# Your first AI agent, start to finish

A step-by-step walkthrough: describe one job, read the draft, try it safely, put it on a schedule and check its first runs

Oct 23, 2026 · Guides · 8 min read · https://arclyx.ai/blog/ai-agent-tutorial

## Summary

This AI agent tutorial for beginners builds one real agent from start to finish, with no code: a competitor watch that checks three competitors every Monday and posts what changed. You describe the job in one sentence, answer a quick question, read the draft Arclyx writes, try it once without anything being sent, then let it run on a schedule. Each step below takes a few minutes, and the last one shows what to check after its first runs.

## Key Takeaways

- **No code, no flowchart.** The whole agent comes from one sentence and your answers. Agent frameworks in Python are for developers; this tutorial isn't that.
- **Start with one job that repeats.** An agent earns its place on work that comes back every week, with nobody there to ask.
- **Name four things in the sentence.** The job, the tools, when it runs and where the result goes. Arclyx asks about anything you leave out.
- **Try it before it runs alone.** Try it on a draft is a dry run: it reads with your real tools and shows you what it would post instead of posting it.
- **A draft never runs on its own.** Schedules only start agents you've created, and Arclyx never invents a schedule you didn't name.
- **Check the first few runs.** Every run is recorded step by step, and an agent that stops short shows up as Needs you on Home.

## What is an AI agent, in simple terms?

An AI agent is software that does a job for you with tools, not just answers a question. You give it the job once; it decides the steps, uses the tools it was given, such as web search or your inbox, and hands back the result. IBM defines it as a system that autonomously performs tasks by designing workflows with the tools it has [Source 2].

The difference from a chat assistant is who starts the work. A chat answers when you ask. An agent can start itself on a schedule or an event, do the same job every time, and leave a record of what it did. Anthropic draws a related line: workflows follow predefined code paths, while agents direct their own process and tool use [Source 1].

In this tutorial the job is a competitor watch. Every Monday it reads three competitors' pricing, product, news and careers pages, compares what it finds with your own notes, and posts a short digest with a link for every change.

## What do you need before you build one?

You need an Arclyx account, one job you want done, and the tools that job uses. For the competitor watch, that's less than it sounds: web search and reading web pages work without connecting anything, so the only setup is somewhere to post the digest and, if you have them, your competitor notes.

| You need | For the competitor watch | Why |
|---|---|---|
| A job that repeats | Check competitors every Monday | Agents pay off on work that comes back |
| Tools | Web search and web pages | Built in, nothing to connect |
| Your notes (optional) | Upload them to Documents | So it reports what's new, not what you know |
| A place for the result | A Slack channel, or your email | Where the digest lands |

If you'd rather not set up Slack yet, end the sentence with "and email me the digest" instead. An agent's email only goes to the person it runs for, so that email goes to you.

## How do you describe the job in one sentence?

Name four things: the job, the tools it works in, when it runs and where the result goes. Plain words are enough. Open New agent in Arclyx and type the sentence into the "What should it do?" box. The ready-made jobs listed under the box are another way in, once you've seen how one is built.

![The New agent screen in Arclyx with example data: a one-sentence inbox job typed into "What should it do?", ready-made jobs below it, and a draft preview on the right noting that nothing runs on its own until you create it.](https://arclyx.ai/blog/ai-agent-tutorial/img/new-agent-screen.6685e529-1600.webp)

Here is the sentence this tutorial types in place of the inbox job in the picture, the same one the [Competitor watch agent](https://arclyx.ai/agents/competitor-monitoring) starts from:

```text
Every Monday at 8, check Competitor A, B and C for pricing, product, news
and hiring changes since last week, compare them with our competitor notes,
and post a cited digest to #competitive.
```

Read it against the four parts. The job is "check for pricing, product, news and hiring changes" and "compare them with our notes". The tools follow from that: the web and your documents. When is "every Monday at 8". Where it goes is "#competitive". Our page on [how to create an AI agent](https://arclyx.ai/product) has more examples of good and weak sentences.

## What does Arclyx ask, and what does it draft?

Arclyx asks at most three quick questions about what the sentence leaves out, each with an answer already picked, so you can accept it or change it. Then it drafts the agent: a name, the instructions it follows on every run, the tools it may use and, because the sentence named one, its schedule.

In the example below, the competitor sentence gets one question: where your competitor notes live, with your documents picked as the answer. Then the draft appears: Competitor watch, with document search, web search, web fetch and Slack as its tools, running Mondays at 8:00.

