Technical note 03 · Strategy layer
The AI strategy layer belongs above the model
Models will keep changing. Durable advantage comes from the firm context, rules, tools, controls, and feedback loops that make intelligence useful.
Practical outcome
Adopt better models without rebuilding how the firm works
9 minute read
The thesis
The foundation model is a component, not the product strategy.
General intelligence is improving quickly and will come from multiple providers. An AEC firm should benefit from that progress without encoding its project operations inside a single model or vendor interface.
The durable layer is the one the firm owns: connected project context, approved standards, reusable skills, workflow definitions, access controls, evaluations, and the record of what people accepted or corrected.
System design
Put stable firm knowledge beneath replaceable intelligence.
AEC Hub is designed as a strategy and orchestration layer across existing project systems. Source applications remain authoritative. The platform turns selected records into a shared project context, exposes bounded tools to AI, and returns outputs to a human workflow.
Systems of record
Models, documents, communication, and business systems remain the authoritative sources.
Project context
Typed objects and source relationships give every workflow a consistent working set.
Firm logic
Standards, checks, skills, and approval policies encode how the firm wants work performed.
Model layer
Choose intelligence for the task while keeping tools, evidence, and controls stable.
The firm-owned AI strategy layer
Engineering judgment
Use deterministic software, model judgment, and human review for different jobs.
A calculation, permissions check, record lookup, or file identifier should not be guessed by a language model. Those belong in deterministic code. Interpretation, comparison, drafting, and explanation can benefit from model judgment. Decisions with contractual, safety, financial, or publishing consequences stay with a person.
The strategy layer routes each part of a workflow to the right mechanism. That is more reliable and usually less expensive than sending every step through the largest available model.
Model choice
Provider choice should improve the system, not redefine it.
Today the platform uses provider-neutral routing for AI visualization and a bounded agent runtime for project assistance. The broader direction is pragmatic portability: evaluate models against real AEC tasks, use the strongest fit, and keep the firm-owned context and control plane independent.
That is not a claim that every model is interchangeable. It is a commitment to measure the whole workflow rather than treating a provider logo as the architecture.