Subsoil

Project information across systems

Research on common data environments, who controls project records, and what new tools need from the data they use.

What the studies found

Interoperability costs
$15.8B NIST’s estimate for US capital facilities in 2002. Owners and operators bore about two-thirds. [21]
AI in architecture
74% of practices in RIBA’s 2026 survey reported using AI on at least some projects. [7]
Systems integration
37% of RICS construction survey respondents cited integration among their top three AI adoption barriers in 2025. [8]

Architecture, engineering, and construction (AEC) organizations need project information to remain useful as applications, partners, and responsibilities change. Common data environments help organize delivery, but information still has to cross organizational boundaries. AI and the growing ability to build small applications make those boundaries more consequential: a new tool needs understandable data, appropriate access, and a way to return results that other people can trust.

The Open Container Initiative (OCI) offers a possible shared foundation for portable published versions and linked results. Organizations can choose compatible applications, registries, and hosting without adopting a single provider's product. ORAS supplies tools for working with OCI artifacts; Subsoil is a usable demonstration of this approach. The opportunity depends on maintained integrations, agreed content formats, and sound information management as well as reliable transfer.

Collaboration across several environments

The Infrastructure Client Group describes large projects using several disconnected systems despite attempts to standardize on one CDE. Mandating a client's system can leave delivery partners duplicating information or adapting to a different platform for every customer. Its paper, informed by a November 2025 expert discussion, recommends open information structures, portability, and governance across connected environments. It also recognizes the added integration and governance complexity of an open ecosystem. [1]

Academic research supports the concern about information crossing boundaries. A literature review of 46 publications combined with interviews with 15 BIM experts identifies project complexity and simultaneous use of multiple CDEs as sources of accountability, transparency, and reliability problems. The study is qualitative evidence about recurring difficulties, rather than a measurement of their prevalence across the industry. [2]

The UK BIM Framework distinguishes the CDE workflow from the technical solutions that support it. The client and delivery teams can operate their own systems within one agreed project workflow. Its transfer example shows what happens when one system has a classification field and another models that classification with folders. Someone has to agree on the mapping and may have to enter the metadata again. The receiving team needs the permitted use and revision alongside the file. [3]

NIST's 2004 capital facilities study estimated that inadequate interoperability cost the US industry segment $15.8 billion in 2002. Owners and operators bore about two-thirds of the estimated costs, mostly during operation and maintenance. The estimate covers information exchange and management across the asset life cycle, including paper records. It gives historical context for the cost of recreating usable information at handover. [21]

For a project team, the practical questions are familiar. Which version is this? What was it issued for? Which inputs produced this report? Who can use it? Those questions offer a clearer introduction to a shared delivery architecture than its storage terminology.

Data ownership across organizational boundaries

Keeping control of project data means being able to choose where it is stored, grant access to the information a team is responsible for, and keep usable copies of the records it is entitled to retain. A design firm can keep its working files and issue selected versions. An owner can keep accepted deliverables and their approval records in an environment it controls. Project agreements determine which records are authoritative and what each party can retain, share, and reuse. The Infrastructure Client Group recommends specifying responsibility for transfers, consistent permissions and audit records, and portable handover data when teams use several systems. [1]

Construction research has tested parts of this approach. Petri and colleagues used a UK infrastructure project trial to exchange IFC contributions through a common interface across separate cloud nodes. The participating disciplines could keep their source data where they managed it and share contributions with information about their intended use. Werbrouck and colleagues demonstrated a different exchange in a prototype that assembled issue data from distributed stores and presented it through buildingSMART's BIM Collaboration Format (BCF) API. The studies cover model coordination and issue exchange, respectively. [22][23]

A common interface gives a team a way to replace one compatible system while keeping the way its partners send and receive information. Where an existing CDE requires an adapter, the adapter maps to the shared interface. The parties also need agreement on how to identify a release, interpret its permitted use, and verify who issued or accepted it. OCI can identify and transfer exact file versions. For a governed exchange, test whether an owner can verify the authorized version and its approval history from a retained copy after a system change or the closure of the original project account. [1][15][16][17]

One release across two CDEs

  1. Engineering firmIssues a version from its own CDE.
  2. Shared interfaceCarries the exact version and its approval record.
  3. OwnerKeeps the release and checks what it was approved for.
Proposed workflow. The project defines who can issue, accept, and retain each release.[1]

Dependence on software suppliers

The Nordic architectural associations' September 2022 open letter to Autodesk describes dissatisfaction with licensing costs, unwanted application bundles, interoperability, performance, and responsiveness to user needs. The four associations represented more than 14,000 architects. The letter is evidence of organized historical dissatisfaction; its individual product complaints should not be presented as a current assessment without fresh verification. [4]

Recent practitioner discussions illustrate the operational side of dependence. A February 2026 discussion describes updating training materials, BIM execution plans, and client communications around platform changes. Participants disagree about whether those changes are justified. A separate March 2026 feature request asks for better continuity of responsibilities, reviews, and issue history when a user leaves a project. These are firsthand accounts and requests, not verified product audits or representative surveys. [5][6]

The implication for a shared delivery foundation is practical freedom to choose and change compatible tools. A credible demonstration should preserve exact publications and the relationships it claims to carry when applications or hosting providers change. It should also make the ongoing integration and operating work visible. An open interface alone does not eliminate migration costs or make proprietary native files universally editable.

