How to Build Your First Personal AI Agent

01 · What to automate first

What should I automate first?

Automate the routine you already do the same way every week. Rule-based, repeatable, low-ambiguity tasks make the best first builds, because you can tell inside a single run whether the agent got it right or not. The vacation, the dinner with friends, the final sign-off on the house, that part is yours. The logistical work around it is what you automate.

Two questions, about a minute. Every answer routes to a specific build.

Question 1 of 2

What eats the most repetitive time in your personal week right now?

Triaging personal email, bills, and newsletters
Organizing flight confirmations, hotel bookings, and travel details
Chasing contractor updates, permits, and invoices on a home project
Logging food delivery receipts, reservations, and loyalty balances
Wrangling RSVPs and logistics for a group dinner or a weekend trip
None of these. My personal admin is light, or someone else handles it

01 → 02 → 03

Your starting build

Answer the two questions and the one agent to build first appears here, with a direct link to its recipe. Answer both for the sharpest match.

If you would rather just read the routing:

What eats your personal time	Start here	Why
Triaging personal email, bills, and newsletters	Personal Chief of Staff	It drafts your replies, flags the bills, and briefs you twice a day. You still hit send.
Managing flight upgrades, hotel statuses, and travel confirmations	Travel Concierge	It builds one itinerary and flags every perk your status earns.
Tracking contractor bids, timelines, and invoices	Home Project Overseer	It logs the paperwork and flags any invoice that beats the original bid.
Logging delivery receipts, reservations, and loyalty balances	Dining & Delivery Manager	It keeps the running log so you stop guessing what you spent.
Organizing group dinners and weekend getaway logistics	Event Coordinator	It tracks who is coming and drafts the confirmations. You just host.
None of these, mostly light admin	Personal Chief of Staff	It drafts and briefs, never sends, so it is the lowest-risk agent to try. If a pile of receipts is more your edge, start with the Dining & Delivery Manager instead.

Take the builds with you

Want these five builds in your inbox?

Drop your email and I will send you all five agent builds, paste-ready, with your starter one in full. Set the first agent up tonight and keep the rest for when you need them.

One email with the builds. Reply anytime and it reaches me directly.

02 · The parts

What is a personal agent actually made of?

A Gumloop agent is an assembly of six kinds of parts, a Trigger, Connectors, Skills, Knowledge Sources, Subagents, and Abilities, rather than one large model that somehow already knows your life. Everything the agent knows, one of these six parts put there, and you can open any part and read it.

1

Trigger

The heartbeat that starts the agent running: a schedule, an event in a connected app, or a check you describe in plain language. For a Sunday meal plan it might be 8:00 AM every Sunday. For a receipt logger it might be a new email that looks like a receipt. Gumloop offers three kinds: Create With AI for custom polling Triggers you describe yourself, Event-Based Triggers built for a single service, and Scheduled Triggers that run on a set schedule or once at a set time.

2

Connector

An authorized link to an outside app the agent reads from or writes to. For personal agents this is usually Gmail, Google Calendar, Notion, Google Sheets, Google Drive, or Slack. Gumloop documents more than 150 integrations, including Gmail, Outlook, Google Calendar, Sheets, Docs and Drive, Notion, Airtable, Slack, Cal.com, and Zoom.

3

Skill

A reusable set of instructions, and optionally templates and scripts, that teaches an agent how to do one specific task, like pulling the total, date, and vendor from a receipt and formatting it as a row. You can build one three ways: Create With AI, Write Skill Instructions, or Upload Files.

4

Knowledge Source

Your personal brain: a preferences doc, your home warranties, appliance manuals, past itineraries, a loyalty-program list. Gumloop's Brain connects Notion, Google Drive, Slack, and more, plus uploaded files. Attach knowledge and the agent gains a Search Company Brain tool that returns relevant snippets with citations instead of guessing.

5

Subagent

A separate agent your agent calls to handle a distinct job, like your Chief of Staff calling a Travel Concierge to look up flight times. An agent can clone itself, which appears in selection lists as (Me), with a depth limit of one so clones cannot spawn clones. It can also invoke a different named agent, with no depth restriction.

6

Ability

A built-in capability the agent has without connecting anything: Web Search to look up current information, Web Fetch to read a page you named, Image Generation, a Code Sandbox that runs quick calculations, and Ask Question, which pauses and asks you for input through a structured choice card.

