Shopify
Stripe
Slack
Notion
Google Workspace
Confluence
How Do You Build an AI Agent with Macha's AI Agent Builder (Sidekick)?
Macha's AI Agent Builder is Sidekick, the assistant that opens when you click Build with AI on the Agents page. You describe the agent in plain language, and it proposes the name, instructions, tools and triggers from your connected tools, then builds the agent only after you confirm.
Key takeaways
- Macha's AI Agent Builder is Sidekick: clicking Build with AI on the Agents page opens it, and a plain-language description returns a complete agent proposal.
- Sidekick proposes the agent's instructions, specific tools such as zendesk_get_ticket and zendesk_update_ticket_tags, and a ticket.created trigger before creating anything.
- A good brief for the AI Agent Builder names three things: the job, the trigger that starts the agent, and the actions it takes with the result.
- Build with AI is one of three ways to create a Macha agent, alongside creating one manually from a blank configuration and starting from a template.
- After Sidekick builds an agent, Edit with Sidekick refines it in plain language and Test run tries it on a real ticket before customers see it.
To build an agent with Macha's AI Agent Builder, click Build with AI on the Agents page, describe the job, the trigger and the actions in plain language, and Sidekick proposes the agent's name, instructions, tools, triggers and sub-agents, then creates it only after you confirm.
"AI Agent Builder" and Sidekick are the same thing: Build with AI opens Sidekick, Macha's built-in assistant, and it is available on every plan, including the trial. Build with AI is one of three ways to create an agent:
| Method | How you start | Best for |
|---|---|---|
| Build with AI | Click Build with AI and describe the agent | First agents and standard workflows |
| Create manually | Click Create Agent and fill a blank configuration | Precise control over every field |
| Start from template | Pick a pre-built template such as ticket triage | Common workflows you want to adapt |
What does building an agent look like?
Where do you start?
On the Agents page, click Build with AI. That opens the Sidekick panel with a prompt, and you describe the agent you want. You can also open Sidekick from the sidebar footer, or with ⌘K on Mac and Ctrl K on Windows or Linux, from any page.
Here's Sidekick after a real prompt: "Build an agent that triages incoming Zendesk tickets and tags them by topic and priority." It doesn't just acknowledge the request. It proposes the whole agent: the instructions, the exact tools (zendesk_get_ticket, zendesk_update_ticket_tags, zendesk_update_ticket_priority) and a ticket.created trigger, then says "If you approve, I'll create it."
How should you describe the agent?
The quality of what you get back depends on how clearly you brief it. A good description covers three things:
- The job: what the agent should do. "Create a voice-call transcription agent that transcribes the audio attached to a ticket."
- The trigger: when it should run. "Fire it whenever a new Zendesk ticket comes in."
- The actions: what it should do with the result. "Leave a summary of the transcription as an internal note."
The builder works from your connectors, so it knows what's possible. It recommends a ranked list of tools from the connectors you have connected, with a picker you can adjust. In the example above, ElevenLabs is connected for transcription and Zendesk for the trigger and the internal note, and the builder wires all of that together from the description.
Does it build anything before you approve?
No. This is the part that makes building with Sidekick safe to use: it proposes a plan and waits for your confirmation before building anything. After you describe the agent, it shows you what it intends to create:
- the name and description it'll give the agent,
- the full instructions,
- and the tools and triggers it recommends, based on your connectors.
You review the plan at your own pace. Nothing is created until you say "yes." Confirm, and the builder assembles the agent with connectors filled in, tools wired and the trigger set.
How do you check and refine what it built?
Because the agent is AI-generated, treat it as a strong first draft, not a finished product. Open the new agent and read the instructions. Do they cover your edge cases? Are the tools scoped right? If you spot a gap, you don't start over: click Edit with Sidekick on the agent's configuration page, say what to change ("also capture the customer's name"), confirm, and it updates the agent. (This conversational build and edit is part of Sidekick.)
Then test it before it touches a customer. The Test run button on the same bar has Sidekick draft a realistic scenario, create a real ticket or message on the connected workspace, run the agent autonomously against it, and write a review with concrete tweaks. Sidekick also offers a test on its own right after it creates an agent.
How do you prompt it well?
- Name the trigger and the actions, not just the goal. That's how it knows to wire a trigger and which tools to use.
- Reference what you've connected. It builds from your connectors, so an agent that needs Shopify works best once Shopify is connected.
- Be specific about behavior: when to escalate, what to ignore, what "done" looks like.
- Iterate by asking. Get the first version, then refine in plain language rather than hunting through settings.
When should you build with Sidekick instead of manually?
- Use it for your first agents, for anything standard, and any time describing is faster than configuring (which is most of the time).
- Build manually when you want precise control over every field, or your workflow is unusual enough that adapting a proposal is slower than building from scratch. (See the full build walkthrough for both.)
What agents can you build just by describing them?
Here are three briefs you could paste straight into Build with AI:
- "Create a triage agent: on every new Zendesk ticket, classify the topic, set priority from how urgent it sounds, tag it, and assign billing issues to the Billing group."
- "Build a WISMO agent: when a customer asks where their order is, look it up in Shopify and reply with the status and tracking link; escalate to a human if the order looks lost."
- "Make a summarizer: on every ticket, write a one-paragraph internal note summarizing the issue and what's been tried so far."
For each, the builder reads your connectors, proposes the name, instructions, tools and trigger, and once you confirm, assembles a working agent. You go from a sentence to something you can test in a couple of minutes.
Frequently asked questions
What is the AI Agent Builder? It's Sidekick, Macha's built-in assistant, building an agent from a plain-language description. Click "Build with AI," describe the agent, and Sidekick proposes the name, instructions, tools and triggers, then builds it after you confirm.
Does it set up triggers too, or just instructions? Both. If you mention when the agent should run, it proposes and wires the trigger.
Will it build something without my okay? No. It proposes a plan and always waits for your confirmation first.
Do I still need to review the result? Yes. It's AI-generated, so review the instructions and tools, and use Test run, before going live.
Is the AI agent builder the same as Sidekick? Yes. "Build with AI" opens Sidekick, and "the AI agent builder" is simply Sidekick building an agent for you.
Which plans include the AI Agent Builder? Every plan, including the trial. The trial gives you $50 of free usage with no credit card and no time limit.
Build by describing: open Agents, click Build with AI, and brief it. $50 of free usage, no credit card required. Start free.
