Tutorials, case studies and field notes from the team building the evaluation and governance layer for the agent economy.
An MCP server is the plug a model uses to reach a real system. Here's what that actually means — and what an enterprise-grade version of it has to do that a vanilla one doesn't.
Two analysts, two definitions of Revenue, both right, neither agreed. Here's an agent that queries the data directly and shows you the SQL.
Dashboards sit on a server, perfectly correct, and nobody opens them. The fix isn't a better dashboard — it's an agent that emails you what matters.
Two refund agents, same prompt. The one with 16 prompt rules buried the duplicate-detection signal under noise. The other caught it cleanly.
I built a chaser. Then I tried to make it mean. It refused — with bullet points on why ultimatums backfire. The whole 30-minute build, end to end.
Two stories I couldn't stop thinking about last week. The agent decided. The thing it had access to executed. Nothing in the middle paused to ask 'are you sure?'.
I built a weekend agent in ContextGate that watches a Gmail label, reads PDF receipts, extracts vendor/date/amount/category, and writes clean rows to a Google Sheet — flagging anything it isn't sure about. Here's how it works, and why governance lives outside the agent.
This guide walks through building a governed AI agent in ContextGate that reads your source documents, populates a long-form template, and emails a summary flagging every place the sources disagree, so you catch contradictions early.
Clinical trial submissions are death by a thousand Google Docs. Here's how a 5-agent ContextGate pipeline turns chaos into a compliant, auditable submission package.
How Enterprise AI Governance Differs from Managed Agent Infrastructure 09 April 2026 Executive Summary On April 8, 2026, Anthropic launched Claude Managed
Bank reconciliation sounds simple until you're staring at a CSV and an invoice list, matching rows by hand. Here's how to build a governed AI agent that reconciles Xero invoices against bank transactions — with PII redaction, human approval, and a full audit trail.
Sales teams sit on some of the most commercially sensitive data in any company — revenue figures, deal values, commissions. Here's how to give them a powerful AI agent with CRM and email access, while automatically blocking it from ever seeing the numbers that matter.