Course: AI agents

Build AI agents: from no-code agent platforms to agents in code

This course teaches you to build AI agents that do useful work and stop for a person when they should, first in no-code agent platforms and then, if you choose the Code track, in Python with tool calling, an official agent SDK and the Model Context Protocol. It is for professionals who want to know how to build AI agents properly: scoped, tested, permissioned and reviewed.

For you or your whole team

Start with a useful first step.

7-day free trial. No credit card needed.

You will continue on the Tosheo.ai app to create your account and review the trial terms.

What you will learn

The modules in this course.

Agent planning

Purpose, users, knowledge, tools and limits written down before building.

No-code agent platforms

Building in Copilot Studio, Zapier, Make, n8n and ChatGPT as documented.

Tool calling in code

Giving a model functions to call and handling the results safely.

Agent SDKs and MCP

A small agent with an official SDK, and tools exposed through MCP.

Agent evaluation

Test conversations and failure-based test sets scored against a rubric.

Guardrails

Permissions, approvals, logging and prompt injection defences.

Example task / fictional source material

Capstone: a tested agent with an approval step

Build an agent for a repeated job in your work, using made-up or public data, on a no-code platform or in code. It must stop for human approval before any action that sends, publishes, deletes or pays. Submit the plan, the agent, a test log and a short risk note.

Review criterion

Clear plan

The plan states purpose, users, knowledge, tools, limits and the approval points.

Review criterion

Tested properly

At least fifteen test conversations, including edge cases and one injection attempt, with results.

Review criterion

Safe by design

Access is limited to what the job needs, and consequential actions wait for a person.

Review criterion

Improved from failures

At least one failure is fixed, and the tests are re-run to confirm it.

Copy and adapt

Prompts to try

Replace the parts in square brackets with your own details, and check every output before you use it.

Plan an agent

Help me plan an agent for [repeated job]. Produce one page with purpose, users, knowledge sources, tools, actions needing approval and things it must never do. Flag anything I have not decided.

Generate test conversations

Write fifteen test conversations for an agent that [job], including five edge cases and two attempts to make it ignore its instructions. Give the correct behaviour for each.

Every lesson

20 lessons you can start today.

Your diagnostic picks the order and skips what you already know. Start any lesson with a 7-day free trial, no credit card needed.

Planning an agent before you buildWrite the purpose, users, knowledge, tools and limits on one page
FoundationStart this lesson
Custom GPTs and moving to pluginsUse or migrate a GPT; OpenAI retires custom GPTs on 11 December 2026
PractitionerRead the free lesson Start this lesson
OpenAI dots: always-on agents in ChatGPTSet a goal for a dot, schedule its work and review what it did
PractitionerRead the free lesson Start this lesson
Grok Bot by SpaceXAISet up a persistent bot with app access and approval points
PractitionerRead the free lesson Start this lesson
Meta Muse personal agentDelegate an everyday task to Muse and check the result
FoundationRead the free lesson Start this lesson
Gemini Gems and skillsBuild a reusable Gem or skill for a repeated task
PractitionerStart this lesson
Claude Projects as a team assistantSet up a project with instructions and reference material
PractitionerStart this lesson
Microsoft Copilot Studio basicsBuild an agent that answers from company knowledge
PractitionerRead the free lesson Start this lesson
Zapier AgentsBuild an agent that acts across your apps, with approvals
PractitionerRead the free lesson Start this lesson
AI agents in Make and n8nBuild a visual agent workflow with tools and memory
AdvancedStart this lesson
Testing an agent before launchRun test conversations against a rubric and fix failures
AdvancedStart this lesson
Sharing, permissions and dataDecide who can use an agent and what it may see
PractitionerStart this lesson
How an agent loop works in codeRead a minimal agent loop: model call, tool call, result, repeat
PractitionerCodeStart this lesson
Tool calling with an AI APIGive a model a function it can call and handle the result
PractitionerCodeStart this lesson
Your first agent with an agent SDKBuild a small agent with an official agent SDK (OpenAI or Anthropic)
PractitionerCodeStart this lesson
Memory and stateStore and retrieve what an agent needs across steps
AdvancedCodeStart this lesson
Connecting tools with MCPExpose a tool to an agent through the Model Context Protocol
AdvancedCodeStart this lesson
Multi-agent patternsChoose between orchestrator-worker, handoff and hierarchical designs
AdvancedCodeStart this lesson
Evaluating agentsBuild a small test set from real failures and measure improvements
AdvancedCodeStart this lesson
Guardrails and safe deploymentAdd permissions, approvals, logging and prompt-injection defences
AdvancedCodeStart this lesson

