A $1.4B regional builder spent eighteen months and several million dollars standing up an AI program: licenses for Procore Assist, a pilot of Augmenta routing, a Document Crunch deployment, a data lake, a steering committee that met monthly. Then the audit came. Nobody could say who owned the verification gates that were supposed to sit in front of the stamp, the pay app, and the safety plan. The model-governance addendum had been drafted but never assigned. The risk register existed as a spreadsheet that three people edited and none owned. When a hallucinated code citation slipped into a permit set and the AHJ caught it, the post-mortem found the program had no single owner, so it belonged to no one, and the gates that the program had so carefully designed were operating on the honor system. The lesson the firm learned the expensive way is the one this lesson teaches up front: AI does not run on tools, it runs on roles, and a role that is not on the org chart with a name, a reporting line, and a compensation band is a role that does not exist. By the end you will be able to define the five new roles an AI-enabled AEC firm needs, place each one in the org chart so it reinforces the line organization instead of forming a parallel silo, set a defensible compensation band for each, and produce the deliverable: the org chart and the job descriptions.

Why AI Creates Roles the Firm Must Actually Define

The reflex when a firm adopts AI is to treat it as a tooling decision: buy the licenses, run the pilots, train the staff, and let the existing organization absorb the new capability the way it absorbed email or BIM. That reflex fails predictably, because AI introduces standing obligations no existing role owns. Someone has to own the data the models run on, its quality, lineage, and access. Someone has to own the verification gates that stand in front of the stamp, the schedule, the pay app, and the safety plan. Someone has to own the bias and fairness questions and the workforce-displacement questions the firm's leadership has decided are not optional. Someone has to own the prompts and patterns that determine whether the firm gets reliable output or expensive noise. And someone has to own the risk register and the compliance posture that keeps the firm defensible when a regulator, an insurer, or an opposing expert asks who was in charge. These obligations do not map cleanly onto the PM, the VDC manager, or the general counsel, which is why the firm must define new roles rather than hope the work gets done in the seams.

The controlling analogy is the trade. When a firm takes on a new scope it has never self-performed, it does not ask the existing trades to absorb it in their spare time; it hires or subcontracts the trade, and either way it names who is responsible, where they report, and what they are paid. AI is a new scope. It has its own materials (data), quality control (the gates), code compliance (governance and law), craft (prompting and pattern design), and safety officer (the compliance and risk function). Treating it as a hobby the existing trades pick up between tasks produces exactly what the $1.4B builder discovered: real money spent, real capability built, and no one accountable when it fails. The five roles this lesson defines are the named trades of the AI scope, and the discipline is to put them on the chart with the same seriousness the firm gives any responsible position.

There is a second reason to define rather than improvise the roles: the program's gates and obligations from earlier levels need owners or they decay. The five verification gates (design intent, code, contract authority, dollars, life-safety), the cardinal rule that the human verifies before the stamp, the schedule, the pay app, and the safety plan, the data and model governance from the CDE addendum, and the workforce and ethics commitments are all designs that assume an owner. A gate with no owner operates on whoever happens to remember it, which is to say no one. The new roles are how the firm operationalizes the designs it has already made, turning policy documents into standing accountabilities held by named people on the chart.

The Five Roles and What Each One Owns

The first role is the Director of VDC + AI, the evolution of the existing VDC or BIM manager into the operational owner of how AI is applied to the work: model coordination, clash detection, scan-to-BIM, quantity takeoff, drawing comparison, generative design exploration, and the integration of tools like Augmenta, Forma, and the firm's BIM stack into the production workflow. This role owns how AI gets used on projects day to day and whether it makes the model and production process faster and more reliable. It is deliberately built on an existing competency because the firm already trusts its VDC leadership with the federated model and the LOD progression; extending that trust to the AI-augmented version of the same work keeps the role inside the line organization rather than off to the side.

