Shopify
Stripe
Slack
Notion
Google Workspace
Confluence
What Is Macha's AI Agent Builder? Describe an Agent and Macha Builds It
Setting up a support agent used to mean writing instructions, picking tools, wiring a trigger, deciding on sub-agents and choosing a model by hand. Macha's AI Agent Builder does all five from one plain-English description, using the integrations your account already has connected.
Key takeaways
- Macha's AI Agent Builder, shipped April 3, 2026, turns a plain-English description into a configured agent with instructions, tools, triggers, sub-agents and a model in one pass.
- Agents generated by the Macha AI Agent Builder run on GPT-5.4 by default, and the model can be changed per agent without changing what a ticket costs.
- A refund-triage agent generated by the Builder used Zendesk's Get Ticket, Add Comment and Update Ticket tools alongside Stripe's Get Customer, List Payments and Create Refund.
- The Builder runs automatic circular-dependency checks when it assigns sub-agents, so a generated agent cannot be wired into an infinite loop by accident.
- Every Macha customer gets the AI Agent Builder, because Macha sells one plan with every feature, priced by ticket volume from $299 a month for 750 tickets.
Macha's AI Agent Builder, shipped on April 3, 2026, turns a plain-English description into a configured support agent in one pass: instructions, tools wired to your actual connected integrations, triggers, any sub-agents, and the model. You write something like "triage incoming Zendesk tickets, tag the billing ones, and draft a reply for everything else", and what comes back is a real agent you can run, inspect and edit, not a stub you have to finish by hand. It is included in Macha's single plan, which carries every feature, and the same build-from-a-description ability now also sits in Sidekick, the AI helper built into the Macha dashboard.
| Question | Answer |
|---|---|
| Shipped | April 3, 2026 |
| What it sets for you | Instructions, tools, triggers, sub-agents, model |
| Default model | GPT-5.4 (change it per agent) |
| Where you use it | The AI builder on the Agents page, and Sidekick |
| Who gets it | Every Macha customer; one plan, priced by ticket volume |
| Companion | AI Tool Builder, for custom API tools |
What does the AI Agent Builder do?
The AI Agent Builder is a conversation that ends in a configured agent. The launch was deliberately plain about it: describe what you want in plain language and the AI creates a fully configured agent with the right tools, instructions, triggers, and sub-agents. A follow-up refinement added the last piece it sets for you, model selection, so the agent that lands is ready to test rather than half-built.
Three things make this more than a text box:
- It knows your account. The Builder already knows which integrations your org has connected (your help desk, Notion, Stripe, your custom API tools) and suggests configurations that use those tools, not a generic catalogue you don't have. Hover a suggestion pill and it previews the exact prompt it will run.
- It builds the whole agent. Instructions, tools, triggers, sub-agents and model, in one pass. It can assign sub-agents while it creates an agent, with automatic circular-dependency checks so you can't accidentally wire an infinite loop.
- The output is real and clickable. Created agents, triggers and tools appear as interactive artifact cards in the chat. Click an agent card and you land on its configuration page, where every field the AI filled in is yours to keep, change or throw away.
How does the Builder turn a description into an agent?
You start by telling the Builder what you want, in your own words. There's no schema to learn; you write the way you'd brief a new teammate.
Macha reads the request against your account, drafts the agent, and streams the result back as artifact cards. Because the Builder knows your connected stack, a request like "refund triage agent" reaches for the connectors you actually have (your help desk for reading and replying to tickets, your billing tool for the refund, your knowledge sources for policy) instead of inventing capabilities you'd then have to go set up.
From there you open the agent and review it like any other. Nothing is locked. The Builder gets you from a blank screen to a sensible first draft in one message. Tuning and shipping are still yours: chat with the agent in the Try it sandbox, where you can reject confirmation cards to preview writes without firing them, or replay past Zendesk tickets through Agent Simulations with write tools stubbed. Both go faster from a populated config than an empty one.
What does a generated agent look like?
Type "build me a refund-triage agent for Zendesk that processes straightforward refunds in Stripe and replies on the ticket" into the Builder with Zendesk and Stripe already connected, and the agent that lands is fully populated. Here is the config it produces, field by field:
| Field | Generated value |
|---|---|
| Name | Refund Triage |
| Handle | @refundTriage (auto-generated camelCase from the name) |
| Instructions | "When a customer requests a refund, read the ticket to understand the request, look up the customer's recent payment in Stripe, and confirm eligibility. For straightforward, fully refundable orders, issue a full refund to the original payment method, reply on the ticket confirming the amount and a 5–7 business-day timeline, then mark the ticket Solved and tag it refund-processed. If the request is ambiguous, partial, or outside policy, draft a reply and hand off to a human instead of refunding." |
| Tools | Zendesk → Get Ticket, Add Comment, Update Ticket · Stripe → Get Customer, List Payments, Create Refund |
| Trigger | Zendesk Ticket Created, condition: customer requests a refund |
| Model | GPT-5.4 (the default) |
Every one of those fields is editable. If you'd rather the agent never refund on its own and only ever draft, delete the Create Refund tool and tighten the instructions, by hand on the form or by asking the AI editor to "make this draft-only." The point is that you start from this filled-in config instead of a blank one. The deep-dive linked below walks through building, testing and shipping an agent like this end to end.
