His wife planned a Halloween pumpkin craft with her OpenClaw agent, which she calls Betty. Betty made the templates. Then his own agent, running on OpenAI's Codex, sent the files to their Bambu Lab 3D printer, and the kids loved the result. It's a nice picture of two family agents splitting a job. You don't need a 3D printer for this: the same request can produce printable paper stencils for carving.
PromptPlan a [HOLIDAY] craft for kids aged [AGES] that takes under an hour. Make printable templates sized for...
He was looking for an animation for a small OLED screen that he'd once saved on X but only half remembered. He started scrolling his bookmarks by hand and asked his OpenClaw agent at the same time. The agent found it before he got a quarter of the way through. Searching your own saved stuff is a small but real time saver if your agent can read your bookmarks.
PromptSearch my saved [X bookmarks / browser bookmarks / notes] for a post about [WHAT I REMEMBER]. I think I saved...
On the Make Community forum, someone wanted a scenario to run whenever they checked off a task in Todoist. Their webhook fired when a task moved between sections, but nothing happened when a task was marked complete. A forum helper explained why: in Todoist, finishing a task is its own event, called item:completed, separate from item:updated. So you add that event to the webhook and give it its own route with a filter in the scenario. Recurring tasks are the trap. Checking one off just moves it to the next date, so Todoist sends an update event instead, marked with a completed intent. The original poster hadn't confirmed the fix yet, but Todoist's own documentation describes the events the same way. The general lesson for Make: when a trigger doesn't fire, check which exact events the app sends, because "changed" and "finished" are often separate.
PromptHelp me set up a Make scenario that runs when I check off a task in Todoist. Tell me which Todoist webhook...
He runs a website of profiles of well-known Indian figures, and keeping each profile current took him about an hour a day: descriptions, photos, birth dates, awards and timelines. He set up Hermes Agent with the API of Directus, the content system behind the site, so the agent now does those updates. It even checks each photo with image recognition to make sure it's a portrait of just one person. He says he spent about $5 experimenting with a paid OpenAI model and is moving to cheaper open models through OpenRouter. One thing didn't work: Hermes couldn't sign in to Google in its browser, even with the login saved in its vault. That's a common limit, since Google often blocks automated sign-ins, so an official API is usually the steadier way in. For a small site owner, the takeaway is that repeat edits through an API are a good fit for an agent, while anything behind a Google login may need a different route.
PromptEvery day, check the profiles in my [CMS] for missing or out-of-date fields: description, photo, birth date,...
He asked his Hermes Agent to find the official mobile beta of Hermes. Search results were full of unofficial apps using the Hermes name. The agent skipped them and found the real sign-up on Nous Research's own site. The Android test group was already full, while the iPhone one was still open. With popular open-source agents, lookalike apps are common, and some ask for your API keys or passwords. Getting the link from the maker's own site is the safe habit.
PromptFind the official app or download for [TOOL]. Only use links from the maker's own website or their verified...
He texted his Grok Bot, which he calls his chief of staff, just "cancel it" about his Xbox Game Pass. The bot signed in to his account, cancelled the plan and sent back the confirmation, so $9.99 a month stops from the next billing date. It also remembered that the family likes the subscription for game nights, so it noted that turning it back on later would take one more text. The useful part is the short loop: the bot already knew which subscription he meant from earlier chats, did the clicks itself and returned proof. If you try this, ask for the confirmation number or email every time, and check your next card statement, because a cancellation that didn't go through is easy to miss. Also ask whether you keep access until the end of the paid month, which many subscriptions allow.
PromptCancel my [SUBSCRIPTION] plan. Sign in with my saved login, find the cancel option and tell me the date the...
He shared what his Grok Bot did in one afternoon of household admin. It compared his car insurance against quotes from GEICO, Progressive and Plymouth Rock and found he already had the best rate, so nothing changed. It filed two class action settlement claims he qualified for, found an open recall on his car and booked the dealer visit together with a brake appointment, kept a log of spam callers so he can report them to the Do Not Call registry, and flagged bills before they were due. He also promotes a copy of his landlord bot, so treat him as someone who sells bot setups. Still, the list is a fair picture of what these agents are good at: lots of small forms and lookups that nobody gets around to. Settlement claims are a common case. You usually need proof you bought the product, and a claim made without it can count as a false statement under penalty of perjury.
PromptHelp me with household admin this week. 1) Check open class action settlements I may qualify for, based on...
She tried Dots, the always-on agent in ChatGPT Pro, and found it less proactive than she hoped. It didn't use the background she had fed it, and when a request was blocked by a missing app connection, it stayed blocked after she added the connection until she asked again. She says Meta Muse is better at following up on its own. If you use Dots, a simple fix is to tell it what to do when it gets unblocked.
PromptIf a task is blocked because you need access to an app or a file, tell me exactly what you need. When I give...