Living Context Layers Become AI Infrastructure
Claim
AI-native companies will need living context layers that continuously capture workflows, operating artifacts, decisions, and system state. Static documentation and file piles will be insufficient for reliable AI products and internal agents.
Raised by
- 2026-03-23-a16z-two-paths-left-for-software - david-george / a16z recommends process-capture sprints and treating documentation as product infrastructure.
Supporting evidence
- 2026-03-23-a16z-two-paths-left-for-software - explicitly calls for harvesting SOPs, tickets, transcripts, requirements docs, policies, CRM notes, support logs, event data, and approval paths into a living context layer with evals for accuracy, exceptions, latency, and cost.
Counter-evidence
- (unknown - needs source) It is not yet proven that buyers will purchase a context layer directly rather than expect it to be hidden infrastructure inside a vertical product.
Implications
- context-os-brain is directionally validated as infrastructure, but not automatically validated as the product.
- The team should keep improving ingest, dedup, conflict handling, evals, and human approval because these become reusable substrate across vertical wedges.
- The commercial wedge still needs to be an output surface: AM capacity, revenue-cycle loop, construction margin protection, or another workflow.
Ideas this favors
- context-os-brain as internal infrastructure beneath a vertical product.
- human-in-the-loop-accountability as the trust layer that turns context from “summaries” into accountable operating knowledge.
- vertical-use-case-led-brain where the vertical determines the context schema.
Ideas this weakens
- A passive document repository or static wiki positioning.
- A pure RAG-on-uploads product without maintenance, evals, conflict resolution, or human accountability.
Confidence
Medium. Multiple internal Brain discussions already point this way, and the a16z article is an external investor signal. It still needs buyer evidence that this infrastructure drives budget.
What would change our mind
- Production agents become reliable enough using existing SaaS APIs and search without a maintained context layer.
- Buyers reject context-maintenance workflows as operational overhead.
- Vertical products win without building durable cross-source context.