How to Create an AI Agent, Without Writing Code
How to create an AI agent without code: pick the job, write the brief, connect your tools, set permissions, test and schedule it. Plus when you need code.
Two years ago, building an AI agent meant writing software. Today most people who create an agent never see a line of code: they describe a job in plain English, connect the apps the agent should work in, and let it run.
This guide covers that no-code route step by step, then explains the two other ways to build an agent (a visual workflow builder, or code) and when each is worth it. If you are not yet sure what an agent is, start with what is agentic AI.
Before you start: three ways to build an AI agent
| Route | How you build it | Best for | Effort |
|---|---|---|---|
| Agent app | Describe the job in plain English | Varied work: research, writing, triage, reports | Minutes |
| Workflow builder | Draw the steps on a canvas; AI fills some boxes | The same process, run identically many times | Hours |
| Code | Write the agent with a developer framework | Products and custom systems | Days or more |
Most people should start with an agent app. You can move to a workflow builder or code later for the jobs that need it. The trade-off is explained in agent or workflow builder?
The steps below use the agent-app route. We use Burrak AI to show each step, because it is ours, but the same steps apply in most agent products.
Step 1: Pick one job, not a whole role
The most common mistake is starting too big: "be my marketing assistant". An agent does best with a job that has a clear finish line.
Good first jobs share three traits:
- It repeats. Weekly, daily, or every time something happens. That is where the time savings add up.
- It is easy to check. You can tell in a minute whether the output is right.
- Nothing breaks if it is wrong. A summary you read, not an email sent to a customer.
Examples: a Monday summary of competitor news; a morning list of which emails need your reply; a Friday revenue check; turning call notes into a proposal draft.
Step 2: Write the brief like you would for a new hire
The brief is the agent. A good one answers four questions:
- What should exist when the job is done? "A one-page summary", not "look into competitors".
- What context matters? Who it is for, the format, the length, the tone, what to ignore.
- Where is the material? The website, the folder, the inbox, the spreadsheet.
- What does done look like? "Every competitor on the list is covered, each with a link."
Here is a complete brief:
Every Monday at 8am, check the news and blog pages of Acme, Globex and
Initech for anything they announced in the past 7 days: new products,
price changes, hires, funding. Write a one-page summary with a heading
per company and a link for every claim. If a company announced nothing,
say so in one line. Email it to me.
That is the whole "build". No triggers, no flowchart.
Step 3: Create the agent and give it a role
In Burrak:
- Create a free account on the web, in the desktop app, or in the iOS or Android app. All of them are free to start.
- On the Overview page, create an agent and name it, for example "Competitor watch". Choose a remote agent if you want it to work while your computer is off. Remote agents run in the cloud around the clock.
- Optionally, give it a role from the Marketplace, such as Market Researcher or Chief of Staff. A role is a written persona that shapes how the agent approaches every task, so you start from a specialist rather than a blank prompt.
You can create several agents side by side, each with its own name, chat history and memory: a researcher, an inbox assistant and a report writer, for example.
Step 4: Connect the tools it needs, and only those
An agent can only act in the apps you connect. In Burrak these are called connectors: Gmail or Outlook, Google Calendar, Sheets, Docs, Notion, Stripe, and others. You connect one by signing in to it and approving what the agent may do.
Two rules keep this safe:
- Connect only what this job needs. The competitor-watch agent needs web search and your email, not your accounting software.
- Start with the smallest permission. Where an app offers read-only access, take it. For email, ask for drafts rather than sending until you trust the result.
Step 5: Run it once and read everything
Before you schedule anything, run the brief once in chat and read the result closely:
- Did it cover everything in the brief?
- Are the facts right? Click the links.
- Is the format what you wanted?
Then open the run's transcript, the step-by-step record of what the agent did. It shows you where it looked, what it skipped and why. Most fixes are one sentence added to the brief: "ignore job postings", "use bullet points, not paragraphs".
Expect to adjust the brief two or three times. That is normal, and it is much faster than redrawing a workflow.
Step 6: Put it on a schedule
Once one run is right, make it recurring. In Burrak you do this from the Tasks page: describe the cadence in plain English ("every Monday at 8am", "every weekday at 9am"), pick the agent, and choose the AI model it should use. Routine work can run on a cheaper model; important work on a stronger one.
Two settings are worth turning on for anything that runs unattended:
- Verification, so the agent checks each run against your goal and retries if it falls short.
- Notifications, so you hear when a run finishes. Free accounts get them on mobile; paid plans add desktop and email, and you can reply to the email to give the agent its next instruction.
Our article on what breaks when nobody is watching covers why these matter.
Step 7: Review weekly, then widen its job
For the first few weeks, read every result. When the agent has been right several times in a row, give it a little more: a second competitor list, a draft reply instead of a summary, a new data source.
Agents with memory improve here. Burrak keeps notes on your preferences between runs, which you can read and delete on the Memory page, so corrections you make once tend to stick.
What it costs
Agent products bill for the work the AI does, usually in credits, because a ten-step job uses more computing than a single chat message.
In Burrak, the account is free on every app, with pay-as-you-go credits and up to two remote agents. Paid plans add a monthly credit allowance: Pro is $9.99 a month with 12,000 credits, and the first week of any paid plan is $1. You are only charged while an agent is working; an idle agent costs nothing. The full breakdown is on pricing.
When to use a workflow builder or code instead
Choose a workflow builder (such as Zapier, n8n or Lindy's flow editor) when the process must run the same way every time, thousands of times, and someone needs to audit each branch: copying form submissions into a CRM, routing invoices. An agent can call Zapier through its connector when you want both.
Choose code when the agent is your product: something your customers use, embedded in your app, with your own interface. Developer frameworks for this include the OpenAI Agents SDK, Anthropic's Claude Agent SDK and LangGraph. Expect to handle hosting, model costs, security and monitoring yourself.
For an agent that does work for you or your team, the no-code route above is almost always the faster start.
Frequently asked questions
Can I create an AI agent for free? Yes. Several agent apps have free tiers. Burrak's free account works on the web, desktop, iOS and Android with pay-as-you-go credits, so you pay only for the work your agents do.
How long does it take to create an AI agent? With an agent app, a few minutes to create it and an afternoon of small adjustments to the brief. With a workflow builder, hours. With code, days to weeks.
Do I need to know how to code to build an AI agent? No. In an agent app you write what you want in plain English, choose a role from a list and sign in to the tools it should use.
What is the difference between an AI agent and a chatbot? A chatbot answers and waits for your next message. An agent takes a goal, works through the steps using tools, and hands back finished work. See AI agents vs chatbots.
Can an AI agent run while my computer is off? A cloud (remote) agent can. In Burrak, remote agents run 24/7 and notify you when the work is done. A local agent runs on your own computer and works with your files, but only while the computer and the desktop app are on.