Some people asked me what I think of AI in the new Revit 2027. Other people asked me what I think of the entire Autodesk Cloud ecosystem being rebranded as Forma. More people requested I express an opinion on the lack of mention to agentic AI within the new proposed ISO 19650. As it turns out, these three themes tie together. Let’s see how, starting from ISO 19650.
1. Two More Words about the Norm
There’s a moment, in every standards revision or drafting, where the gap between aspiration and reality becomes visible, and we’re in that moment now.
The draft for ISO 19650-1: 2026 has been in public consultation since early March (I talked about it here and here), and a part of the industry is fighting about a supposed rebrand where BIM is being replaced by Information Management, which means treating the standard as a terminology update. But beneath an language shift is a more fundamental question, as we have started to see: who decides? Who owns the data? Who is responsible when decisions are made? These, however, aren’t questions the standard can answer alone: they’re questions about power, control, and the alignment (or misalignment) between the frameworks we write and the tools we build, whether they’re software, contractual tools or smart aids for management.
And now, with agentic AI rampaging through the AEC industry, there’s a third player in this conversation, one that the standard doesn’t yet know how to think about and refrains from addressing, possibly out of awareness that the scenario is rapidly shifting and this is a norm built to last several years.
So, let’s try to unbox what’s happening in relation to automation and a certain simplification of tools that seems to be the trend, shall we?
1.1. The Standard’s Ambition
ISO 19650-1:2026 seeks to revise key principles of information governance in construction. As we have seen, the current framework (2018) defined roles — appointing parties, appointed parties, and lead appointed parties — and defined documents — OIRs, AIRs, PIRs — and created a structure for who asks for information and who produces it. The revision goes deeper in asking what the purpose of this information is, who actually needs to decide, and what assumptions are we baking into the system.
To some people, the move from BIM to Information Management signals a reclamation, as if BIM meant ‘models’. BIM as a practice has always been about managed information. But there’s been a chronic conflation: the model as container vs. the model as one source of data among many. ISO 19650-2026 is pushing hard toward distributed information organised around stakeholder need, not around model convenience. Information is produced for purposes and through the awareness that a structural engineer needs different data than a facility manager needs different data than a cost estimator. The standard wants to name this explicitly: you’re not producing a model that everyone uses; you’re producing information that serves different purposes across the lifecycle.
This is important because it decouples information management from the modeling tool, just as the Common Data Environment is explicitly reframed as a framework to keep pushing that the term was never meant to signify a sharing platform. When it comes to information management, the standard in theory is saying: organize your information around what stakeholders actually need, not around what your information model happens to contain. The tool serves the information strategy; the information strategy doesn’t serve the tool.
In theory.
2. The Vendor Problem and Forma’s Evasion
Autodesk Construction Cloud achieved BSI Kitemark certification for compliance with ISO 19650 in early 2026. On the surface, this is a win. It suggests that this major vendor is genuinely engaging with governance and control, building tools that support the standard’s intent.
To understand what’s really happening, you have to zoom out, forget the damn certifications and see what’s being built around the stamp.
The document management system that passed the ISO 19650 audit has been rebranded as Forma Data Management, Autodesk Construction Cloud has been folded into something called the Forma Industry Cloud and and at the center of this ecosystem, Autodesk is building design tools (Forma Building Design and Forma Site Design), coordination tools (Forma Board), and analysis tools (the embodied carbon tracking in Forma Site Design), all feeding into the same data environment.
This is a fundamental architectural choice that reveals something important about what “compliance” actually means when the vendor controls the entire stack.
2.1. Enters Autodesk Forma
ISO 19650-1:2026 emphasises governance. The revision proposes stronger whole-life integration from project initiation through operation, greater emphasis on appointing party accountability and governance, and a clearer separation between managing information and producing information. These are not trivial requirements. They imply a degree of independence. They imply that the party managing information (the client, the asset owner) has insulation from the party producing it (the designer, the contractor).
But when Autodesk provides the design tools and the data management system and the coordination layer, that separation collapses: a single vendor is simultaneously producing information, managing it, validating it, and hosting it. The common data environment — which ISO 19650 imagines as neutral territory — becomes proprietary infrastructure that also hosts analysis, production, validation.
Is this necessarily a violation of ISO 19650? The BSI certification seems to say no. But the certification obviously didn’t assess the philosophical alignment between the standard’s intent and the positioning of the tool within its larger architecture. The standard tells the client they should decide what information they need. The tool ecosystem poses itself as the single place where information is produced, flows, is validated. The tool vendor isn’t denying governance, but they’re making certain decisions automatic, and automation is harder to govern than choice.
This would be less concerning if there were regulatory pressure. But there isn’t.
2.2. The Context we need to understand
ISO 19650 adoption is mandated in the UK, driven by the Information Management Mandate that has been in force since 2016. It’s mandated in Singapore, Germany, the UAE, across Europe and the Middle East. In these markets, vendors have to build to an external standard. They can’t treat the Common Data Environment as their proprietary asset. Clients can audit compliance. If Autodesk’s approach to data governance doesn’t match the standard, clients have alternatives.
