— Free workshop for engineering leaders

Your team is shipping AI agents without half the guardrails.

In 90 minutes I will show you what breaks in production and how to fix it. Real code. Real failures. Real cost.

01 — The failures

Seven things that go wrong without an agent harness

No sandbox

Agent runs actions against your live customer database.

No guardrails

Agent refunds $10,000 instead of $10.

No cost controls

Ticket spike drains your AI budget overnight.

No memory

Agent asks the same customer to re-explain their issue every time.

No observability

You have no idea why it issued 47 refunds at 3am.

No context engineering

Agent does not know your refund policy or escalation rules.

No task orchestration

Agent tries to do everything in one step and gets confused.

02 — Reference architecture

What a production agent harness looks like

flowchart TB user[Customer ticket] --> guard1[Input guardrails] guard1 --> agent[Agent loop] agent --> ctx[Context engineering
policies + RAG] agent --> mem[Memory store] agent --> tools[Tool access
via MCP] tools --> sandbox[Per-ticket sandbox] sandbox --> db[(Customer DB)] agent --> guard2[Output guardrails] guard2 --> reply[Customer reply] agent -.cost.-> proxy[LiteLLM proxy] agent -.trace.-> obs[Langfuse]

Eight components. Each one prevents a specific failure mode.

See the full architecture reference →

03 — What you get

After 90 minutes you walk away with

A working agent repo you can clone and run locally
The recording to share with your team
An architecture you can adapt to production

Read the ramp-up guide → · View the slides →

04 — Format

Free. 90 minutes. Zoom. Engineering leaders only.

I run this for engineering leaders shipping AI agents. No sales pitch, no fluff. Working code, real failures, production patterns.

DM me on LinkedIn for the next session →