How to create an AI agent with the Swiftask agent builder?

Since the update of Swiftask's agent builder, any employee can configure their own AI assistant in a few minutes — simply by describing what it should do, in natural language. Here is how to proceed, step by step.

Create an AI agent on Swiftask: three starting points

When you click "Create agent" from the Swiftask interface, a window opens with three distinct options. Each one matches a different user profile.

Create with AI

You describe your goal in everyday language. A dedicated agent — The Maestro — automatically generates the instructions, the tools, and the complete configuration of your future assistant. It is the fastest and most accessible method.

Use a template

You start from a pre-configured agent available in the Swiftask library — SEO writer, community manager, competitive analyst, and more — and adapt it to your needs. Ideal when a proven workflow already matches your use case. Browse the AI agent templates library to see what is available.

Start from scratch

You define every parameter manually: instructions, behavior, language model, skills. This option gives you total control, but it is designed for users who know exactly what they want to configure.

This tutorial focuses on the first option — the most popular among enterprise teams.

What the "Create with AI" option actually does?

The "Create with AI" option relies on The Maestro, a configuration agent built into Swiftask. Its role: guide you through building your virtual assistant with a structured conversation — not a technical form.

The Maestro extracts, throughout the dialogue, all the information it needs:

• The agent's exact mission (write, analyze, monitor, respond, etc.)
• The target audience and the expected level of technicality
• The tools to connect (internet search, internal knowledge base, code generation...)
• The language model best suited to the use case

With every answer you give, the configuration schema fills in automatically on the right side of the interface. You watch your agent take shape in real time: selected language model, written instructions, associated skills.

Step 1 — Describe your agent in natural language

Click "Create with AI". The Maestro appears and immediately invites you to describe what your future virtual coworker should do.

How do you write a good description?

No need to look for technical phrasing. Write what feels logical, as if you were explaining to a new colleague what is expected of them. The more precise you are, the better your agent will perform from its very first run.

Here is what to include:

• The main role: what should it do concretely? (write articles, analyze data, answer emails, monitor competitors...)
• The audience: who does it work for? (just you, a team, customers?)
• The business context: marketing, customer support, legal, HR, sales...
• The constraints: expected tone, output format, length, language

A description like this, free of technical jargon, is enough to generate a complete writing agent — with a recommended language model, system instructions, and activated tools.

Step 2 — Talk with The Maestro to refine every parameter

Once your description is sent, The Maestro analyzes the available information and asks targeted questions to fill in what is missing.

These questions follow a progressive logic:

1. Type of content or tasks — what concrete deliverables should the agent produce?
2. Target audience — who will read or use the results?
3. Specific goals — generate traffic, convert, inform, monitor?
4. Work process — should it research before acting, remember user preferences?
5. Language model choice — The Maestro compares the available options and recommends the one that best fits your use case and your budget

On the language model choice, The Maestro presents a reasoned comparison: writing quality, balance between creativity and rigor, cost per use. You decide with full knowledge of the trade-offs.

On the right side of the screen, the schema evolves with every exchange. You see appear in real time:

• The agent's identity card (name, description, generated icon)
• The primary language model + a fallback model
• The system instructions, written automatically
• The activated skills: internet search via Perplexity, user memory, code and data visualization generation, and more

Step 3 — Validate the final configuration

When the conversation with The Maestro is over, your agent is fully configured. A structured summary is displayed before deployment.

At this stage, the connected tools are also listed with their explicit role in the agent's process. You know exactly what your virtual assistant can do and how it works.

Two options are then available:

Option A — Try it right away: click the "Preview & Test" tab at the top of the interface. Start a first conversation with your agent to check it behaves as expected.

Option B — Switch to manual configuration: click the "Manual configuration" tab. You access the full set of detailed settings to adjust the instructions, the connected tools, the behavior constraints, or the data the agent can access.

These three tabs — Agent builder, Manual configuration, and Preview & Test — remain accessible at all times at the top of the interface, so you can navigate freely at any moment.

Step 4 — Test your AI agent and refine the instructions

A created agent is not a finished agent. The value of an AI assistant is built through iteration.

Recommended process after creation:

1. Open the "Preview & Test" tab from the Swiftask interface
2. Start a conversation with your agent on a real use case — not an ideal one
3. Observe the responses: tone, format, consistency, tool usage
4. Go back to "Manual configuration" to adjust the instructions if needed
5. Run another test and repeat until you get the expected behavior

What to check during testing:

• Does the agent respect the defined tone and format?
• Are its answers consistent across similar topics?
• Does it use the connected tools correctly (search, knowledge base)?
• Are its outputs suited to the target audience's level?

Once you are satisfied with the behavior, Swiftask lets you lock the agent to protect its configuration, or share it with your teams directly from the interface. The step-by-step agent documentation covers each of these settings in detail.

Every agent created on the Swiftask platform can be connected to your business tools, powered by your internal databases, and deployed across your organization — with full governance over access, usage, and costs. No tool sprawl. No shadow IT.

You can even manage a fleet of AI agents to cover all your team's needs — each agent specialized in one specific task, while sharing the same data and knowledge. Compare the team plans and pricing to get started.