EDB, Adobe, Box Move AI Agents Into Databases and Content Repos, Not Chat Windows

Agents Move to the Data Layer

The clearest shift in enterprise AI this week is architectural: platforms are building agents into databases, content repositories, and CX systems rather than offering standalone chat interfaces. EDB acquired Altinity to combine PostgreSQL and ClickHouse for agentic workloads, targeting Q4 2026 for integration. Adobe embedded its CX Enterprise Coworker across ChatGPT, Claude, Gemini, Copilot, and Databricks Genie. Box added Nvidia Nemotron models and a new retrieval layer for agent workflows. All three developments signal that vendors expect agents to live where enterprise data already sits, not in separate applications.

The problem for buyers is that none of these announcements include the figures needed to build a business case. EDB disclosed no purchase price, customer count, or migration benchmark. Adobe provided no pricing, deployment count, or productivity metric. Box released no latency result, retrieval-accuracy benchmark, or measured productivity gain. The strategic direction is clear; the commercial justification is not.

EDB-Altinity Targets Real-Time Agent Workloads

EnterpriseDB's acquisition of Altinity on November 5 positions the combined company against Snowflake, Databricks, Oracle, Microsoft, and MongoDB in the race to become the data and governance layer for AI agents. The integration will combine transactional PostgreSQL data with ClickHouse's real-time analytical engine, specifically for agentic workloads. The target is Q4 2026.

The buyer calculation is whether consolidating operational and analytical infrastructure reduces total cost enough to justify migration risk. PostgreSQL customers already committed to EDB should treat this as a roadmap signal. Buyers comparing database platforms for new AI projects face a choice: bet on a 2026 integration target with no current pricing or performance data, or select a competitor with production deployments today. The absence of migration benchmarks, integration details, and measured customer outcomes makes near-term ROI unproven.

Adobe's CX Agent Goes Cross-Platform

Adobe announced that its CX Enterprise Coworker is now available inside Anthropic Claude, Google Gemini for business, Microsoft Copilot, OpenAI ChatGPT, and Databricks Genie. AWS, IBM, and Snowflake integrations are planned. The architectural bet is that enterprises will standardize on one CX workflow system while allowing employees to use different AI assistants.

This approach competes directly with Salesforce Agentforce, Microsoft Dynamics 365 Copilot, ServiceNow AI Agents, and HubSpot, all of which try to lock customers into a single AI front end. Adobe is offering portability instead of lock-in. The risk for buyers is complexity: procurement teams must verify identity controls, audit logging, data residency, permission inheritance, and whether third-party access creates additional consumption or connector fees. Adobe did not disclose whether using the CX Coworker inside ChatGPT or Claude triggers separate usage charges from OpenAI or Anthropic, nor whether Adobe bills per user, per workflow, or per API call.

Box Adds Nvidia Models and a Retrieval Layer

At BoxWorks 2026, Box introduced Box Content Retrieval, a retrieval engine designed to structure enterprise content for AI agents without forcing a separate vector database or data-migration project. Box also made Nvidia's open-weight Nemotron models available through Box AI. The products are immediately available to Enterprise Advanced and Shield customers. Box Architect enters beta later in 2026, while Box Content Retrieval and advanced threat detection reach general availability in early 2027.

Box competes with Microsoft SharePoint/Copilot, Google Workspace/Gemini, Dropbox, Egnyte, OpenText, and specialized retrieval platforms built on Databricks or Snowflake. The purchasing question is whether Box can deliver a governed context layer at a lower total cost than native Microsoft or Google offerings. Existing Box Enterprise Advanced and Shield customers have the clearest adoption path. Other buyers should compare total licensing and model-consumption costs, especially if they already pay for Microsoft 365 or Google Workspace licenses that include AI features.

What Buyers Should Demand

The shift from chat assistants to agents embedded in systems of record is real. The commercial evidence is not. Every announcement this week lacked at least three of the following: total cost of ownership, automation rates, error rates, deployment counts, and independently measured productivity gains.

For near-term evaluations, prioritize products with existing identity, permission, and audit integrations in your current infrastructure. If a vendor cannot show you a customer deployment with measured results, assume you are buying a roadmap, not a product. Ask for per-user pricing, per-workflow pricing, and model-consumption pricing separately. Ask whether using an embedded agent inside a third-party AI platform triggers additional fees from that platform. Ask for error rates, not just accuracy claims.

The architecture is moving in the right direction. The business case is still missing.