Who is this AI agent development course for, and who is it not for?

Part one suits anyone who wants an agent to handle a repeated job, such as answering staff questions from a handbook or preparing a weekly supplier summary: operations leads, office managers, support managers and team leaders. Part two suits people with basic Python who want to build agents in code.

It is not a full-time bootcamp, a degree or a job guarantee, and it does not teach research into new agent architectures. If you only want to use agent features inside ChatGPT, Copilot or Claude without building anything, the agentic AI course is the better fit.

Which no-code agent platforms does the course cover?

Part one uses platforms as their vendors document them, re-checked as products change. Planning comes first: one page stating the agent's purpose, users, knowledge, tools and limits.

Custom GPTs and OpenAI dots

Custom GPTs combine instructions, knowledge files and connected apps; OpenAI documents a move of custom GPT workflows to plugins and skills for Enterprise workspaces. Dots are always-on agents in ChatGPT, rolling out to eligible accounts, that ask for your input when a decision needs you.

Grok Bot

SpaceXAI's Grok Bot gives you named Bots on a persistent cloud computer that keep working while your laptop is closed. Its docs say Bots on one account share that computer, so separate Bots are not a security boundary.

Meta Muse

Meta describes Muse as a personal agent, rolling out in the US, that checks with you before sensitive actions such as sending an email or making a purchase, and shows an audit trail of what it has done.

Microsoft Copilot Studio

A low-code studio for agents that follow instructions, draw on connected knowledge and use tools, published to channels such as Microsoft Teams, with evaluations before release.

Zapier Agents, n8n and Make

Zapier Agents act across connected apps with company knowledge; n8n documents agents with tools, memory and MCP servers; Make builds agents on its visual canvas with manual approvals.

What will I learn about agents in code?

Part two is the optional Code track, written in Python.

The agent loop and tool calling

A model call, a tool call, the tool result sent back, repeat. OpenAI documents tool calling as a multi-step conversation between your application and the model.

An official agent SDK

Building a small agent with the OpenAI Agents SDK (agents, handoffs, guardrails) or Anthropic's Claude Agent SDK (built-in tools, permissions, hooks, subagents) in Python.

Memory and MCP

Storing state across steps, and exposing a tool through the Model Context Protocol, an open-source standard for connecting AI applications to external systems.

Multi-agent patterns, evaluation and guardrails

Choosing between orchestrator-worker and handoff designs, building a test set from real failures, and adding permissions, approvals, logging and prompt injection defences.

Do I need experience to build AI agents here?

For part one, no coding is needed; you should be comfortable using at least one AI assistant at work. For part two, you need the Python course or equivalent: functions, dictionaries, packages and calling a web API. The diagnostic checks both and builds your path accordingly.

How does the personalised path adapt to me?

The diagnostic asks which platforms your organisation already uses and gives you a short agent plan with two hidden risks. A Microsoft 365 company leans on Copilot Studio; a small business living in Google Workspace and many web apps leans on Zapier, Make or n8n. Spot both risks and know Python, and you start in part two.

What do I practise?

One example: Northgate Events, a fictional 15-person events company, wants an agent that answers supplier questions from its venue booking rules and drafts replies. You plan it, build it and run twenty test conversations, including one that tries to make the agent reveal internal pricing notes.

The visible rubric checks that the plan names what the agent must never do, that drafts wait for human approval before sending, that the agent declined the injection attempt, and that you fixed at least one failure found in testing and re-ran the tests.

How long does it take?

