Observed Signal · Mar 10, 2026 · Opinion · Source: a16z · Impact: 3/5 · Sentiment: Positive
Unlocking Data Agents: The Power of Context Layers
This a16z opinion piece argues that AI data agents cannot operate autonomously without a maintained context layer that captures business definitions, data-source provenance, tribal knowledge, and governance. It traces the evolution from the modern data stack and the 2024–25 agent frenzy to the common failure modes—brittle workflows and missing context—and explains why semantic layers alone are insufficient. The article outlines a five-step approach for building a modern context layer: ensure access to the right data, automate initial context construction (using LLMs), apply human refinement, connect agents via APIs or MCP, and implement self-updating context flows. It highlights existing players and paths forward (databricks/snowflake AI analyst products, Palantir ontologies, OpenAI internal work) and maps market categories including data-gravity platforms, AI data-analyst vendors, and new dedicated context-layer companies.
Analysis highlights an operational gap (context layers) that affects how enterprises deploy LLM-based data agents; relevant to data infrastructure and MarTech vendors but not a major platform policy or product launch.
Track Snowflake Signals & Market Shifts in Real-Time
Polaris7 autonomous intelligence agents track regulatory filings, primary sources, executive changes, and deal flow 24/7. Create your free Explorer workspace to monitor these entities.
Key Takeaways & Evidence Grounding
- a16z states that AI/data agents frequently fail due to lack of up-to-date business context and messy, disparate enterprise data.
- The article distinguishes traditional semantic layers from a broader 'context layer' that should include canonical entities, identity resolution, tribal knowledge and governance.
- Recommended steps for a modern context layer: access the right data, automated context construction, human refinement, agent connection (API or MCP), and self-updating context flows.
- Databricks (Genie) and Snowflake (Cortex Analyst) are cited as data-gravity platforms building AI data-analyst products on top of warehouses.
- OpenAI published details of its internal data agent; Palantir is noted for building ontologies to provide context from messy data.
Connected Companies & Entities
7 Entities mappedOntology Mapping & Concepts
Related Market Signals & Shifts
Recent verified developments and strategic activity across this market segment.
AI Agents Need a Semantic Layer
The article argues that giving AI agents direct text-to-SQL access to data warehouses exposes a lack of shared business semantics, causing inconsistent answers, missing access controls, and no auditability. A semantic layer (an intermediary metadata/metrics layer) solves these problems by exposing governed metric definitions, enforcing row-level security and multi-tenancy, providing an API (MCP/tool APIs) rather than raw DB connections, and enabling metrics-as-code workflows (YAML + Git). The piece distinguishes 'agentic semantic layers'—designed for programmatic agent consumption—from traditional BI semantic layers, lists operational requirements (MCP support, pre-aggregation, schema-as-code, warehouse coverage), and names existing vendors/tools. It promotes discovery and query APIs and recommends versioned, governed metric definitions to ensure consistent, auditable results across dashboards, agents, and APIs.
Making Audience Data Usable for AI Agents
The article argues that agentic AI is reshaping advertising beyond isolated tool improvements — moving systems toward autonomous research, planning and execution. It warns that AI agents need structured, semantically enriched context rather than raw, large datasets, and that quality and context will matter more than sheer data volume. The piece introduces Model-Context-Protocol servers (MCP) as a new standardized access layer that can connect business logic to raw signals and act as a commercial interface between partners. Frameworks such as AdCP and AAMP are named as attempts to structure agent-to-platform interactions. For agencies and AdTech vendors the shift implies prioritising semantic layers, selective high-quality integrations, and new measurement of agent usage and outcomes. The transition is presented as gradual but already changing how advertising is planned, governed and monetised.
Context Is a Platform Capability
The essay argues that organizational context for AI agents should be treated as a platform capability rather than a per-session developer burden. Developers currently spend time gathering scattered, sometimes-stale information (runbooks, standards, ownership) for each agent interaction. The author proposes platform teams provide a trusted context layer with six properties — canonical, versioned, fresh, attributable, accessible, and safe — exposed via an interface, owned by named stewards, and enforced by platform tooling. Practical first steps include identifying the top questions agents ask, naming owners, making canonical sources queryable by agents, and letting the platform generate shared agent instruction content to keep context fresh and enforced.
Track Real-Time Market Signals & Shifts
Set up custom watchlists to receive automated, evidence-grounded executive digests whenever material signals or shifts occur across your tracked landscape.