Advanced: Agent Dev 16 Takes Your Agent From Notebook to Production

Running something in a notebook is only the start. Going to production means adding everything beyond the model: hosting, identity, state, retries, cost, evaluation, and approvals.

Remember These Points

• The model accounts for roughly 20% of a production-grade agent; the other 80% is operational scaffolding.
• Three deployment options: running inside your own process, a managed service, and using a workflow to orchestrate multiple agents.
• Lifecycle: create, version, deploy, evaluate, observe, retire.
• Scaling considerations: model routing, caching, concurrency, stateless design.
• Cost is controlled by choosing the right model, routing, and evaluation gates.

Going Deeper

Prototype vs. production.

Concern	Prototype	Production
Running	In a notebook	Managed service, with versions and releases
Identity	Your own login	A managed identity with restricted permissions
State	In memory, gone on restart	Stored externally
Failures	You read the error	Retries, fallbacks, alerts
Cost	'A few cents'	Per-request tracking, caching, budgets
Quality	Eyeball it	Automated evaluation before every release
Trust	You approve every step	Policies plus human approval for risky actions

Three deployment options. Inside your own process: full control, but you manage scaling and stability yourself. Managed service: the agent is registered to a platform as a resource, and the platform handles sessions, content safety, and permissions. Workflow: orchestrate multiple agents into a graph with branching, approvals, and resumable runs.

Ways to save money. Use small models for simple tasks and large models only for complex ones; cache repeated content; use evaluation gates before release to block versions that get worse.

Image from the course repository.

Try It Yourself

Using the table above, check off an agent you are working on: which rows still sit in the 'Prototype' column?

Let Your Agent Teach You

Give the text below to your agent (Claude Code, ChatGPT, and Gemini all work), and it will guide you step by step like a mentor.

References

• Content adapted from Lesson 16 (16-deploying-scalable-agents) of microsoft/ai-agents-for-beginners, originally under the MIT license.
• For the original and full code: 16-deploying-scalable-agents
• All courses are listed in the Agent course roadmap