Summary

• At SimScale's September Agentic Unlocks event, 31% of respondents prioritised engineering hours saved as their main AI performance indicator, followed by product performance at 22% and faster time to market at 17%.
• Just 5% selected direct cost reduction, while 22% had yet to define AI performance indicators.
• Separate adoption polling found that 70% had not progressed beyond informal experimentation, with 16% reporting AI embedded in daily work or moving into scaled deployment.

Engineering organisations are increasingly assessing artificial intelligence through the time it saves and the additional design work it enables, rather than direct cost reduction, according to polling conducted by Munich-based engineering software company SimScale.

During the company’s Agentic Unlocks event in September 2026, 31% of respondents selected engineering hours saved as their most important measure of AI success. Improved product design or performance followed at 22%, with faster time to market chosen by 17%.

Only 5% selected cost reduction as their primary indicator. The results suggest that participating engineering teams were placing greater emphasis on completing development work more quickly and investigating additional technical possibilities than on reducing expenditure alone.

However, the same event exposed considerable differences in deployment progress. In a separate question about adoption, 70% of respondents said their organisations had not moved beyond informal experimentation, while only 16% reported AI embedded in daily engineering work or progressing towards wider deployment.

The findings were gathered through live polling at an event attended by engineers and technology specialists interested in AI-assisted development. They provide evidence of the participating audience’s experiences rather than a representative measure of adoption across the engineering industry.

Engineering time becomes a performance measure

Product development involves repeated cycles of design, simulation, testing, and revision. Engineers examine whether proposed components and systems meet requirements involving performance, reliability, safety, cost, and manufacturing feasibility.

Computer-aided engineering software can reduce reliance on physical prototypes by allowing teams to examine how designs behave under specified conditions. Computational fluid dynamics models the movement of fluids, while structural analysis and thermal simulation assess mechanical and temperature-related behaviour.

These methods still require specialist preparation. Engineers must select appropriate models, define materials and operating conditions, establish boundaries, and assess whether the resulting calculations are suitable for the problem being investigated.

AI can support different parts of that process. Software agents may automate repetitive preparation tasks, while machine learning models trained on engineering data can estimate particular physical behaviours without running a complete conventional simulation for every design variation.

The two approaches involve different capabilities. An agent configuring a simulation must apply the appropriate tools and operating instructions, while a predictive model must produce results sufficiently accurate for the conditions under examination.

Both could shorten particular development activities, although their commercial benefit depends on whether the resulting information is reliable enough to influence decisions.

Reducing the time required to prepare or complete an individual calculation may allow engineers to investigate more alternatives before committing to a design. The resulting development programme could produce a better component, reach a decision sooner, or reveal problems before expensive physical testing begins.

Those outcomes explain why engineering hours and product performance may be more useful measures than a narrow calculation of software operating costs.

More design options do not guarantee better decisions

Conventional development processes often limit the number of alternatives examined because detailed simulations require computing resources and specialist time.

Where preparation and analysis become faster, engineers can compare additional geometries, materials, or operating conditions within the same project schedule.

That can be valuable when a design must satisfy competing requirements. Reducing the mass of a component may affect structural strength, while improving thermal performance may introduce additional pressure losses, equipment requirements, or manufacturing complexity.

Investigating those relationships across a broader set of alternatives can provide information that would otherwise be unavailable before a design decision is made.

However, evaluating more possibilities does not independently establish that the selected design is better. The validity of the underlying models, the range of conditions tested, and the criteria used to compare results determine whether additional analysis improves the decision.

SimScale’s separate State of Engineering AI 2026 research, based on 350 engineering leaders at large organisations in the United States, United Kingdom, and Germany, reported that teams using AI-enabled workflows examined more than three times as many design variants as those relying on conventional approaches.

That survey also identified faster responses to requests for quotations among organisations using AI-assisted processes. Engineering teams preparing a technical quotation may need to assess a product’s feasibility, performance, and manufacturing requirements before a commercial proposal can be completed.

Shortening that work could allow a business to respond to more opportunities, although the commercial result will still depend on quotation accuracy, customer demand, production capacity, and the contracts ultimately secured.

The 350-person study is a separate research exercise from the September event polling. Its findings should not be combined with the live poll percentages to calculate a single adoption or performance rate.

Pilots remain more common than routine deployment

SimScale’s detailed analysis of the Agentic Unlocks poll found that relatively few respondents had embedded AI within established engineering workflows.

When asked about their current adoption stage, 70% said they had not progressed beyond informal experimentation, including 37% who had not started. Only 16% described AI as embedded in daily work or moving towards wider deployment.