The second role is the Chief Data Officer, who owns the firm's data as an asset: the Common Data Environment, the IFC and COBie deliverables, the model element authorship under AIA E203 and G202, the training data, the prompt libraries, and the AI-generated geometry, plus data quality, lineage, access control, and the data-governance addendum the governance lessons produced. Where the Director of VDC + AI owns how AI is used, the CDO owns what it runs on. This split matters: a firm that lets its production leaders also own the data tends to optimize it for today's project at the expense of the asset, and a firm with no data owner finds its models run on whatever is convenient, which is how lineage and access decay into the audit finding.

The third role is the AI Ethicist. This role owns the bias and fairness questions and the workforce questions that the ethics-and-equity lesson established as firm commitments: whether AI-assisted decisions disadvantage protected groups, whether the firm's MWBE and DBE participation holds under AI-assisted procurement, whether labor displacement is being managed with retraining rather than ignored, and whether prevailing-wage obligations on AI-assisted federal work are honored. The fourth role is the Prompt Engineer-in-Residence, who owns the firm's prompts and patterns as reusable craft: building, testing, and maintaining the prompt libraries, the retrieval patterns, and the evaluation harnesses that determine whether the firm's AI produces reliable output. The fifth role is the AI Compliance Lead, who owns the verification gates and the risk register: the standing accountability for whether the gates are actually operating, whether the cardinal rule is being followed before the stamp, schedule, pay app, and safety plan, and whether the risk register is current and owned rather than an orphaned spreadsheet. These two functions, the ethics owner and the compliance owner, are the ones the firm is most tempted to leave undefined, and whose absence produced the audit finding.

A role that is not on the org chart with a name, a reporting line, and a compensation band is a role that does not exist; the gates the firm so carefully designed will operate on the honor system until someone owns them.

How These Roles Operationalize the Program's Gates

The reason to define these roles is not org-chart tidiness; it is that they operationalize the gates and commitments the program has spent five levels building. The AI Compliance Lead owns the five verification gates and the risk register, the load-bearing assignment. The gates (design intent, code, contract authority, dollars, life-safety) and the cardinal rule that the human verifies before the stamp, the schedule, the pay app, and the safety plan are not self-executing; they require someone whose standing job is to confirm they are operating, maintain the risk register that tracks where AI is used and what could go wrong, and escalate when a gate is being bypassed. Without this owner, the gates are documentation, and documentation that no one owns is what the $1.4B builder had when the hallucinated citation reached the AHJ.

The AI Ethicist owns the bias, fairness, and workforce questions, the operational form of the firm's ethics-and-equity commitments. These do not resolve themselves either: someone has to actually run the fairness checks on AI-assisted decisions, track the MWBE and DBE participation under AI-assisted procurement, own the retraining plan rather than let displacement happen by default, and confirm prevailing-wage compliance on AI-assisted federal work. Assigning these to a named role with authority is the difference between a firm that has ethics commitments and one that honors them. The Chief Data Officer operationalizes the data and model governance from the CDE addendum, the Prompt Engineer-in-Residence operationalizes the prompt-engineering discipline so reliability is built rather than hoped for, and the Director of VDC + AI operationalizes the production application of the four engines (generative AI, computer vision, predictive ML, generative design). Each role is a gate or a discipline given an owner, the only way a design becomes a standing practice.

This is also why the roles cannot be hollow titles. An AI Compliance Lead who cannot stop a deliverable from going out, or an AI Ethicist who cannot halt a procurement pattern that fails a fairness check, operationalizes nothing. The authority has to match the ownership: the compliance and ethics roles in particular need a real escalation path and a real stop, or they become the advisory function the line organization politely ignores, which brings us to the central placement question.

Where They Sit: Reinforce the Line, Do Not Build a Silo

The most common failure in AI org design is not failing to create the roles; it is creating them as a parallel innovation silo the line organization ignores. A firm stands up an AI center of excellence, staffs it with the new roles, gives it a budget and a charter, then watches the project teams keep working the way they always have because the AI group is something that happens over there, to other people. The roles exist, the org chart shows them, and they operationalize nothing because they have no purchase on the actual work. This silo failure is more insidious than the no-owner failure because it looks like success on paper.