Walking the builder screen

Agent Preferences is where a build starts. Pick the model that powers the agent, then use the Add instructions for the agent box to write what it is, who it works for, and how it should behave. This box is the agent's job description, and most of what people expect from a Skill actually belongs here.

Beneath the instructions sits the Allow Self-Updates toggle, which lets the agent revise its own instructions. It is on by default. Leave it on and you can correct the agent in conversation, tell it to update its instructions so it always works that way, and watch the change land in the box you just wrote.

Triggers is where you decide what starts it. The panel carries an AI Managed setting, and the agent can create and manage its own Triggers during a conversation, so you can set a schedule by asking for it in plain language.

Connectors is the authorization step, shown with an AI Discovery setting. Add only what this agent needs. An agent holding every Connector you own is the signal that you should have split it into specialists.

Skills holds the reusable procedures, with an AI Skill Editing setting on the panel. Skills can be updated by the agent itself when you correct it, or edited by hand by anyone with agent editing access.

Knowledge Sources is where you attach the documents and workspaces the agent should read before it answers. Subagents is where you add the other agents it can call, including its own clone as (Me). Abilities holds the built-in switches, Web Search and Web Fetch among them.

Now see how the parts combine

03 · Build it or write it

When do I create a Skill, use a Subagent, or just write it into the instructions?

Write the logic into the instructions when it governs everything this one agent does and fits in a few lines. Make it a Skill, a reusable set of instructions that teaches an agent one specific task, when it is a standing procedure you would otherwise paste in as a long document, or when a second agent will need the same procedure. Hand it to a Subagent, a separate agent called to do a distinct job with its own tools, when the work is a different job with its own steps, its own connections, and its own back-and-forth.

Gumloop's documentation states the same test more bluntly:

"If you are about to paste a long SOP into the agent instructions, make a skill. If you are about to give one agent every connector in the company, split specialists and call them as subagents."
Gumloop documentation

Object	Build it when	What it is	Where it runs
Instructions	The rule applies always, to this agent only, and fits in a few lines	The text in the Add instructions for the agent box	Every turn, automatically
Skill	The procedure is long, repeatable, or shared across agents	A reusable set of instructions, plus optional templates and scripts	Inside the current conversation, when relevant
Subagent	The work is a separate job with its own tools and its own reasoning	A different agent, or a clone of this one shown as (Me)	In its own conversation, then reports back
App Rules	An action must be impossible, not merely discouraged	An enforcement policy, set by agent editors or org admins	At the tool call, before it executes

App Rules are the fourth object, and the one people skip. Skills suggest, App Rules enforce. Gumloop's own comparison page puts it this way:

"A skill can ask the agent not to delete Slack messages. An app rule can stop the delete call."
Gumloop comparison page

Worked example: Personal Chief of Staff, instructions

The Chief of Staff drafts replies to routine scheduling and briefs you twice a day. "Draft, never send. Match my voice from the Personal Preferences doc. Stop and ask before anything about money, health, or a direct request from family." That is a few lines, it applies to every draft the agent produces, and it belongs in the instructions box. Building a whole Skill for it would add a file to maintain and change nothing about the output.

Worked example: Dining log, a Skill

The receipt rules go the other way. Every delivery service formats its receipt differently, and the rules for pulling the restaurant, the items, and the total out of each one, and where to file a loyalty reload, add up to a procedure, not a sentence. Writing all of that into the instructions box means pasting a document into a field meant for a job description. You will also want the same one the day you point a second agent at your statements. Make it a Skill.

Worked example: Travel Concierge, a Subagent

Travel research does not belong on the Chief of Staff agent. It needs the web, your calendar, and your loyalty-program doc, and bolting all of it on leaves you with one overloaded agent holding every Connector you own. Keep travel as its own agent with its own connections. The Chief of Staff calls it as a Subagent when a trip comes up, and it hands back the itinerary.

The Chief of Staff is also where App Rules earn their keep. It can reach your live inbox, so the rule that blocks Send has to be enforcement, not a politely worded instruction.

Jump straight to a build: Personal Chief of Staff, Travel Concierge, Home Project Overseer, Dining & Delivery Manager, or the Event Coordinator. Not sure which fits your week? Take the assessment.