A separate question asked how organisations were progressing against their expectations for 2026. In those responses, 37% said they were on track, 34% had not started, 20% were behind plan, and 9% were ahead.

Because the questions examined different aspects of adoption and the company did not publish respondent counts for each question, the percentages should not be treated as interchangeable measures of the same population.

The results nevertheless indicate that moving from experimentation to operational use remains difficult for many participants.

A successful demonstration may involve automating one task under controlled conditions, while routine deployment requires the software to work consistently with established design systems, engineering records, and approval procedures.

Organisations frequently use different applications for computer-aided design, simulation, product lifecycle management, manufacturing, and commercial processes. Information produced in one application may need to be transferred or adapted before another system can use it.

Connecting those systems introduces practical requirements involving data quality, permissions, interfaces, and the management of changes. A simulation agent that works reliably with a single demonstration model may require additional development before it can handle the variety of projects encountered by an engineering team.

The difficulty increases when AI systems are permitted to make changes rather than merely provide recommendations. Organisations must determine which actions can be automated and when an engineer needs to approve the proposed result.

Measurement remains incomplete

Although engineering hours saved emerged as the leading performance indicator, SimScale’s detailed event analysis found that 22% of respondents had not defined AI performance measures.

Without an agreed baseline, teams can struggle to establish whether a pilot has improved their development process or simply introduced a new way of completing familiar work.

Engineering time provides one possible measure, but it does not capture every benefit. A project may consume a similar number of hours while producing a more effective design or examining more alternatives before testing.

Conversely, faster simulation does not necessarily produce a shorter product development cycle if later stages remain constrained by physical testing, procurement, or manufacturing requirements.

Useful evaluation therefore depends on the purpose of the technology. A system intended to accelerate design exploration might be assessed through the number of validated alternatives examined, while another automating simulation preparation could be measured against time saved and the consistency of its setup.

Predictive models also require accuracy measurements appropriate to their intended application. Their usefulness depends on whether the training data represents the conditions being investigated and whether engineers can identify circumstances in which the model’s estimates become unreliable.

For consequential design decisions, conventional simulation or physical testing may remain necessary even when AI provides a faster initial assessment.

Event results require careful interpretation

SimScale distributed an October account of the Agentic Unlocks polling describing 300 participants and reporting that 92% identified measurable AI value. It stated that 46% had experienced faster engineering cycles, 39% higher productivity, and 23% reduced simulation or testing effort.

The company’s more detailed event analysis used different figures for the question about where respondents saw the greatest value: 95% of those answering identified at least one benefit, while 32% selected faster development cycles and 26% higher engineering productivity.

The differences may reflect question wording, response options, or the groups who answered individual questions, but the published material does not provide sufficient question-level counts to establish a precise reconciliation.

Consequently, the figures cannot support a single combined estimate of the proportion of engineering organisations already obtaining measurable value from AI.

The more specific findings on success indicators and adoption stages can be considered within their respective questions without treating every percentage from the event as directly comparable.

The event itself attracted more than 1,000 registrations, but registration does not establish participation in every poll. Its audience also had a particular interest in engineering AI, which may differ from the broader engineering workforce.

Integrating AI requires engineering assurance

SimScale’s event included examples of organisations applying AI to specialised development tasks. Speakers from businesses including Convion, Dexory, Dolphin, and Thornton Tomasetti discussed applications involving simulation preparation, predictive modelling, and engineering workflows.

Such demonstrations can show how a particular task might be accelerated, but commercial deployment requires repeatable performance across the wider range of work undertaken by an organisation.

That includes the ability to detect unsuitable inputs, retain engineering assumptions, and provide information that allows specialists to investigate the results.

Where software generates a design or predicts its behaviour, engineers must understand the conditions under which that output is reliable. An apparently plausible result may still be unsuitable if the model has been applied outside its validated operating range.

AI systems that coordinate other engineering applications introduce additional requirements because they can carry out several actions through one workflow. Permission controls, records of tool use, and review procedures help organisations determine how those actions were performed.

The appropriate level of automation will differ between tasks. Preparing a routine simulation may present different risks from approving a design for manufacture or confirming that safety-related requirements have been satisfied.

The September polling indicates that participating organisations are increasingly evaluating AI through engineering outcomes rather than direct cost savings. At the same time, the relatively small proportion reporting routine deployment suggests that demonstrating an improvement in one task is not equivalent to integrating it into an established development process.

Further progress will depend on whether engineering teams can measure the quality and speed of their work consistently, connect AI with existing tools, and maintain the technical assurance required for dependable design decisions.