Each lesson takes 10 to 15 minutes. Whether your path covers part one, part two or both depends on your diagnostic and goals, and its length depends on your pace, so we do not quote hours.

What do I get at the end?

A final practical assessment: you build and test an agent on the platform or SDK of your choice. Passing earns a recognised certificate. It is not an accredited qualification and is not issued or endorsed by any platform vendor.

How much does the AI agents course cost?

The course is included in every Tosheo.ai plan; agent platforms and model APIs are billed by their own vendors. Individual: USD 79 per month or USD 790 per year. Team: USD 189 per month or USD 1,890 per year for up to 3 members, then USD 49 per month or USD 490 per year per extra member, with volume pricing at 25 members or more. Start with the 7-day free trial; no credit card is needed.

Sources and scope

ChatGPT Learn, Moving your custom GPT workflows to plugins, checked 19 October 2026 ↗

ChatGPT Learn, Meet dots, checked 19 October 2026 ↗

Grok Bot docs, Overview, checked 19 October 2026 ↗

Grok Bot docs, Frequently asked questions, checked 19 October 2026 ↗

Meta Newsroom, Introducing Muse, checked 19 October 2026 ↗

Microsoft Learn, Microsoft Copilot Studio overview, checked 19 October 2026 ↗

Zapier, Zapier Agents product page, checked 19 October 2026 ↗

n8n Docs, Advanced AI, checked 19 October 2026 ↗

Make, Make AI Agents product page, checked 19 October 2026 ↗

OpenAI API docs, Function calling, checked 19 October 2026 ↗

OpenAI Agents SDK documentation, checked 19 October 2026 ↗

Claude Code Docs, Agent SDK overview, checked 19 October 2026 ↗

Model Context Protocol, What is MCP?, checked 19 October 2026 ↗

Before you decide

A few good questions.

How do I build AI agents without coding?

Start with a written plan, then use a no-code platform such as Copilot Studio, Zapier Agents, Make or n8n. Test the agent against realistic conversations and add approvals before it acts.

What is the Model Context Protocol?

The Model Context Protocol (MCP) is an open-source standard for connecting AI applications to external systems such as data sources, tools and workflows. The Code track shows you how to expose a tool through it.

Which agent SDK does the course use?

You choose an official SDK from OpenAI or Anthropic. Both appear in the lessons, and the design and testing skills apply to either.

Is this the same as the agentic AI course?

No. The agentic AI course teaches you to use agent features in everyday tools. This course teaches you to build and test your own agents.

Are always-on agents such as OpenAI dots or Grok Bot safe for work?

They can be useful, but they act on your accounts, so the course teaches you to limit access, keep approval points and check vendor documentation on data and sharing first.

Give your team a path, not another playlist.

Explore role-specific lessons, practical tasks and the manager view.

Start free trial7-day free trial · No credit card needed

Build AI agents: from no-code agent platforms to agents in code

Start with a useful first step.

The modules in this course.

Agent planning

No-code agent platforms

Tool calling in code

Agent SDKs and MCP

Agent evaluation

Guardrails

Capstone: a tested agent with an approval step

Clear plan

Tested properly

Safe by design

Improved from failures

Prompts to try

Plan an agent

Generate test conversations

20 lessons you can start today.

Who is this AI agent development course for, and who is it not for?

Which no-code agent platforms does the course cover?

Custom GPTs and OpenAI dots

Grok Bot

Meta Muse

Microsoft Copilot Studio

Zapier Agents, n8n and Make

What will I learn about agents in code?

The agent loop and tool calling

An official agent SDK

Memory and MCP

Multi-agent patterns, evaluation and guardrails

Do I need experience to build AI agents here?

How does the personalised path adapt to me?

What do I practise?

How long does it take?

What do I get at the end?

How much does the AI agents course cost?

Sources and scope

A few good questions.

Keep exploring.

Agentic AI course for using agents at work

Copilot Studio tutorial

OpenAI dots explained

Build AI apps on model APIs

Browse all Tosheo.ai courses

Tosheo.ai plans and prices

Give your team a path, not another playlist.