The agent workshop.
One day, your team, your stack. Training on building agents that survive production — evals, operations, deployment discipline — taught by people who run their own.
Built around your workflows, not a demo repository.
This is a private session for one team — engineers, the operators who own the workflows, and whoever will answer for the system in production. Before the day, we ask for a short questionnaire and one workflow you actually want automated; everything taught gets applied to it.
The material is vendor-neutral. Model providers give away tool tutorials, and the tools change quarterly — what holds is the practice: how you evaluate, how you operate, how you decide what not to build. That practice is the curriculum.
The working example is ours. Agent Floburn is the marketing-operations agent we built and run for our own products — it reads and writes across campaign planning, a CRM, and nightly content discovery. You’ll see how it’s evaluated, monitored, and operated, including the parts that went wrong.
Five blocks, one working day.
- 01
Where agents fit — and where they don't
The honest triage before any architecture: which of your workflows wants an agent, which wants a script, and which wants nothing. When a workflow doesn't need an agent, we say so in the room.
- 02
Evals before agents
The discipline that separates demos from systems: writing the evaluation before the agent, building the test set from your real cases, and knowing what “working” means in a number before anyone ships.
- 03
Agent operations
What running agents actually costs: monitoring that catches drift, handling model deprecations and upgrades without a rewrite, failure modes that only appear in production, and who gets paged.
- 04
Integration reality
Agents are only as good as the systems they touch. Authentication patterns, data quality, rate limits, and the gap between what the API docs say and what the database contains.
- 05
A working session on your workflow
The afternoon is applied: we take one of your actual workflows and work it end to end — triage, eval design, architecture sketch, operating plan. Your team leaves with the method applied to something real.
Not a tool tutorial. Not certification theater.
We don’t teach clicks in a vendor console, and nobody leaves with a badge. We’re not selling a platform, ours or anyone else’s — if the right answer for a workflow is a hundred-line script and a cron job, that’s the answer you’ll hear.
It’s also not a sales call in disguise. Some teams leave the workshop wanting a build; most leave with enough method to do the next piece themselves. Both outcomes are fine with us.
One day. Fixed fee. In writing before we start.
Up to 12 participants, on-site in California or remote. The fee is fixed and quoted before anything is scheduled; it covers preparation against your questionnaire and workflow, the full day, and the materials — which your team keeps. Multi-day formats are scoped on request.
If the working session surfaces a build worth doing, the next unit is the two-week build diagnostic — fixed scope, fixed fee, and a deliverable you can take to any vendor.