Course
Build AI Agents in Go
Build AI agents in Go that actually work: tools, guidelines, guardrails, honest evals, and any model behind one flag.
Who it's for
Backend engineers who know some Go and want to build agents themselves — not just call an SDK from another language.
Not for: People who want a no-code agent builder, or a tour of every agent framework.
What you'll build
- An agent loop written from scratch, no framework
- Tools that are safe by default and give the model useful errors
- One flag to switch between Claude, Bedrock and local models
- Evals with an answer key, so you can compare models honestly
- An agent you can ship as a CLI and run in CI
Planned curriculum
This is the plan today. Waitlist feedback will shape the final version.
- 01
What an agent really is
Model + tools + loop, and why Go is a great fit: one binary, concurrency, easy to ship.
- 02
Your first agent loop
Tool use with the Claude API, written from scratch so nothing is magic.
- 03
Designing tools
Schemas, read-only by default, and errors the model can learn from.
- 04
Guidelines
Giving your agent taste with embedded instruction files.
- 05
One flag, any model
A provider interface for Claude, Bedrock and local models via Ollama.
- 06
Context and cost
Context limits, trimming and token budgets — including a full code review for 16 cents.
- 07
Evals
Answer keys, scoring, and comparing models honestly, using real experiments from Build It in Go.
- 08
Guardrails
Permissions, secrets, and human approval for anything that writes.
- 09
Shipping it
Packaging your agent as a CLI and running it on every pull request.
Before you start
- Comfortable writing Go
- No prior AI or agent experience needed
Your instructor
I'm Marian Montagnino: a senior software engineer, author of Building Modern CLI Applications in Go, international GopherCon speaker, and a founding engineer on Netflix's cloud workstation platform. On my YouTube channel, Build It in Go, I build real things with Go and AI and show the real numbers — what worked, what broke, and what it cost.
Questions
When does it start?
When enough people join the waitlist. Everyone on the list hears first and gets the founding price.
How do I pay?
When enrollment opens, checkout runs through Gumroad. Joining the waitlist is free and commits you to nothing.
Want this for your whole team?
I also run it as a workshop for engineering teams. See workshops.
Want to go deep on the tools agents use? Agent-Ready Go CLIs →