs. Software interoperability showed the opposite pattern, rising from 39% among pilot-stage groups to 84% for mature organizations. That suggests the challenge changes from preparing data to integrating AI across existing engineering software and workflows. The survey found that organizations with the right foundations in place are moving faster, with the three top enablers being secure data governance and access controls at 70%, clear cross-functional ownership for moving pilots into production at 65%, and an executive mandate with a defined budget at 56%. Infrastructure was also reported to make a difference, with 55% saying cloud-native tech stacks are a key enabler for mature AI. Despite those foundations, the transition to production can take considerable time. Respondents reported an average of eight months to move an AI or agentic AI initiative from pilot to mature deployment. For 55%, the process took seven to 12 months, while 9% said it required more than a year. SimScale co-founder and CEO David Heiny said the difference in execution speed is creating another divide, with the fastest teams reaching scaled deployment in as little as three to six months while others remain stuck in experimentation. Even with many programs still in the pilot phase, the survey shows AI is being applied across a range of engineering workflows. Ninety percent of respondents reported some use of agentic copilots or autonomous agents, although 83% described those deployments as limited and only 7% as extensive. Autonomous agents were most common in simulation and computer-aided engineering, where they appeared in 19% of workflows, compared with 11% in design and CAD and 10% in requirements engineering. Surrogate models, which approximate the results of more computationally demanding simulations, were more widely adopted. SimScale found 92% of respondents had deployed them, although only 8% reported extensive use. Respondents also associated AI workflows with reducing average simulation turnaround from 17 hours to six and increasing the number of design variants evaluated from 17 to 56 per program. Those gains suggest AI is increasing engineering capacity by shortening analysis time and allowing teams to evaluate more designs within the same development cycle. Yet most firms are keeping that added capacity under human control: only 8% [reported](https://www.simscale.com/research-reports/state-of-engineering-ai-2026/) frequent autonomous decision-making, while 71% require human review of all AI outputs.