THE METHOD

Three Claude "Agents" You Can Set Up in Ten Minutes, and Where They Stop Being Agents

2026-10-09

The short answer: you can have three genuinely useful helpers running in about ten minutes each, without writing any code. One briefs you before meetings, one sends a weekly competitor report, and one builds a list of prospects to contact. All three use features already built into Claude: connectors, skills and scheduled tasks.

A caution before you start. Most of what gets called an agent right now is really a prompt plus a few connected apps plus a schedule. That is still useful. It just isn't the same thing as a system that decides its own next steps. If you want the difference spelled out, we cover it in our explainer on chatbots, workflows and agents.

AI Agents, Explained: The Difference Between a Chatbot, a Workflow, and an Agent

The setup path, in four steps

• Connect the app. In Claude, open Customize, then Connectors, and connect the service you want, such as Google Calendar or Gmail.
• Turn it on in a chat. Click the plus sign in the message box, choose Connectors, and switch on the ones you need for that conversation.
• Save your instructions as a skill if you will reuse them. Skills live under Customize too, and they save you from retyping the same prompt.
• Schedule it. For anything that should run on its own, create a scheduled task: in Cowork, choose Scheduled, then New task, and pick hourly, daily, weekly or weekdays.

Helper 1: The meeting briefer

What it does: before each meeting, it looks at who is attending and gives you a one-page brief, so you walk in knowing who they are and what you last discussed.

Connect Google Calendar, and Gmail too if you want it to find past conversations with the same people. Then give it instructions like these:

"Look at my calendar for tomorrow. For each meeting with someone outside my company, tell me who is attending, what we last discussed if you can find it in my email, and three questions worth asking. Keep each brief to ten lines. If you can't find anything, say so instead of guessing."

Run it by hand a few times first. Once the briefs look right, schedule it for the evening before. This helper only reads, so the worst it can do is give you a thin brief.

Helper 2: The weekly competitor report

What it does: once a week, it checks a short list of competitors and reports what changed, such as new products, pricing, announcements and hiring.

Keep the list short. Three to five companies gives you a report you will actually read. Name each company and the pages worth watching.

"Every Monday, check the websites and recent news for these companies: [list]. Report only what is new since last week: product changes, pricing changes, announcements and notable hires. For each item, give the source link. If nothing changed, say 'No changes this week' and stop."

The line that matters most is the last one. Without it, a quiet week produces filler. Also ask for source links so you can check anything that looks important. Competitor claims are exactly the kind of thing worth verifying before you act on them.

Helper 3: The lead researcher

What it does: given a description of your ideal customer, it builds a short list of prospects and writes a personal opening line for each one.

This one works well as a saved skill, because you will run it again with different targets. Describe who you sell to and what makes a good fit, then ask for a small batch.

"Find ten companies that fit this profile: [your ideal customer]. For each, give the company name, why it fits, the role of the person I should contact, and a two-sentence opening message that mentions something specific and recent about them. Mark anything you couldn't confirm."

Two cautions. First, check that every company and detail is real before you send anything, because AI will sometimes produce a convincing but wrong example. Second, check the rules that apply to your outreach, such as anti-spam laws in your country. Review every message yourself before it goes out.

What these can't do

People often ask for two more helpers: one that answers your inbox for you, and one that writes and sends launch emails. Those are riskier, and the tools may not be ready for them.

Anthropic's own documentation is currently inconsistent on Gmail. One help article says Claude can draft, send and reply with your approval. A separate documentation page says the Gmail connector is read-only. Because the two disagree, test what your own connector can do before you build a plan around it.

Even where sending is possible, we would keep a person in the loop. A message sent to a customer can't be unsent.

The NoHypeAI Reality Check

These three are worth setting up, and they are cheap to try. But be clear about what you have built: scheduled prompts that read your connected apps and produce a report. They do not take independent action, they can be wrong, and they only know what your connectors and the web show them.

A good rule: start with helpers that only read, check the output by hand for a few weeks, and add anything that sends or changes things only after you trust it.

Sources: Anthropic's pages on Google Workspace connectors, the Gmail connector and scheduled tasks in Claude Cowork. Features and plan limits change often, so check them before you build.