Is it an agent? The four tests

Build-along · ~8 min at your own keyboard

Vendors now call everything an agent. The word covers a lot: in the broad sense, any AI that takes actions with tools is an agent, and you'll meet that meaning in the Builder track. This level is about one kind: the personal agent, which has its own memory, logins and schedule, and asks before anything irreversible. A chatbot answers when asked. A copilot lives inside one app. Four tests show where a tool sits. They are a capability check, not a pass/fail label: a tool can be partly there. And a tool that acts in your accounts with no approval step isn't 'not an agent' — it's an agent you shouldn't trust yet.

Ladder · Level 3The long version: Personal agents: what they are ↗

Do this

0/6

Checkpoints

▸You should now have two tool names and four marks for each: yes, partly, no, or can't tell yet.
▸You should now be able to say, in one line, why each tool passes or fails Test 1.
▸You should now have a count of yeses and a label for both tools.

The three ways this goes wrong

You scored a tool yes on every test because it has a feature with that name. — Test the behavior, not the label. Close the tab for an hour, then check whether it did anything on its own.
You are not sure whether memory counts when the tool only recalls inside one long chat. — Memory means it carries across sessions. Start a new chat and ask what it knows about you. If it knows nothing, that is a no. If it recalls only what you saved to a memory feature, that's 'partly'.
You cannot tell if an approval gate exists. — Ask the tool: 'What will you do without asking me first?' Then check the settings for an approval step. No gate you can find means no on Test 4.

Dated facts in this lesson: ChatGPT agent: Plus gets 40 messages a month, Pro gets 400. Secondary source (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Claude Cowork is available on Pro ($20/month) and above. Claude in Chrome also needs Pro or above. Cowork scheduled tasks run in the cloud with the laptop off (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Gemini Spark is a beta 24/7 agent. It needs Google AI Pro or above. It is not available in the EEA, UK or Switzerland. It can edit Calendar, Docs, Sheets and Drive, and draft email (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Copilot Cowork has been generally available since June 16, 2026. It needs a Copilot license plus usage credits. Secondary source (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Grok Connectors launched May 6, 2026: Outlook, OneDrive, SharePoint, Gmail, Calendar, Drive, Notion, GitHub and Linear. Most can read and write, including sending email and creating invites. Bring Your Own MCP is supported. Automations (daily, weekly, email triggers) need a paid tier (secondary source) (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it); Perplexity has Gmail, Calendar and Outlook connectors. Comet browser is free. Perplexity Computer is on Max. Scheduled Tasks run at most hourly. Secondary source (researched Oct 4, 2026, not yet confirmed by a person: check the source before you rely on it).

Share what happened

Paste two tools, each marked yes / partly / no / can't tell yet against the four tests (not applicable is allowed only on Test 4 for a tool that can't act), with a count of yeses and a one-line label (chatbot, copilot or personal agent).

Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].

🔒 Don't paste confidential, personal-health, or payment info.