AI adoption and the information it depends on

RIBA's 2026 survey reports that 74% of architectural practices use AI in at least some projects. Its accompanying discussion describes productivity benefits alongside concerns about intellectual property, careers, and other effects on professional practice. [7]

RICS's 2025 construction survey received more than 2,200 responses globally, 48% from the UK. When asked to select their top three adoption barriers, respondents most often selected skills shortages (46%), integration with existing systems (37%), and data quality and availability (30%). Implementation cost (29%), unclear return on investment (28%), and privacy and security (22%) were also concerns. These are reported perceptions. The survey covers a different population and period from RIBA's and is not directly comparable. [8]

The resulting opportunity is to reduce the effort required to connect useful applications to understandable information. Exact versions can anchor an analysis to known inputs. Linked results can help people investigate how an answer was obtained. Structured content can support queries when its fields, units, and meaning are agreed. These are ingredients for an AI workflow; correct answers still require appropriate methods, data quality, and review.

Six most-cited barriers to AI adoption

  1. Skills shortages46%
  2. Integration with existing systems37%
  3. Data quality and availability30%
  4. Implementation cost29%
  5. Unclear return on investment28%
  6. Lack of standards and guidance25%
RICS, Q1 2025. The survey received more than 2,200 responses globally. Respondents could select up to three barriers. Bar lengths use a 0–50% scale.[8]

Teams are beginning to build their own tools

Stantec's design-computing leader identifies AI-assisted creation of small, specialized applications as a 2026 trend. His account describes rapid prototyping while cautioning that improvements to individual tasks have not yet produced an equivalent acceleration of whole projects. He also identifies inconsistent information across design and business systems as a barrier to deeper analysis and AI, and raises questions about data ownership. This is a practitioner's account and outlook, rather than an industry-wide productivity study. [9]

Broader evidence shows substantial use of AI for application development. Anthropic's April 2025 analysis of 500,000 coding interactions found frequent work on interfaces and web applications. Its sample concerns its own products and excludes Team, Enterprise, and API usage from the Claude.ai portion; it cannot establish AEC adoption or production software quality. [10]

An implication for infrastructure is that more people may be able to prototype a useful tool while still needing help with its connection to real work. Access control, version selection, file interpretation, error handling, and result publication remain necessary. A common contract can make some of that work reusable. Supported examples and maintained integrations are needed to turn the contract into something an in-house developer can use.

Agents need information and capabilities

Whether an agent should operate on software or data is best understood as two related responsibilities. Data supplies the context and evidence for a task. Software supplies the operations that can be performed on that information. The Model Context Protocol (MCP) reflects this distinction through resources that expose context and tools that can query systems, invoke APIs, or perform computations. MCP is one possible integration mechanism; it does not require OCI, and OCI does not require MCP. [19][20]

For an AEC workflow, the useful contract is an authorized input version, a defined operation, and a result connected to its inputs. An agent might request quantities from a supported model, invoke an engineering calculation, and prepare a report for review. Authoring changes may need the source application's geometry engine, constraints, or transaction API. Structured access can reduce dependence on a particular screen layout, while interface automation remains a possible route where better integrations are unavailable.

OCI can deliver the input and output artifacts. It does not execute the operation, make a model understandable to an agent, or provide the agent's authorization and execution policy. Query services, adapters, domain engines, and access controls complete the workflow. Keeping these responsibilities explicit allows teams to change agents and applications while retaining inspectable project records.

Security and responsibility across tools

Construction information crosses organizational and contractual boundaries. NCSC's construction joint-venture guidance addresses how participating businesses should protect sensitive information through a coordinated security approach. Its advice places responsibility on management as well as technical specialists. [11]

Agentic tools introduce additional questions about what software can access and do. NCSC's August 2026 interim advice recommends constrained environments, limited credentials, observation of activity, and appropriate human oversight. It explicitly advises against relying on prompts alone. [12]

Access should be explained through concrete actions: an application can read a selected version, or publish to a permitted destination. Claims about secure automation need additional evidence about the application, its execution environment, and operating procedures. A content fingerprint detects changed bytes; it does not establish that a model is correct or approved. Hosting under an organization's control does not, by itself, establish legal ownership, permitted AI training, or compliance. These remain separate responsibilities.

