About This Video
A 25-minute video from Matthew Berman on the 12 ways he uses a personal AI assistant. His primary assistant is Grok bot. He also uses Muse and Dot, and he says every use case applies to whichever assistant you run.
Summary
Berman describes personal AI assistants as having taken AI mainstream: they help people get more done and skip the tedious parts of life. He shows twelve bots he has built, each defined by a plain-language prompt that tells the assistant to figure out what it needs first, only ask questions when something cannot be inferred, and follow firm rules about never acting without his okay. The bots also feed each other, so the daily brief can suggest a DoorDash order and the email bot can pull from his to-do list.
The 12 Use Cases
- Money saver. Searches email for subscriptions and negotiates them down. It found over $1,000 a month in fixed seats across his three Gmail accounts, drafted a note to Beehive about a $459 a month plan, and searched California's unclaimed property site, finding hundreds of dollars.
- Tesla bot. Connected through Tesla's official fleet API. "Get the car ready" sets the climate and seat heat, sets the destination from his calendar, and can open the garage door.
- Computer cleanup bot. Runs every Monday, finds caches and stale files, and lists what it would clear. One run flagged about 50 GB as safe to clear with 371 GB free, and nothing was deleted.
- Daily brief bot. Separate work and family versions that read calendar, email, and to-do list each morning and lead with what needs action.
- Family bot. Summarizes school and sports emails, posts them to a family Telegram group, auto-archives them, and adds events to the family calendar with his wife invited.
- Lead bot. Checks DMs on Facebook, LinkedIn, X, TikTok, and Instagram, sorts sponsor requests from interesting projects, archives spam, and drafts replies for approval.
- Meeting summary bot. Pulls the Fathom transcript after each meeting, sends a summary to Slack, extracts action items, and updates his to-do list.
- To-do manager. Works against Todoist and checks the list whenever it reads email, meetings, or the calendar.
- Email triage bot. Scans three accounts, auto-archives low-risk mail with a receipt, walks him through the rest, and drafts replies.
- Utility bill bot. Reviewed 12 months of PG&E usage, compared plans, and found one that saves $1,000 per year. It made the switch after he approved.
- Calendar assistant. Syncs work, personal, and family calendars, flags conflicts, and proposes times when someone asks for a meeting.
- Uber and DoorDash bots. Order a ride or reorder a meal on request, and other bots can call them.
The Pattern Behind Every Prompt
- Infer first. Each prompt opens by telling the bot to work out what it needs from connected accounts, and to ask only when something truly cannot be inferred.
- Clear job, clear rules. A short job description is followed by rules such as never send, delete, buy, switch, or book anything without his okay.
- Stay quiet. Several prompts tell the bot to say nothing when there is nothing new.
- Receipts. The email, to-do, and calendar bots report what they changed in a short summary.
- Risk levels. The cleanup bot sorts deletions by risk, and the email bot never auto-archives contracts, legal mail, or anything needing a signature.
Key Takeaways
- The biggest wins come from connecting the assistant to the accounts where your obligations already live: email, calendars, and a to-do list.
- Write approval rules into every prompt so the assistant drafts and recommends while you decide.
- Bots become more useful when they can call each other and share context.
- Dollar savings are the clearest proof: the utility plan and subscription negotiations each tied to specific amounts.
Reflect & Discuss
- Which of these twelve would save you the most time or money this month, and which accounts would it need access to?
- Where would you draw the line between "draft for my approval" and "just do it"?
- What risks come with giving an assistant access to email, calendars, a car, and utility accounts, and how do the rules in these prompts address them?