A private note to leadership
A read on where AI actually fits in Bilzin Sumberg's stack, and where it does not.
Written for Bilzin Sumberg and the partners who signed off on the January 2026 reshuffle. One page, no deck, no vendor gloss.
The January 2026 reshuffle of the Land Use, Land Development & Government Relations, and Real Estate Capital Markets groups reads like a firm responding to the pace of Miami's density and transit build-out. Bilzin Sumberg's own Tech Talk work on data-center permitting and AI in homebuilding makes the same point from the client-advisory side. The document and regulatory surface is moving faster than the hour-by-hour research that produces entitlement packages, compliance memos, and CMBS closing binders. Every law firm in South Florida is looking at the same shift. The question is which tools belong where inside the practice.
The honest frame
Most of what a practice group calls "AI" is three different problems wearing one label.
-
60%
Traditional database and retrieval.
How does an associate find the right precedent, the right ordinance, the right prior entitlement package, fast. Properly indexed, this is not an AI problem.
-
30%
Rule-based logic.
The permit checklist, the closing checklist, the diligence matrix, the FinCEN residential real-estate compliance sweep. Deterministic, auditable, no model required.
-
10%
Genuine language-model work.
Drafting assistance, summarization across thousands of pages, cross-document reconciliation. This is where a model earns its keep.
Firms that skip this distinction end up putting a large-language model where a properly indexed precedent library would do the job faster, cheaper, and with a better audit trail.
What the first engagement looks like
A four to six week read of one practice group. Land Use, or Construction, or CMBS. Pick the group whose Fridays hurt the most. The deliverables:
- A layer map of that group's weekly document work, sorted onto the 60/30/10 view with the specific retrieval, rule, and model tasks named.
- A first version of the context architecture that makes precedent retrieval work inside the firm's existing DMS and M365 tenancy. No vendor cloud on day one.
- A training pass with the associates and the knowledge team so the architecture is used, not shelved.
- A published internal methodology in the firm's voice, so the work is portable to the next practice group.
A relevant case study
VigilOre. A multi-agent compliance platform for a mining operation in the Democratic Republic of the Congo. Document-heavy, regulated, permit-driven.
The compliance workflow compressed from over 160 hours to under 5 minutes by separating retrieval, rule checking, and drafting onto the layers where each actually belongs.
The industry is different from a Miami land-use practice. The shape of the problem is the same. Scoped, measured, sitting on top of existing systems rather than replacing them.
The methodology, published
The work is anchored in a peer-reviewed paper: Interpretable Context Methodology (ICM), submitted to ACM TiiS. The paper argues that agent context organized as a layered filesystem is more interpretable and more reproducible than prompt-soup architectures. For a firm with a Chief Knowledge Officer, the methodology is the part that matters. It gives the Bilzin Sumberg team a framework they can publish under their own name, not a vendor lock-in.
Repository: github.com/RinDig/Interpretable-Context-Methodology-ICM-
Adjacent credibility, briefly
- KPMG UK (one of the Big Four): 40+ executives trained through a scoped program in regulated-industry AI practice.
- Correlation One enterprise program: 1,500+ people trained across Pacific Life and Colgate-Palmolive since May 2025. 95% still using the tools thirty days after the workshop.
- NLP Logix partnership for work that sits below the orchestration layer. NLP Logix has been in machine learning since 2011 and runs over 150 data scientists. If a Bilzin engagement eventually needs production-grade ML pipelines or data-platform infrastructure, that partnership covers it.
A half hour with Matt Creamer
Matt is Eduba's Chief Revenue Officer and the right first conversation for firm leadership. Thirty minutes, on video. Bring one document workflow that currently eats a full associate day every week. We will map it onto the 60/30/10 view and tell you which piece belongs on which layer.