An active and changing market

Incumbents are actively adapting to AI. Autodesk's April 2026 announcements describe granular data APIs, MCP integrations including a Revit preview, and an expanding marketplace for applications and agent tools. Capabilities in preview and planned integrations should be distinguished from generally available products. [13]

Other providers also address information access and reuse. Speckle describes model-based AI queries, visibility into their inputs, and analysis across projects. These are the provider's product claims, not independently assessed performance findings. Meanwhile, buildingSMART's openCDE Documents API already defines file selection, download, and upload between compatible clients and CDEs. [14][15]

An OCI-based proposal therefore needs to explain its particular role and how it can fit existing workflows. OCI provides content distribution and artifact associations; domain profiles define what the delivered files mean. The proposed value is portable publications and linked results across independently operated stores using a shared interface. It must be demonstrated with maintained applications and explicit content agreements. [16][17]

Subsoil is one usable demonstration, not a required product or exclusive gateway. ORAS is an OCI artifact tool, rather than a separate domain standard. Other compatible clients and registries can participate. Common transport does not require every participant to use Subsoil's model conversion, data profile, browser, or application-launch flow. Those are implementation choices and additional agreements, distinct from the OCI specifications. [18]

A useful way to explain the opportunity

Begin with continuity: project information should remain useful as teams and tools change. Show the effort involved in moving it between systems, then introduce the growing ability to build specialized software. The foundation becomes relevant when people see why each new application needs dependable inputs, appropriate access, and a place for results.

Federation, version comparison, specialist applications, and learning across projects can then serve as practical examples. A federation application selects exact contributions and defines their alignment. A comparison application explains changes it can identify. A focused application publishes a result against its inputs. Portfolio analysis requires consistent measures and rights to use the information. The delivery layer supports these activities without supplying their domain logic.

The smallest persuasive demonstration connects two maintained applications to an exact publication, produces a useful result linked to that input, and transfers the supported package between independently operated stores. Assess the time needed to connect the applications, the information preserved, and whether another person can reproduce the handoff. The current implementation and integration contracts are described in the project overview and development guide.

Sources

  1. Institution of Civil Engineers, Infrastructure Client Group. The Future of Common Data Environments. Expert discussion held November 2025.↩1 ↩2 ↩3 ↩4
  2. Jaskula, K., Kifokeris, D., Papadonikolaki, E., and Rovas, D. Common Data Environments in construction: State-of-the-art and challenges for practical implementation. UCL author manuscript, 2024.↩1
  3. UK BIM Framework. Guidance Part C: Facilitating the common data environment workflow and technical solutions. Edition 1, September 2020.↩1
  4. AiN, ATL, Danske Arkitektvirksomheder, and SAMARK. Open letter to Autodesk. September 12, 2022; reproduced by AEC Magazine.↩1
  5. r/bim. Practitioner discussion of Autodesk cloud platform changes. February 2026.↩1
  6. Autodesk Community. Autodesk Construction Cloud user level handover. User-submitted feature request, March 23, 2026.↩1
  7. Royal Institute of British Architects. RIBA AI Report 2026.↩1 ↩2
  8. Royal Institution of Chartered Surveyors. Artificial intelligence in construction report 2025. Q1 2025 Global Construction Monitor responses.↩1 ↩2 ↩3
  9. Mullins, B., Stantec. 2026 AEC trends: AI in architecture, vibe coding, real-time analysis, and data. February 19, 2026.↩1
  10. Anthropic. Economic Index: AI's impact on software development. April 28, 2025.↩1
  11. National Cyber Security Centre and industry partners. Joint Ventures in the Construction Sector. 2022.↩1
  12. National Cyber Security Centre. Managing the cyber risk of agentic AI. August 20, 2026; interim advice.↩1
  13. Autodesk Platform Services. Building for Agentic AI: What's New in Autodesk Platform Services. April 15, 2026.↩1
  14. Speckle. Design data and intelligence platform. Product descriptions accessed November 4, 2026.↩1
  15. buildingSMART. openCDE Documents API. Release 1.0 specification.↩1 ↩2
  16. Open Container Initiative. Distribution Specification 1.1.1.↩1 ↩2
  17. Open Container Initiative. Image Manifest Specification 1.1.1, including artifact usage and subject associations.↩1 ↩2
  18. ORAS. Pushing and pulling OCI artifacts and copying artifacts between registries.↩1
  19. Model Context Protocol. Resources. Specification dated November 25, 2025.↩1
  20. Model Context Protocol. Tools. Specification dated November 25, 2025.↩1
  21. Gallaher, M. P., O'Connor, A. C., Dettbarn, J. L., Jr., and Gilday, L. T. Cost Analysis of Inadequate Interoperability in the U.S. Capital Facilities Industry. NIST GCR 04-867, 2004. Estimates for 2002.↩1 ↩2
  22. Petri, I., Beach, T., Rana, O. F., and Rezgui, Y. Coordinating multi-site construction projects using federated clouds. Automation in Construction, 83, 273–284, 2017.↩1
  23. Werbrouck, J., Schulz, O., Oraskari, J., Mannens, E., Pauwels, P., and Beetz, J. A generic framework for federated CDEs applied to Issue Management. Advanced Engineering Informatics, 58, 102136, 2023.↩1