Field Notes on Building AI Employees
How I build AI employees that do real work - the engineering, the reliability discipline, and what I've learned running my own agency on them. Written by Christoph Sauerborn.
Notes from practice
WhatsApp customer support with AI that does more than a chatbot
Why decision-tree chatbots frustrate customers, and how an AI agent that knows your company answers on WhatsApp: sources, handoffs, approvals, GDPR.
WhatsApp CRM Integration: Send a Voice Note, Skip the Data Entry
One voice note after the meeting, and the agent files everything into your CRM, structured. Here's how WhatsApp CRM integration works, GDPR-compliant over the Business API.
Market monitoring with AI: a wiki that keeps itself current
An AI employee reads your industry sources, summarises them, and keeps a wiki up to date in Notion, Confluence, or SharePoint. With the source, the link, and clear limits.
Create LinkedIn Posts With AI From a Voice Note
Send a voice note, get a draft in your voice, approve it, done. That's how you create LinkedIn posts with AI and skip the hour in front of a blank editor.
Automate CRM updates so every meeting lands in the CRM
CRM upkeep fails on timing, because the work piles up right after the meeting. Here is how an AI employee writes the summary, notes, and tasks into your CRM.
The AI Productivity Paradox: Why Your Team Isn't Faster Yet
I keep watching teams adopt AI and ship at the same speed. Here's why the bottleneck moved, and the few norms you have to rewrite to catch up.
Stop Babysitting Your AI Agents: Build a Verification Loop
I stopped babysitting my AI agents by giving each one a verification loop and a clear definition of done. Here is how the loop works and how I build it.
AI Subagents: When One AI Employee Becomes a Team
I explain when to split one AI employee into an orchestrator plus specialist subagents, and the more common case where you should not.
What an AI Employee Actually Is (And Why It Isn't a Chatbot)
I build AI employees for a living. Here is the real difference between one that does a task and a chatbot you operate yourself.
Why AI Gives Confident, Wrong Answers About Your Data
AI gets your data wrong because it knows business in general, not your definitions. What fixes it is a semantic layer that maps the model to how your business actually works.
AI Agent Evals: How to Tell If Your Agent Actually Works
I never ship an AI employee without an eval set. Here is how I turn "it feels worse lately" into a number I can act on.
AI Agent Skills: Teaching an AI Employee One Job Properly
I explain what a skill is when you build AI employees: a packaged, versioned procedure that teaches an agent to do one job your company's way, every time.
Trustworthy AI Agents: Trust the Mechanism, Not the Output
I trust an AI agent because I can read how it reached an answer, not because the answer looks good. Here's the engineering behind that.