![Competitor watch in four steps with example data: the sentence and its one question, the draft with its tools and Monday 8:00 schedule, a dry run of five rounds with the Slack post marked not sent, and the digest it posts once created, with a source link for every change.](https://arclyx.ai/blog/ai-agent-tutorial/img/competitor-watch-four-steps.99f9e982-1600.webp)

Read the draft before you do anything else, especially the instructions and the tools. If something is off, say what should change in plain words and Arclyx rewrites the draft. Our guide to [writing AI agent instructions](https://arclyx.ai/blog/ai-agent-instructions) shows what good instructions include.

## How do you test your first agent without sending anything?

Press Try it on the draft. It's a dry run: the agent reads with your real tools and does every step, but anything it would post, send or write through a connector is held and listed for you as not sent. You see exactly what the Monday digest would say, and nothing reaches #competitive.

In the competitor watch's dry run, each round shows as it happens: it reads your notes, searches the week's news, reads each page fresh, compares with your notes, and holds the Slack post. Check three things:

- **Did it read the right pages?** Each round shows the tool and what it asked for.
- **Is every change linked?** Each item in the digest should cite the page it came from.
- **Is the post going to the right place?** The held post shows where it would have gone.

A dry run uses your real tools, so it counts toward your usage. If the result isn't right, change the draft in plain words and try again. The [test runs page](https://arclyx.ai/product/test-runs) covers how to test edge cases, such as a week where nothing changed.

## How do you put it on a schedule?

If your sentence named a schedule, such as every Monday at 8, the draft already has it. Press Create agent, and from then on it runs on that schedule in your time zone. If you didn't name one, Arclyx doesn't invent one: the agent runs only when you ask, and you can add a schedule on its page later.

![An agent's page in Arclyx with example data: Add a schedule set to Weekdays at 8:00 in your time zone, and the Triggers panel listing a weekday schedule, a webhook and "Any time you ask", each schedule and webhook with Pause, Edit and Remove.](https://arclyx.ai/blog/ai-agent-tutorial/img/agent-triggers.87d7fdf8-1600.webp)

Schedules can be daily, weekdays, weekly or monthly, at the time you pick. If the draft can't set the schedule you named, such as every other Monday, it says so and sets none, rather than guessing; add one under Triggers instead. Each schedule sits in the agent's Triggers panel with Pause, Edit and Remove. A draft never runs on its own: schedules and webhooks only start agents you've created.

For the digest to reach Slack, pick the agent's one channel on its page and invite the Arclyx app to that channel with `/invite @Arclyx`. The agent can only post there. Schedules aren't the only way to start an agent; the [triggers page](https://arclyx.ai/product/triggers) covers webhooks and Slack mentions too.

## What should you check after its first few runs?

Open its first real runs and read them, the way you read the dry run. Each run's page shows what happened in a sentence, the numbers, and every round with the tool it used and what came back. After a few Mondays, look for a run that took much longer or cost more than the rest.

![Home in Arclyx with example data: running now, done this week, one agent that needs you and spend this month, the agent cards with Meeting brief marked Needs you after a failed run, and recent activity below.](https://arclyx.ai/blog/ai-agent-tutorial/img/home-needs-you.985d85ac-1600.webp)

You don't have to check every day. If a run stops short or a connection breaks, the agent shows as Needs you on your home page until you look, and a schedule that fails five times in a row pauses itself. When something needs fixing, change the agent in plain words with Improve, then press Run it now on its page. Run it now is a real run, so it posts for real. Our guide on [how to debug AI agents](https://arclyx.ai/blog/debug-ai-agents) walks through reading a run that went wrong.

## What should you build next?

Build the next agent from a job you already do by hand every week. Good second agents read where your work already lives: your inbox, your calendar, your CRM. Each one needs its tool connected once with your own sign-in, and the draft lists exactly which tools it gets.

- **Start from a ready-made job.** The [AI agents for business](https://arclyx.ai/agents) page shows jobs that run in Arclyx today, each with the sentence behind it.
- **Connect your tools.** Gmail, Google Drive and Calendar, Outlook, Notion, GitHub, HubSpot and more each connect with your own account; the [connectors page](https://arclyx.ai/product/connectors) lists what each one can do.
- **Let your team ask it.** Our guide to using a [Slack AI agent](https://arclyx.ai/blog/slack-ai-agent) shows how people @mention Arclyx in a channel and get the answer in the thread.

## Sources

1. Anthropic — *Building effective agents* (19 December 2024). [https://www.anthropic.com/research/building-effective-agents](https://www.anthropic.com/research/building-effective-agents)
2. IBM — *What are AI agents?*, Anna Gutowska (IBM Think, checked 23 October 2026). [https://www.ibm.com/think/topics/ai-agents](https://www.ibm.com/think/topics/ai-agents)