Enterprise agent platforms are shipping cost attribution tooling, stateless protocol upgrades, and production-grade observability this week. Here's what matters for teams running AI in CI/CD and business workflows.
Announcing the Databricks Big Book of AgentOps
Databricks published a comprehensive playbook covering agent architectures, deployment patterns, and stakeholder management with real enterprise outcomes. FactSet improved accuracy 44%, DXC cut costs 30%, and Block saw $10M in productivity gains. The guide focuses on governance primitives like Unity Catalog and MLflow for teams moving agents from prototype to production at scale.
Expanding Genie Agents: Deep analysis, file reasoning, and more
Genie Agents now support multi-step analysis via API and can reason over unstructured data (PDFs, images) stored in Unity Catalog volumes alongside structured sources. The Genie Code tooling helps teams build and diagnose agents faster, while conversation sharing lets agent managers distribute insights across orgs. This bridges the gap between data analysts and developers who need programmatic access to agentic workflows.
How we eliminated $1 million a year of wasted AI agent spend in one hour
Databricks traced $1.2M in annual waste to seven small bugs in MCP tool servers that caused silent retry loops. Agents passed semantically valid JSON (arrays instead of comma-separated strings) that tools rejected with cryptic errors, burning tokens and engineering time. The fix: design MCP tools to handle how LLMs actually call them, not just how you expect them to be called.
// Before: rigid tool signature causes retries
function searchDocs(tags: string) { // expects "python,ai,docs"
if (Array.isArray(tags)) throw new Error("Invalid format");
}
// After: flexible handling eliminates retry waste
function searchDocs(tags: string | string[]) {
const tagArray = Array.isArray(tags) ? tags : tags.split(',');
// process normally
}Cloud Copilot Code Review Reaches Azure Repos, Billed Per Review with Reporting
Microsoft brought GitHub Copilot code review to Azure Repos with per-review billing through Azure subscriptions and project-level tags for cost attribution. Costs surface in Azure Cost Management 48 hours after review completion, giving enterprises the observability they need without migrating repos. This acknowledges that compliance and tooling constraints keep many teams on Azure Repos despite GitHub's AI-first features.
MCP in LangChain: Stateless Protocol, Elicitation, and More!
LangChain rebuilt MCP support around the July 2026 stateless spec, moving it to langchain.mcp and enabling client-side caching of tool catalogs. The update adds elicitation via LangGraph interrupts for human-in-the-loop workflows and automatic protocol negotiation for backward compatibility. Stateless core eliminates session management overhead, making production deployments more reliable at scale.
The pattern is clear: enterprise agent platforms are converging on cost observability, flexible tool design, and stateless protocols. Teams running agents in production need tracing infrastructure that surfaces retry waste and governance layers that enforce budget constraints before runaway spend happens.