Bentley Systems brings AI into infrastructure design and digital twins

Bentley Systems applies AI to infrastructure engineering, where a useful answer must remain connected to a design model, site conditions and engineering constraints. Its current offer spans OpenSite+ for civil site design, digital-twin services through iTwin and connections between AI agents and engineering applications. The practical decision is which part of the design or operating workflow needs help and what evidence will make the result acceptable.

01 / ProductEngineering applications and platform services sit at different layers

OpenSite+ is Bentley’s AI-enabled application for site planning and land development. Its model-first approach uses site objects and constraints to support layout, grading and related design work. Bentley describes Copilot access to project documentation and design interaction. The product should be evaluated as an engineering environment rather than as a free-standing conversational assistant.

iTwin Platform provides developer services for digital-twin applications. That is a different route from buying a finished design application. A developer must still build the intended user experience and connect the appropriate engineering data. Bentley’s Cesium acquisition in September 2024 adds geospatial technology to that broader platform story; Cesium is covered here within the company identity.

Bentley also publishes MCP servers that connect AI tools to engineering applications and data. The page marks the STAAD.Pro route as available and MicroStation MCP as restricted Early Access. A working connector for one application does not establish availability or identical permissions for another. That distinction matters when planning a multi-tool engineering workflow.

02 / AudienceCivil teams should begin with an identifiable design bottleneck

A land-development team exploring site options has a concrete OpenSite+ use case. An infrastructure owner building an operational digital twin has a different need, more likely centered on data integration and a platform application. Both may use AI, but they should not be assessed with the same acceptance criteria or budget assumptions.

The Autodesk blueprint offers an adjacent design-software comparison. Examine existing project formats, collaborators and delivery requirements before deciding which AI experience fits. Infrastructure projects often outlast a particular software release, so the ability to preserve engineering meaning across handoffs is as important as the convenience of the first design session.

The NVIDIA blueprint explains a lower layer of AI and simulation infrastructure. Bentley’s relevance is closer to civil engineering tools and the information describing built assets. A project can use both, but a computing platform does not remove the need to define terrain, drainage, structural or asset-data assumptions.

This is less suitable as an entry point for someone who simply wants to produce a conceptual image of a development. Site design involves geometry, quantities and constraints that affect real engineering work. A team needs an appropriate professional reviewer and a known delivery process before it can judge whether an AI-assisted result is useful.

03 / WorkflowA proposed site-design pilot compares alternatives under the same constraints

Consider a civil team evaluating a commercial site with a building pad, access roads and parking. The following is a proposed evaluation, not a hands-on test. Start with a controlled project containing the approved survey, site boundary and applicable design inputs. Record which constraints are fixed and which may vary between alternatives.

Before asking for design assistance, inspect the terrain and coordinate system. A unit mismatch or an incorrect reference surface can undermine quantities throughout the study. Have the engineer establish a baseline option using the team’s current process. This gives the pilot something concrete to compare with AI-assisted work rather than a general impression that the new interface feels faster.

Use the enabled OpenSite+ tools to explore a small number of layout and grading alternatives. Keep the same assumptions for each candidate so that differences remain interpretable. If one option moves the building pad or changes a retaining condition, record the change explicitly. Otherwise a lower earthworks estimate can reflect a different problem definition rather than a better solution.

Ask the assistant to locate relevant project requirements and explain the supported design operations. Check the referenced material before applying the recommendation. An answer can be useful as a guide to the model without becoming an engineering approval. Preserve that distinction in the review record, especially when a requirement is ambiguous or has changed since the baseline was prepared.

Evaluate the resulting design objects, quantities and drainage implications through the team’s normal technical review. Look for details that a high-level comparison can hide: transitions near existing surfaces, access constraints and effects at the site boundary. The objective is an alternative that survives professional review, not merely a model that renders without obvious visual defects.

Finally, produce the intended handoff and ask another engineer to review it without relying on the original conversation. Can they identify the assumptions, inspect the chosen option and understand why it was selected? Measure preparation time, correction effort and completeness of that handoff. This tests whether the workflow improves the actual delivery process.

For a developer building an iTwin application, run a separate pilot around data freshness and object identity. Confirm that updates to an engineering model appear against the correct asset and that users understand what has changed. The success of an OpenSite+ design exercise does not establish that a custom operational twin has been integrated correctly.

04 / PricingApplication quotes and developer credits must remain separate

The OpenSite+ page directs buyers to request pricing without establishing a universal public amount. The iTwin pricing page publishes USD developer-platform plans. Its free Community route is expressly for personal, non-commercial and non-production use, so it cannot be treated as the entitlement for a commercial customer deployment.

iTwin credits do not roll over and cannot pay for other Bentley software. The page lists additional-credit rates while its FAQ says Bentley contacts customers about excess usage and the appropriate subscription. Confirm how that combination applies to the agreement instead of treating an unlimited overage calculation as a guaranteed billing rule.

The economic comparison should begin with the route being purchased. An engineering firm adopting a finished application budgets licenses, training and delivery effort. A software team using platform services also budgets development, data integration and operation of its own application. The same company name on both offers does not make those implementation costs interchangeable.

Use a representative month of workload to estimate consumption. Repeated synchronization, storage, visualization and analysis are different service activities. Measure the actual operations required by the proposed application, then test how the estimate changes when a project grows or users revisit a large model. This is more useful than assuming that the number of end users alone determines the bill.

05 / DistinctionsInfrastructure context makes the assistance worth testing

Bentley’s infrastructure AI approach emphasizes connected data, engineering context and user control. The concrete opportunity is to connect language-based interaction to structured engineering work. A question about a model can be more useful when the answer can refer to the actual objects, data and calculations involved.

The infrastructure AI announcement places new applications within a broader engineering portfolio. That breadth matters to owners who need information to survive from design into construction and operation. It does not establish that every application already shares the same AI capabilities or that all transitions are automatic.

Our editorial assessment is that Bentley is most interesting where the organization has valuable infrastructure data and a repeatable engineering question. The evaluation should show that context improves the accepted result. Promotional speed claims and customer demonstrations are not used here as expected productivity multipliers for another project.

06 / QuestionsHosting and feature stage can change the practical fit

The current OpenSite+ FAQ says its AI services use U.S.-hosted Azure infrastructure and that AI features currently require data transfer or storage in the East US region. It also retains language about expanding localization before general release. Confirm the selected product release, region and AI availability in the quote; the public page contains both current product information and forward-looking language.

MCP access also requires a product-specific check. Bentley’s page distinguishes generally accessible routes from restricted Early Access. Review which actions the connector can perform and where they are confined. A connector that can modify design elements should be piloted on controlled copies with an engineer checking the change, rather than being given a production project merely because it supports natural language.

Ask how the workflow represents missing data and outdated project documentation. A conversational answer should not conceal an unindexed document, an unsynchronized model or an unsupported analysis. During the pilot, deliberately include one known gap and check whether the reviewer can recognize it before relying on the output.

07 / DecisionJudge the complete engineering handoff

Bentley Systems offers several routes into infrastructure AI, from a finished site-design application to developer services and engineering connectors. Choose the route that matches the team’s work and validate the final handoff under real project constraints. Expand when another qualified reviewer can reproduce the reasoning and inspect the resulting design or asset record.