Field Notes · Production AI Engineering

The Agentic Studio Labs Log — field notes from production AI engineering.

Technical write-ups from real client engagements: the architecture decisions, security tradeoffs, and integration details behind agentic systems, MCP connectors, and voice AI that made it into production. Written for engineering leaders deciding what to build, buy, or harden next.

Building a Renewal Risk Command Center with Guild.ai

A Guild.ai TypeScript agent demo that turns renewal-risk spreadsheet data into a Slack approval workflow, using Guild.ai service tools, OAuth credentials, triggers, validation, and publishing to coordinate the handoff.

Read the field note →

Building a Claude MCP Connector for SharePoint and DocuSign

Field notes from an Azure-hosted Claude MCP connector over Microsoft 365 documents: SharePoint, Microsoft Graph, least-privilege access, Key Vault, and DocuSign draft-envelope creation.

Read the field note →

Shipping an OpenClaw Market Research Agent into Production

A production research agent is a pipeline, not a prompt: scheduled runs, citation tracking, third-party APIs, PDF rendering, SharePoint delivery, and Teams integration.

Read the field note →

Setting Up Claude Cowork for Client-Sensitive Work

Claude Cowork is a local desktop AI agent that does production work — which is exactly why setup discipline matters for firms handling sensitive client data. A practical, security-first walkthrough: the right plan, model choice, projects and skills, file-access rules, and the settings to lock down.

Read the field note →

Deploying AI in Your Investment Firm

The firms that will survive the AI transition aren't the ones deploying fastest — they're the ones deploying with an architecture that respects the specific risk profile of an investment operation. Covering trust boundaries, data governance, and what good AI deployment looks like.

Read the field note →

OpenClaw in Production: A Security Guide from a Real Engagement

Lessons from a real client engagement on securing OpenClaw for production — covering the current threat landscape, nine active CVEs, the ClawHavoc supply chain attack, and the ten questions every deployment team should answer before going live.

Read the field note →

Writing Great Anam.AI Persona Prompts

A practical guide to writing system prompts for Anam.AI video personas — covering the five building blocks, the counterintuitive trick that makes them feel real, knowledge base setup, and common mistakes to avoid.

Read the field note →

Building a Voice AI Intake Agent

We compared three voice AI platforms — Vapi, Bland.ai, and Retell AI — to build an inbound intake agent that qualifies leads before the first human call. Here's the platform comparison, conversation flow, and implementation roadmap.

Read the field note →

OpenClaw Security Tooling Landscape

A comprehensive reference cataloging every technical control in the OpenClaw ecosystem — agent firewalls, sandbox isolation, MCP filtering, prompt injection defense, DLP, observability, network segmentation, and compliance tooling across 12 tools.

Read the field note →