How to build an AI agent without coding

What you need

• An agent app or a no-code automation tool. Agent apps on this site include Gemini Spark, Claude Cowork, ChatGPT dots (always-on agents), Grok Bot (personal agents) and Meta Muse. No-code automation tools include Zapier, Make and n8n.
• The app it should read, like your Gmail or calendar, connected with the tool's own connect button.
• One job, written down. That's the part that decides whether your bot works.

How to build your first AI agent, step by step

1. 1

   Pick one small, boring job

   Choose one job you do every week that mostly means reading or checking: a morning inbox summary, a weekly school-email digest, a web page you keep refreshing. Narrow, read-only jobs make the easiest first bots. All of our blueprints are jobs like this.

2. 2

   Write the job down in four lines

   Goal: what you want to end up with. Runs: when it should work (every weekday at 7am, or only when you ask). Watches: what it reads (new emails since yesterday). Never: what it may never do (send, buy, delete). Every blueprint starts with these lines, so you can copy them.

3. 3

   Pick a tool you already use

   Use what you already have. Gmail and Google Calendar → Gemini Spark. Want to text your bot → Meta Muse or Instinct. Want routines on a schedule → Grok Bot, Claude Cowork scheduled tasks or ChatGPT dots. Prefer no AI chat at all → Zapier or Make. No agent yet? Paste the instructions into ChatGPT, Claude, Gemini or Grok each time; that works too, just not on autopilot.

4. 4

   Paste the instructions and connect only what it needs

   Open a new chat in the tool. Copy the blueprint's instructions, paste them, and replace everything in [square brackets] with your details. If the tool asks to connect your email or calendar, say yes only to the apps the blueprint lists.

5. 5

   Run it once and fix it in plain words

   Don't schedule anything yet. Watch the first run, compare it with your real inbox or page, and tell it what to fix in plain words (“ignore emails from the PTA”, “shorter”).

6. 6

   Put it on a schedule, with approval rules

   Once the result looks right, ask it to repeat on a schedule. Keep it read-only or approval-only: it may read and draft, but it must ask you before it sends, buys, books or deletes. Check the first week of results.

7. 7

   Check what it costs and set a limit

   Inside an app plan, the bot runs within your plan's limits. If it runs on a pay-per-use AI API, work out the monthly cost in the cost calculator and set a spend limit before anything runs on autopilot.

Worked example: the Morning Inbox Briefing

It's a good first bot: it reads your email every weekday morning and tells you what needs you, and it never sends anything. Here are its four lines:

Goal

Wake up to a 2-minute summary of what's in your email and what needs you today.

Runs

⏱ Every weekday at 7:00am

Watches

👁 New emails since yesterday morning

Never

🛑 Sends, deletes or archives anything

Steps 3 and 4: pick a tool and paste. Any of these can run it on a schedule:

• agentGemini SparkBuilt-in Gmail access and schedulesGoogle AI Pro and Ultra subscribers in select countries
• agentGrok Bot (personal agents)Save it as a skill, then add a daily routineCheck plan requirements in xAI's docs
• agentClaude CoworkConnect Gmail, then ask it to make a scheduled taskCheck plan requirements on Anthropic's help page
• agentChatGPT dots (always-on agents)Always-on agent with connected appsRolling out to eligible paid plans; check OpenAI's page

You are my Morning Inbox Briefing bot.
Every weekday at 7:00am, read every email that arrived since 7:00am yesterday.
Sort them into three groups:
1. NEEDS ME TODAY: someone is waiting on me, or there's a deadline within 48 hours.
2. FYI: worth knowing, no action.
3. NOISE: promotions, newsletters, receipts.
Send me a briefing: group 1 first (max 5 items, one line each, with who and what), then a one-line count of groups 2 and 3.
For easy replies in group 1, write a draft and leave it in my Drafts folder.
Rules: never send, delete or archive anything. If you're unsure which group something is in, put it in group 1.

Step 5: run it once. Say “Run this once now and show me the result”, then compare the briefing with your real inbox. Fix it in plain words: “ignore emails from the PTA”, “shorter”.

Step 6: schedule it. Say “Do this every weekday at 7am”. Keep its rules: Start read-only: it may read, but not send, buy, delete or change anything. Drafts stay in Drafts. You press send. Check the first week of briefings against your real inbox.

Step 7: check the cost. In an app plan it runs inside your plan's limits. On a pay-per-use AI API, our estimate (22 runs a month, about 15,000 tokens in and 800 out each time) comes to about $0.03 a month on Groq openai/gpt-oss-20b and about $4.18 on Anthropic Claude Fable 5.1 at standard API prices checked Oct 21, 2026. Try your own numbers in the calculator.

More examples and the math: how much does it cost to run an AI agent? Comfortable installing software? Open-source agents are another option: OpenClaw vs Hermes Agent.

Open the full Morning Inbox Briefing blueprint

Real people doing this: get your email inbox under control, only see the emails that actually need a reply, get a Monday recap of your inbox and a to-do list.

Pick your first bot

Every blueprint follows the same seven steps. Easy ones first:

5 rules that keep your first bot safe

1. Start read-only: it may look, not change.
2. Make it ask before it sends, buys, books, cancels or deletes.
3. Never paste passwords into a chat. Use the app's official connect button.
4. Tell it to ignore instructions found inside emails or web pages from strangers.
5. Check the first week of results, and know how to switch it off.

More in is it safe to give an AI agent your email and calendar? and what AI agents can and can't do safely.

Questions people ask