The defense is to place the roles so they reinforce the line organization rather than orbit it. The Director of VDC + AI belongs in operations, reporting through the construction or VDC leadership that already owns production, because that is where AI gets applied to the work; placing it in an innovation silo guarantees it never touches a project. The Chief Data Officer is truly an enterprise function and reports at the executive level, often to the COO or CIO, because data is an asset that spans projects, but the CDO must have working lines into the project teams so the data governance is lived rather than declared. The Prompt Engineer-in-Residence is a center-of-excellence role that should be embedded with the production teams it serves, building patterns for the actual workflows rather than in the abstract. The AI Compliance Lead and the AI Ethicist need independence from the production pressure they police, so they report into risk, legal, or quality rather than into the operations leaders whose deliverables they gate, which gives the stop authority its teeth.

The principle is the matrix, not the silo. The new roles have functional homes (the CDO in the enterprise, the compliance and ethics roles in risk and quality, the VDC + AI director in operations), but every one has a working relationship with the line organization, because the work happens on projects and a role that cannot reach the projects operationalizes nothing. The test for any placement is simple: can this role actually change what a project team does on Tuesday? If not, the role is a silo, however impressive the title, and the firm has rebuilt the failure it was trying to avoid.

What They Are Paid: Defensible Compensation Bands

Compensation makes the role real, and a band that is set by guesswork is a band that either fails to attract the talent or overpays for it. The discipline is to anchor each band to the existing role it most resembles and then adjust for the AI scarcity premium and the seniority of the accountability. The Director of VDC + AI anchors to the firm's senior VDC or BIM director band, adjusted up for the AI scope, because it is a director-level operational role built on an existing competency. The Chief Data Officer anchors to the firm's other chief-officer or vice-president bands, because it is a genuine executive function owning an enterprise asset, and it sits at the high end of the new roles. The AI Compliance Lead anchors to the firm's senior risk, quality, or compliance manager bands, because it is a senior individual-contributor-plus accountability with stop authority.

The AI Ethicist is the role with the widest market range because it is the newest and least standardized; the defensible approach is to anchor it to a senior policy, legal, or risk specialist band rather than invent a number, and to be explicit that the role may start as a fractional or shared responsibility at a smaller firm before it justifies a dedicated seat. The Prompt Engineer-in-Residence anchors to a senior technical specialist or solutions-architect band, reflecting that it is a craft role with a real but bounded scope. The principle for every band is the same: name the comparable existing role, state the adjustment and why, and tie the number to the accountability, so the band survives a CFO's question and a candidate's negotiation. Inventing precise salary figures without anchoring them to the firm's actual bands and market is the kind of unsupported number the program warns against; the deliverable states bands as ranges anchored to comparables, not invented point figures.

Two practical notes keep the bands honest. First, at a smaller firm several roles are responsibilities before they are headcount: one senior person may carry the compliance and ethics accountabilities, and the prompt engineering may live with the VDC + AI director, with the bands reflecting the combined scope. Second, the bands must be defensible internally, meaning a peer in an existing role at the same level should see the new role as comparable in accountability, or the firm creates the resentment that quietly kills the program. Compensation is not a detail; it is the signal of how seriously the firm takes the scope, the difference between a role people compete for and a title no one wants.

Avoiding the Parallel Silo: The Integration Discipline

The silo risk is worth a dedicated discipline because it is the failure that defeats well-intentioned firms. The integration test has three parts. First, every new role has a named working relationship with a line function, written into the JD, so the role's success is measured partly by what the line organization actually does differently. Second, the new roles do not own the work; they own the discipline applied to it. The Director of VDC + AI does not take projects away from PMs; the AI Compliance Lead does not stamp anything; the CDO does not author the model. They own the standards, gates, data, and patterns, and the line organization owns the deliverables, which keeps accountability for the work where it belongs (with the licensed professional, the PM, the estimator) and accountability for the AI discipline with the new roles.