The United States has no federal BIM mandate. There is no ISO 19650 requirement. The US is the world’s largest software AEC market, accounting for $15–51 billion globally, and it’s effectively unregulated on this front.
What does that mean in practice? It means US vendors have no incentive to strive for interoperability, because they see it as relinquishing control. Autodesk can iterate Forma in ways that suit their business model, their product vision, their ecosystem strategy, not necessarily taking into consideration the industry’s governance needs. When there’s no outside pressure, consolidation becomes the natural economic outcome.
A large contractor in the US can use Revit in combo with Forma Data Management (our old friendly Docs) for the entire project lifecycle without ever applying or even knowing ISO 19650, because they get integration without interoperability through a unified data environment that doesn’t need any (as long as you keep things simple). They get what feels like governance because everything is in one place. But they don’t get independence from the vendor, they don’t have the option to swap out components with specialist ones (and God forbid they have complex steel carpentry that requires Tekla), or to audit whether their data management truly aligns with any kind of external standard, such as one needed for running operations with the model they’re building. Also, they don’t get competency in specific fields because, as big as Autodesk might be, their focus on areas such Asset Management is relatively new and still fairly shallow.
The revision to ISO 19650 is being proposed precisely because the industry learned, over eight years of implementation in mandate-driven markets, that governance matters, and that distributed information management is too important to leave to a vendor’s whim, and that you need standards to clarify roles, separate gut-worrying from management concerns, and create space for healthy competition.
Simultaneously, the US seems to have zero pressure to adopt these lessons. The vendors can consolidate. The standards can remain an afterthought. Regardless of smaller software houses advocating for a higher awareness (see Revitzo here), adoption remains optional and concentrated primarily among firms bidding for international projects, and the US AEC industry can continue treating information management as a tooling problem rather than a governance problem.
Moreover, there’s a question any conversation is currently avoiding: who owns the information?
2.3. No Handover, No Party
In the ISO 19650 model, the appointing party (the client, the asset owner) owns the information requirements. They define what they need. The appointed parties produce the information. The common data environment sits in between, governed by protocols. At the end of a phase in the cycle, there’s a handover that’s often mandated by a technological need: the primary solution for the Common Data Environment (say, Forma Data Management) isn’t suitable for asset management, therefore a handover needs to take place. Handover means cleaning and wrapping up information, containers, deliverables, and delivering them to the client alongside rights of use for that data.
But in the Forma ecosystem, Revit data lives in Forma, analysis runs in Forma, coordination happens in Forma, and Forma Data Management wraps everything up. Revit becomes a “Forma Connected Client”, which means the data relationship is one-way. Forma is the source. Revit is the client.
For a firm operating under ISO 19650, this creates a subtle but significant problem: you’ve defined your information requirements, you’ve chosen your tools, but the tools have already defined how information flows through the system in a way that’s appealing because it’s easy. The vendor’s architecture is now manipulating your governance model into thinking you’re working efficiently. You can govern within the system, but you can’t govern the system itself because that’s locked inside Forma’s infrastructure.
This isn’t malice, there isn’t an Autodesk conspiracy to work against ISO 19650 (at least I think there isn’t) but there also isn’t any will to operate towards it, and how can you blame them? Vendors build platforms. Platforms make assumptions. These assumptions, accepted without standards that are tool-agnostic, become constraints. And constraints, over time, become the default path. In spite of standards.
3. Enters AI (and things get messy)
In its governance effort, and out of a strife to stay relevant through an amount of time that’s proportionate to the amount of work it took, DIS/ISO 19650-1:2026 doesn’t mention agents, autonomous workflows and agentic AI in particular.
Standards move slowly, but there are speculations that are reasonable as to the general direction the technology is taking. The ISO 19650 revision, with public consultation underway and final publication expected by late 2026, aims to represent one of the most significant shifts in digital construction standards since 2018. By the time the revision lands in final form, the industry will already be experimenting with agentic workflows that the standard doesn’t contemplate.
3.1. Why Agents and not AI
Traditional BIM workflow assumes human agency, if not always natural intelligence, with a certain degree of automation: a designer makes a choice, a coordinator resolves a clash. Each decision has an author, a timestamp, an audit trail. ISO 19650 mandates all of this: responsible parties, information production schedules, version control, status codes, a responsibility matrix (even if they plan to call it by a different name). It’s built on the premise that decisions can be traced back to decision-makers.
Agents disrupt this model fundamentally, and many industries are still struggling to figure out how to address this issue.
Let’s make up a scenario to clarify this: an agent trained on building code generates 20 layout options for the restrooms in a residential floor plan, and another agent assigns indexes based on parameters that might be functional or aesthetic. The designer selects one based on their experience and on these indices. Did the agent decide, or did it just generate possibilities and evaluate them? And when the first one generated possibilities, who is ultimately responsible for evaluating their quality? The designer is influenced by the agents that assign indices, and picks without being able to audit the logic, not because the agent is secretive, but because explaining why “this layout maximises natural light while satisfying egress requirements while optimising for leasing flexibility” isn’t a question the agent is generally trained to answer.
