Intermediate · 1 day
Build an AI Agent in a Day
A hands-on workshop for developers: build a working agent with tool calling, state, structured outputs, error handling, permissions and a first evaluation.
1 day · Intermediate · Workshops · Developers
Overview
This is a build day, not a survey of agent frameworks. You start with a task that actually needs tools, then implement function calling, state, planning, structured outputs, error handling and permissions. The day ends with a functioning agent and a short evaluation, not with a presentation about autonomy.
The course keeps the agent small enough to finish and strict enough to be worth finishing: validated tool arguments, a stop condition, and a permission boundary. Planning is introduced where it helps the task, not as a default.
Developers who need the deeper two-day agents course still benefit from this workshop as a first implementation. Teams that only need one internal agent often stop here and take the code into their own repository.
The output is a practical agent you ran, with traces and a list of the safeguards still required for production.
Audience
Developers who need to implement a first agent and will learn faster from a complete path than from a lecture.
Prerequisites
Professional software development and prior experience calling an LLM API. Teams new to LLMs should take the engineering or fundamentals course first.
Duration
1 day
Can be scoped to a particular internal tool set. The two-day agents course is the follow-on for production depth.
Learning outcomes
- 01
Implement tool calling with schemas and argument validation.
- 02
Add state and a bounded planning or looping step.
- 03
Handle tool errors, retries and invalid model output.
- 04
Apply a permission boundary so the agent cannot call arbitrary tools.
- 05
Evaluate the agent against a small task set and inspect traces.
Outline
- 01
Tools and structured outputs
- The task for the day and the tools it requires.
- Function calling, schemas and structured outputs.
- Build: a model that can call two or more tools correctly.
- 02
State, planning and errors
- State, memory of intermediate results, and a stop condition.
- Planning where the task needs more than one step.
- Error handling and retries that do not loop forever.
- Build: a bounded loop that can complete the task.
- 03
Permissions, evaluation and a working agent
- Permissions and denying out-of-scope tools.
- Traces and a small evaluation set.
- What would still be required in production.
- Build: finish the agent and run it end to end.
Practical work
You build one agent through the day. Typical tasks are internal operations, research with tools, or a constrained workflow. The agent runs, has traces, and is evaluated on a handful of tasks including a failure case.
Takeaways
- A functioning practical agent
- Tool schemas and a permission map
- Traces and a small evaluation set
- A list of production safeguards still to add
Delivery
Cognitecta delivers private corporate training, on-site or as remote live training. Courses can be run as published, or adapted to your organisation’s stack, domain and experience level.
Instructor
Nicholas Johnson, AI architect and software engineer. He has a degree in Artificial Intelligence and around twenty years of professional technology training, including hundreds of courses for engineering teams and large organisations. About.
Related courses
1 day · Intermediate
RAG in a Day
A focused workshop for developers: build a complete retrieval-augmented generation system in one day, from ingestion and chunking through citations and a first evaluation.
1 day · Intermediate
From Chatbot to Agent
A one-day workshop that starts with a basic LLM chat application and evolves it, step by step, into a useful agentic system with retrieval, tools, state and production safeguards.
1 day · Leadership
AI Architecture Workshop
A facilitated workshop for technical teams, architects and engineering leaders around a real customer problem. The output is an architecture and a next-step roadmap, not a training certificate.