Question 1

What should I hand to an AI agent, and what should I keep doing myself?

Accepted Answer

Hand an agent work that takes you a long time, is easy to check and is cheap to undo: gathering and comparing information, clearing newsletters out of an inbox, filling a routine form, drafting a weekly report from your own files. Keep the steps where a mistake is costly or hard to reverse, such as moving money, accepting terms, writing to people who matter to you, or anything medical or legal, and keep the decisions you enjoy or need to learn from. Ethan Mollick's test weighs how long a task takes you against the time to brief the agent and check its work. The tools draw the same line: dots hand password changes and money transfers back to you, and Google advises against scheduling tasks you find sensitive.

Question 2

How do I brief an AI agent so it does the job I actually want?

Accepted Answer

Brief it like a capable new assistant. Start with the outcome and why it matters: "a comparison of three family laptops so I can choose by Friday." Say what done looks like: the format, the length, where to save it. Give it what it needs: attach the files, connect only the accounts the job uses, and name the sites to trust. Then set the limits: what it may do alone, and when to stop and ask, for example before paying, booking or messaging anyone. Ask it to list its sources and anything it guessed. OpenAI's help page gives the same order for ChatGPT Work. Claude asks when something is unclear, and in Gemini you can type "interview me" in a task to be walked through it.

Question 3

How do I set up an AI agent that works on a schedule or in the background?

Accepted Answer

Three are within reach. In Gemini, click Task, describe the job and when to run it: daily, weekly, or whenever an email matching a Gmail filter arrives. Tasks need Google AI Pro, which is ₹1,950 a month in India. In Claude, any conversation on a paid plan can take a recurring job ("every Monday at 9 AM, summarise my team's Slack"), and it runs in the cloud with your computer off. ChatGPT Work can repeat a task on a schedule or trigger. Dots, the always-on agents OpenAI launched at DevDay on 29 September 2026, keep working between conversations on Pro, from $100 a month. Begin with a job that only reads and reports, like a morning brief, before you let one act.

Question 4

What can AI browsers like Comet, Claude in Chrome and Gemini in Chrome do on websites for me?

Accepted Answer

They read the page you are on and can act on it: click, type, fill in forms, move between pages and add things to a cart, using the logins you already have. Claude in Chrome does this on every paid Claude plan. Gemini in Chrome summarises and compares your tabs, while its auto browse, which books and fills things in for you, started with paid subscribers in the US. Comet is Perplexity's free browser with the assistant built in. The ChatGPT desktop app gives Work its own browser, and Copilot in Edge mostly reads and compares your open tabs. Expect them to pause at sign-ins and purchases and hand those steps to you. Some sites block them outright, and then a dot can fall back to your own browser.

Question 5

How do I check what an AI agent actually did before I trust its work?

Accepted Answer

Read the result, then the trail. Each agent keeps one: a dot shows ongoing and background work in Activity View, and you can open its own computer to watch; a Gemini task's work panel lists every step and the files it read or changed; Claude keeps tasks in Recents and past runs under Scheduled. Check the parts that would hurt if wrong: open the links it cites, recount totals, and confirm the date, amount and recipient before anything goes out. Ask it to say what it skipped, guessed or could not reach. Keep approvals on for anything that sends, buys or deletes until it has done that job well a few times. OpenAI's own advice is to review consequential work, and if checking takes longer than doing, do it yourself.

Question 6

Is it safe to let an AI agent use my logged-in email, bank or social media accounts?

Accepted Answer

It is the riskiest thing you can give an agent. A web page, email or comment can hide instructions, called prompt injection, that steer the agent into leaking data or acting for someone else; Brave showed this against Comet in 2025, and OpenAI and Anthropic both describe their defences as reducing the risk, not removing it. Connect accounts one at a time, use a separate browser profile for agent work as Anthropic suggests, and keep banking and password changes for yourself. Avoid Simon Willison's "lethal trifecta": private data, untrusted content and a way to send things out, all in one agent. And know who answers for the result: X's terms from 9 October 2026 make you responsible for autonomous actions taken on your behalf.

Question 7

What do AI agents cost, and why do I keep hitting usage limits?

Accepted Answer

Agents run on your existing subscription, but one task can use far more of it than a chat, because every step reads pages, runs tools or writes files. ChatGPT Plus is $20 a month and includes Work; dots need Pro, from $100, and talking to your dot does not count against limits while the Work or Codex tasks it starts do. Fast mode uses your allowance at 2.5 times the normal rate. Claude Pro is $20 a month and Max from $100, with a five-hour session limit and a weekly one shown in Settings, Usage. Google AI Pro is ₹1,950 a month and Ultra from ₹6,500. Long browsing and file-making tasks drain limits fastest, so batch requests and keep reference files in a project, where they count for less.

How founders use AI for Working with AI Agents

The questions