ISO 19650 has no vocabulary for this. It assumes decision-makers always understand why they decided and can’t say “the AI made me do this” during the audit trial. It assumes information production follows a prescribed process. Agents invert both assumptions.
3.2. Responsibility Without Competence
The market related to Information Management in Construction (let’s say the “BIM market”) is growing from $9.03 billion in 2025 to a projected $15.42 billion by 2030, which means a 11.3% annual growth. The accessible Cloud has been driving the expansion, and so will AI. Alongside that growth, we’ll see, and we’re already seeing a proliferation of agentic tools: platforms that automatically pre-validate models, detect inconsistencies, and suggest optimisations. Autodesk claims to be releasing Neural CAD (agentic generation of building layouts), and Revit 2027 arrived with an AI partner nobody asked for.
Even if these tools are genuinely useful, and that’s a big if right now, they’re also genuinely dangerous when deployed without competence, as we know.
Let’s make another example.
Imagine a client (appointing party) using a third-party agentic validator to sign off on design submissions without. And you can be sure that these tools are coming. The client might not understand what the tool is validating against, because AI isn’t regulated enough to mandate that they explain what they’re doing. The tool says “code-compliant.” But compliant with what, my friend? The client’s information requirements, regardless of the local building code? The local building code of this place? The local building code of somewhere else? Some trained-on dataset from 10,000 similar projects in the vendor’s database? You don’t know. You can’t know. If you wanted to know, you would be using Solibri. You approve because you trust the tool.
But you’re still liable for what you approve.
ISO 19650 assumes the appointing party knows what they’re requiring and is able to see whether the requirements are met, which is reasonable. An agent-mediated workflow makes that assumption optional. You can outsource to the agent without understanding what you’re outsourcing, and you don’t lose liability when you delegate because that would mean the AI agent (and its developer) is liable in your place.
4. The Control Problem
And here’s where the pieces converge: a vendor who controls the design tool (Revit, Forma Building Design), the data management layer (Forma Data Management), and the coordination/validation layer (Forma Board, clash detection) is planning to introduce an agent into that stack. Planning and failing, because the recent integration of AI within Revit is laughable and massively sucks.
Even while not working, the agent makes suggestions, runs validations, maybe it even implements routine decisions. The vendor frames this as automation, a feature that accelerates workflows, reduces tedium, makes your team faster. Except, automation runs on standards, and this is black-boxish AI: from a governance standpoint, you’ve handed decision-making authority to an AI trained on the vendor’s data, following the vendor’s logic, constrained by the vendor’s architecture.
In this scenario, ISO 19650-1: 2026 tries to clarify who is responsible for information, but it doesn’t contemplate a scenario where information is produced by an agent whose training data is proprietary, whose decision criteria are opaque, whose architecture is locked inside a vendor ecosystem.
In short, the standard assumes human agency, and the industry is moving toward delegated agency. And in that gap, governance becomes a question of power — who has the leverage to enforce transparency, independence, and control — rather than principle.
We Need New Vocabulary (and New Structures)
Three years ago, the problem was tractable: who decides what information is required? ISO 19650-1: 2018 (followed by part 2 the next year) sanctioned an answer already experimented in years of British standards. The 2026 revision refines it. But both were written for a world where decisions are still made by humans with agency and accountability.
That world is already changing.
We need a totally new vocabulary, not just linguistic fine-tuning, for genuinely new concepts, such as:
- Agentic responsibility: if an agent makes a decision (or proposes one), who is liable? The designer who didn’t audit it? The vendor who trained it? The organization that deployed it?
- Competence boundaries: when is it acceptable to use a tool you don’t fully understand? Standards can mandate audits and traceability. They can’t mandate understanding. That’s a practice question.
- Vendor sovereignty: how much control over information management can a single vendor have before they become a governance risk? ISO 19650 assumes vendors are tool providers. What if they’re also decision-makers?
- Data independence: can a firm truly govern its information if the data lives in a vendor’s silo? The standard talks about common data environments. But if “common” means “proprietary,” what have we actually standardized?
These aren’t questions ISO 19650 can resolve alone, as they require changes at multiple levels:
- clearer language around agency and automation in future standards;
- contractual frameworks that allocate responsibility for agent-driven decisions;
- vendor practices that separate the tools that make decisions from the tools that govern them;
- professional education that teaches accountability alongside tool literacy.
For now, the standard clarifies roles, mandates traceability, insists on governance, and these aren’t trivial achievements. They’re also not sufficient for the world we’re entering. The conversation between standards, vendors, and agents is just beginning, while practitioners — designers, engineers, BIM managers — are experimenting with tools they can’t fully understand, on projects where the stakes are real: if we’re smart, we’ll shape this conversation before it shapes us. If we’re not, we’ll wake up in five years and realize that governance was solved by the vendors, in their favor, and without asking our permission.
The time to ask questions is now, while the standard is still being written, while the tools are still being built, while there’s still space to insist that information management is too important to be left to default.
No Comments