Third, the roles are introduced into the project workflow at named points rather than as a general overlay. The Prompt Engineer-in-Residence builds the patterns the PE uses on the RFI workflow; the AI Compliance Lead's gate is the verification step before the stamp; the CDO's governance shows up as the data standard the VDC team follows. When the role's contribution is a concrete step in an existing workflow, it cannot become a silo, because it is woven into the work the line organization already does. The contrast is the innovation lab that produces impressive demos and influences zero pay applications, which is the silo at its most seductive.

The Applied Problem: Produce the Org Chart and the Job Descriptions

Here is the exercise. Produce two deliverables for your firm. First, the AI org chart: place the five roles (Director of VDC + AI, Chief Data Officer, AI Ethicist, Prompt Engineer-in-Residence, AI Compliance Lead) with their reporting lines and their working relationships into the line organization, designed so each role reinforces the line rather than forming a parallel silo. Show the functional homes (VDC + AI in operations, CDO at the executive level, compliance and ethics in risk or quality, prompt engineering in a center of excellence embedded with production) and the dotted-line working relationships into the project teams.

Second, the five job descriptions, one per role. Each JD states what the role owns (the specific gate, asset, or discipline, tying the AI Compliance Lead to the verification gates and the risk register and the AI Ethicist to the bias, fairness, and workforce questions), where it sits (the reporting line and the working relationships into the line organization), what authority it carries (especially the stop authority for the compliance and ethics roles), and the compensation band stated as a range anchored to a named comparable existing role with the adjustment and its reason. Write the JDs so a peer in an existing role would see the band as defensible and so the line organization can see exactly how the new role changes what a project team does.

The deliverable is the org chart and the five JDs, and the lasting product is an AI organization that operationalizes the program's gates and commitments through named, paid, well-placed roles rather than the honor system. The professional who masters this gives the firm's AI program an owner, places that ownership so it reinforces the line organization instead of orbiting it, sets compensation that signals the scope is serious, and ties the most consequential accountabilities (the gates and the risk register to the AI Compliance Lead, the bias, fairness, and workforce questions to the AI Ethicist) to people who can actually stop a bad deliverable, the only way an AI program survives the audit the $1.4B builder failed.

Key Takeaways

• AI is a new scope, not a tooling decision, and like any new scope it needs named trades: a role that is not on the org chart with a name, a reporting line, and a compensation band is a role that does not exist, and the gates the firm designed will operate on the honor system until someone owns them.
• The five roles are the Director of VDC + AI (how AI is applied to the work), the Chief Data Officer (the data the models run on, including the CDE, IFC, COBie, and model element authorship), the AI Ethicist (bias, fairness, and workforce questions), the Prompt Engineer-in-Residence (the firm's prompts and patterns as reusable craft), and the AI Compliance Lead (the verification gates and the risk register).
• The roles operationalize the program's gates: the AI Compliance Lead owns the five verification gates and the risk register and confirms the cardinal rule is followed before the stamp, schedule, pay app, and safety plan; the AI Ethicist owns the bias, fairness, and workforce commitments. A gate with no owner is documentation, not practice.
• The central failure is the parallel innovation silo that the line organization ignores; it is more insidious than the no-owner failure because it looks like success on paper while it operationalizes nothing.
• The defense is the matrix, not the silo: the Director of VDC + AI sits in operations, the Chief Data Officer at the executive level with lines into projects, the Prompt Engineer-in-Residence embedded with production, and the AI Compliance Lead and AI Ethicist in risk or quality for independence and real stop authority; the test is whether the role can change what a project team does on Tuesday.
• The roles own the discipline applied to the work, not the work itself: they own the standards, gates, data, and patterns, while the line organization keeps accountability for the deliverables with the licensed professional, the PM, and the estimator.
• Compensation makes the role real: anchor each band to the most comparable existing role (senior VDC director, chief officer, senior risk or compliance manager, senior policy or legal specialist, senior technical specialist), state the adjustment and its reason, and present bands as ranges rather than invented point figures so they survive a CFO's question and a peer's comparison.
• At a smaller firm several roles are responsibilities before they are headcount, and the deliverable is the org chart plus five job descriptions that state what each role owns, where it sits, what authority it carries, and its anchored compensation band, written so the line organization sees exactly how each role changes the work.