LAF for AI-augmented SDLC

This section addresses the architectural implications of AI participation in the SDLC. It does not define organizational governance for AI models, AI services,  or other technologies; these remain outside the scope of LAF.

LAF assumes that traditional transformation bottlenecks stem from human decision-making velocity, focusing on optimizing interactions between architects, business units, and developers. When AI agents execute substantial portions of the SDLC, the primary bottleneck shifts: human-readable artifacts (RA, ST, CP) are ignored by automated coding tools unless transformed into machine-consumable context. The following additions extend LAF to maintain effective governance in AI-assisted SDLC environments.

5.1 Machine-Readable Architecture Context (MRAC)

Alongside human-readable RA, ST, and CP artifacts, organizations using AI-assisted SDLCs should maintain structured, machine-readable counterparts (e.g., structured JSON/YAML schemas) to inject into AI developer agents as system context. MRAC includes condensed Reference Architecture mappings, machine-validatable Standards rules, and key Component Passport data. Human-readable documents remain the primary source of truth, while MRAC serves as a synchronized derivative.

5.2 AI-Assisted Solution Vision

Using MRAC and business requirements, AI agents can generate initial drafts of Solution Visions. The architect’s role evolves from primary author to reviewer and approver. Within Fast-Track pipelines, AI-generated changes confined to single components that do not change its capabilities can be automatically validated against Standards and Reference Architecture, with RIM attribute  computed programmatically. Non-compliant or ambiguous cases should be escalated to the Architectural Forum.