Can you edit an existing agent by conversation?
Yes. Open the AI editor from any agent's configuration page and you can change any part of it (tighten the instructions, add a tool, set up a trigger, wire in a sub-agent) by asking instead of clicking through the form. The builder and the editor are the same idea pointed at two moments: standing an agent up, and changing it later.
How is it different from Agentforce Builder and other agent builders?
Describing an agent in plain language is the product trend of 2026. Salesforce's Agentforce Builder lets you "describe what you want an agent to do and AI will help generate subagents, instructions, and actions"; ServiceNow and Salesforce have been racing each other to ship agent studios, Talkdesk and Decagon have their own describe-the-agent flows, and horizontal tools like Lindy, Airtable's Omni and Glean do it for general workflows. The idea is good. The question is what the builder starts from.
Most of those tools are one of two things: a horizontal workflow canvas that gives you a blank graph and a thousand possible integrations, or a platform-locked suite that only builds inside its own walled garden. Macha is neither. It is an AI agent layer that sits on top of the help desk you already use (Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom) and connects to your commerce, knowledge and comms tools alongside it.
That changes the first draft. A horizontal builder can't know that your help desk is Zendesk, that your refund logic lives in Stripe, and that you already indexed your policy docs in Notion, so its first guess is generic and you spend the next hour mapping it to reality. Macha's Builder starts from your connected account, so the first draft is already shaped to your stack.
When should you still build an agent by hand?
The Builder is a head start, not a replacement for judgment. Four caveats:
- It drafts; you own the result. Review what it wired up before you turn an agent loose, especially anything that writes back to customers or takes an action. You're accountable for the final config.
- It can only reach what you've connected. Ask for an agent that uses a tool you haven't set up and it can't conjure the integration. Connect the integrations you need first, then describe the agent, and the Builder will pick them up.
- For a surgical change, the form is faster. If you know the exact field you want to change, edit it directly. The Builder is at its best going from zero to a working draft.
- Building by AI still runs a model. Macha bills per ticket: one conversation is charged once however many messages it takes, and the model you pick doesn't change the charge. Refining an agent over ten messages costs the same as getting it right in one. See pricing for the tiers.
Where can you read more?
For the full step-by-step (generated instructions, tools, triggers and sub-agents, with more screenshots and a worked example) read the deep-dive: Build an AI Support Agent Just by Describing It. The docs cover the builder, the editor and Sidekick in reference detail.
FAQ
What is the Macha AI Agent Builder? A conversational way to create support agents. You describe the agent you want in plain English and Macha generates a configured agent: instructions, tools wired to your connected integrations, triggers, sub-agents and model. It launched on April 3, 2026.
Which plans include it? Macha has one plan with every feature, so every customer gets the AI Agent Builder. The price is set only by monthly ticket volume, starting at $299 a month for 750 tickets. See the pricing page for the other tiers.
Does it work with my helpdesk? Yes. Macha is a layer on top of your existing help desk (Zendesk, Freshdesk, Gorgias, Front, HubSpot or Intercom), and the Builder uses the integrations you've already connected when it drafts an agent.
Can I edit what it builds? Always. The generated agent is a normal Macha agent. Every field is yours to change by hand, or you can keep using the conversational AI editor from the agent's config page.
Which model does a generated agent use? GPT-5.4 by default. You can switch the model per agent on its configuration page, and the choice doesn't change what a ticket costs.
Is there a separate tool builder? Yes. A companion AI Tool Builder lets you describe an API, paste your credentials, and have Macha discover endpoints, test them and create the tools. Those tools then show up for the Agent Builder to use.
How do you try the Builder?
Start a free trial ($50 of usage, about 125 tickets, no credit card and no time limit), connect your help desk, and tell the Builder what you want your first agent to do. The Macha team also builds, tests and monitors agents with you as part of the plan.
Written by Abbas (Customer Support & AI, Macha) · Reviewed by Ankeet Guha (Co-founder & CTO) · Published 2026-06-24.
