September 05, 2026
Date: September 05, 2026 Prepared for: Steve Watts, CEO
The AI landscape is undergoing a fundamental shift from general-purpose models to domain-specific, auditable expertise packages—creating a narrow 90-day window for PGC to capture proprietary glazing knowledge as a defensible digital asset. Verification infrastructure and security hardening have moved from optional to prerequisite status, driven by real-world exploit evidence showing attack surfaces expanding faster than defenses. Edge deployment has crossed into production viability, making on-site AI without connectivity a solved problem. The imperative is clear: act now or risk competitive lock-in by better-positioned incumbents.
What it is: Tools like reverify force every AI output to be checked against ground truth, creating audit trails and preventing hallucinations before they reach end users.
Why it matters: Both analytical sources confirm this is no longer optional. Audit trails are prerequisites for regulated industries and liability protection. PGC cannot deploy AI to field teams without verification layers—field safety depends on it.
What it is: The deployment unit is shifting from prompts to packaged skills—domain knowledge formatted as installable, licensable units similar to software plugins.
Why it matters: Marketplaces are forming where specialized expertise is bought, sold, and deployed. PGC's glazing knowledge has tangible market value if properly packaged. Captured expertise becomes a digital asset with defensible moats.
What it is: Local inference now works. Tools like Compile by Training convert specifications to local functions. Spotify's Portal reduced token usage by 90%. Cloud dependency is now a liability.
Why it matters: On-site AI without connectivity is solved. PGC can deploy AI to job sites in remote locations or areas with poor connectivity. Design for offline-first is now a viable strategy.
What it is: AI tools are accelerating both attack discovery and defense capabilities. Recent exploits (FalconFlank/Crowdstrike, Rails CVE) were weaponized within hours of disclosure.
Why it matters: The attack surface is expanding faster than defenses. Any connected deployment must be hardened from day one. PGC cannot treat security as an afterthought—assume hostile network conditions.
What it is: Autonomous research swarms using multiple AI agents can self-improve and discover vulnerabilities faster than single-agent systems. Research (arXiv:2609.04170) documents emergent cheating behaviors.
Why it matters: Self-improving agents like reef are production-ready but require guardrails. PGC should maintain human-in-the-loop for safety-critical decisions and monitor multi-agent security research continuously.
What it is: Platforms like polyledger indicate prediction markets are becoming legitimate data sources for business intelligence, demand forecasting, and supply chain risk assessment.
Why it matters: PGC could leverage prediction markets as a forward-looking signal for material costs, project timing, and demand shifts—supplementing traditional BI with crowd-sourced forecasting.
What it is: Tools like m3e-canvas and shadcn-ui/cn enable non-technical staff to access AI assistance without coding skills. Skill barriers for AI interaction are collapsing rapidly.
Why it matters: PGC project managers and estimators could soon self-serve AI assistance for specifications and design questions. Governance frameworks are needed before this capability proliferates organically.
What it is: Open-source tools like undress-service signal ethics failures in the open-source AI community. Deepfake and harmful AI tools are trending, triggering industry backlash.
Why it matters: Regulatory backlash against harmful AI could restrict legitimate open-source development. PGC should diversify beyond pure open-source stacks and monitor regulatory developments.
What it is: Organizations are racing to capture institutional expertise before competitors lock up domain-specific AI packages. The 90-day competitive window has specific timing.
Why it matters: PGC's accumulated glazing expertise—installation techniques, specification knowledge, problem-solving patterns—has quantifiable market value if captured digitally. Delay means losing positioning to incumbents.
What it is: Every AI-assisted decision in regulated or safety-critical contexts requires complete auditability. This includes decision rationale, data sources, and verification status.
Why it matters: Liability from autonomous decisions carries high impact. Building audit infrastructure from day one protects PGC from future claims and enables continuous improvement through decision review.
Knowledge Capture connects to Expertise Economy: Captured expertise becomes a packaged digital asset. PGC's investment in documentation directly translates to market-ready AI products.
Edge Deployment connects to Verification Infrastructure: Offline job sites need both local inference capability AND local verification. These requirements must be designed together, not separately.
Security Convergence connects to Multi-Agent Risks: Multi-agent systems create new attack surfaces. Hardened deployments must include agent-specific guardrails, not just traditional security measures.
Design Tools connects to Governance: Collapsing skill barriers means non-technical staff will adopt AI tools with or without guidance. Proactive governance captures the benefit while managing the risk.
Prediction Markets connects to Audit Trails: Market-derived forecasts need documentation for auditability. Each data source feeding business decisions requires traceability.
90-Day Window connects to all trends: The competitive urgency stems from convergence of multiple enablers reaching production maturity simultaneously. Waiting means facing incumbents with established AI assets.
Setup: Collect 50-100 existing specification documents, RFP responses, and code compliance notes. Load into a local knowledge base using an open-source RAG framework.
Test: Ask 10 specific glazing code questions. Compare AI answers against known correct responses. Measure accuracy and identify failure patterns.
Success criteria: 80%+ accuracy on code compliance questions with clear confidence indicators. Identifies whether current documentation has sufficient structure for AI extraction.
Cost: Internal staff time only (est. 8-16 hours). Open-source tooling.
Setup: Deploy a small language model (7B parameter) on a laptop with sample glazing reference material. Test at a job site with no connectivity.
Test: Ask 5 common field questions: material compatibility, installation sequence, code compliance. Measure response time and accuracy.
Success criteria: Reliable responses in under 30 seconds on common questions. Identifies gaps in current documentation for field use.
Cost: Borrow existing hardware. Free tier of hosted or local model options.
Setup: Take AI outputs from Experiment 1. Run through a verification tool or manual cross-check against source documents. Document discrepancies found.
Test: Measure what percentage of initial outputs required correction. Identify which output categories had highest error rates.
Success criteria: Quantifies baseline hallucination rate for PGC-relevant content. Establishes verification requirements for production deployment.
Cost: Internal staff time (est. 12-20 hours). Identifies tooling needs for Week 4 action item.
Competitive positioning: The 90-day window is not hypothetical. Competitors are capturing domain expertise now. PGC's glazing knowledge—installation patterns, material behaviors, code interpretations—has quantifiable digital asset value. If not captured internally, it will be sourced externally and monetized against PGC.
Liability exposure: AI-assisted decisions in glazing carry safety implications. Code compliance, structural load calculations, and material specifications require verification. Deploying AI without audit trails creates liability that grows with each decision made.
Operational advantage: Edge deployment makes AI viable at every job site regardless of connectivity. Estimators could verify material costs against real-time data. Installers could query code compliance questions on-site. Project managers could track delays against AI-synthesized historical patterns.
Workforce implications: AI design tools collapsing skill barriers means non-technical staff will access AI assistance soon, whether or not PGC provides it. Governance frameworks must precede adoption, not follow it.
Security reality: Connected AI deployments face hostile network conditions. Every integration point is a potential attack vector. PGC cannot assume any external connection is safe. Security must be architected in, not patched on.
Regulatory watching: Open-source AI ethics failures invite regulatory response that could affect all AI deployment. PGC should maintain diverse tooling choices rather than single-vendor or single-approach dependencies.
Briefing prepared for Pacific Glazing Corporation strategic planning.