See the five builds

The builds

Five agents to build first

These five are common, bounded, low-risk jobs, and each one takes a different corner of your personal admin: your inbox, your travel, your home project, your spending, and your gatherings. None of them lets the agent send to a person, book, pay, or delete on its own.

Each recipe uses the same fields so you can compare them before you pick one. The instructions block in each is written to paste straight into the "Add instructions for the agent" box in Agent Preferences, with your own details swapped in. Each recipe also names the Skills to build and why; the test is the one from the decision table: a reusable process becomes a Skill, a one-off detail stays in the instructions. New to the parts vocabulary? The anatomy section defines all six in a minute of reading. Build one, watch it run, then build the next from what you learned. The Personal Chief of Staff is the one to start with.

01/5

04 · Personal Chief of Staff · the master build

Personal Chief of Staff

Start here. It reviews your inbox and calendar twice a day, drafts replies to routine scheduling, flags the bills and the urgent notes, and writes you a short daily brief. It is the gatekeeper for anyone whose personal admin has quietly become a second job.

Drafts and briefs you. It never sends an email.

Trigger Scheduled, twice daily at 7:30am and 5:00pm

Connectors Gmail, Google Calendar, Google Docs

Expected build time 45 to 75 min

Connectors, and what each one does here

Gmail. Reads your inbox to find what needs you, and saves each reply as an unsent draft. It never sends. Gmail's documented capability includes managing drafts, which is the basis of this recipe.
Google Calendar. Reads tomorrow's schedule so the drafts propose times you are actually free and the brief knows your day.
Google Docs. Writes the daily brief into one document you open with your coffee.

Knowledge Source, and why

Attach a Personal Preferences Google Doc as a Knowledge Source: how you like to work, your household's basic schedule, and your communication style. A Knowledge Source is your own material attached to the agent so it can search and cite it instead of guessing. This is what makes the drafts sound like you and the brief know your life.

Skills to create, and why

Build two Skills. A Newsletter Summarizer Skill through Write Skill Instructions that takes a long subscription email, pulls the two or three lines that matter, and drops the rest. A Calendar Conflict Checker Skill that cross-references an incoming invitation against what is already on your calendar and your buffer times before it drafts a yes. A Skill is a reusable instruction pack an agent follows for a specific task.

Setup steps in order

Create a new agent and name it Chief of Staff.
In Agent Preferences, choose your model and paste the instructions block below into "Add instructions for the agent."
Add the Gmail, Google Calendar, and Google Docs Connectors.
Write your Personal Preferences doc and attach it as a Knowledge Source.
Create the Newsletter Summarizer and Calendar Conflict Checker Skills.
Run the agent manually once and read the brief and every draft.
Set the twice-daily Scheduled Trigger only after that first run reads clean.

Make it yours, then paste it in

When it runs

Always stop and ask about

Nothing you type here leaves your device. It only rewrites the block below.

Paste this into "Add instructions for the agent"

Paste this

You are my personal Chief of Staff. You review my inbox and calendar twice a day. You never send.

Every day at 7:30 AM and 5:00 PM, read my inbox and tomorrow's calendar. Identify the emails that actually need me: a bill due, an urgent note from family, a direct question from a friend. For routine scheduling requests, draft a short, warm reply using my real calendar availability, and save it as a draft. Follow my Personal Preferences doc for my tone and my household schedule.

Write one Google Doc titled "Daily Brief" plus the date. Put three things in it: my schedule for tomorrow, a two-line summary of each newsletter worth reading with the rest left out, and the three items that need me today.

Use only what is in my inbox, my calendar, and my Personal Preferences doc. Do not invent an appointment, a deadline, or a message. Do not use em dashes. Never send an email. Never delete or archive anything. If something involves money, health, or a direct request from family, stop and use Ask Question to show me the message and ask how I want to handle it.

Open Gumloop and paste this block in. Stuck partway through? Email hello@bnedai.com.

What to check before you turn it on

Run it manually once and read the brief and every draft word for word.
Confirm each reply landed as an unsent draft and nothing left your outbox.
Send yourself a test bill and a test request from "family" and confirm it stops and uses Ask Question.
Check the brief against your real calendar: the right events, nothing invented.

Time math (illustrative, swap in your own counts)

If you spend 20 minutes each morning and 15 each evening sorting your inbox and figuring out your day, that is about 35 minutes a day, close to 4 hours a week. If reading the brief and approving the drafts takes 10 minutes a day, the agent gives back roughly 2.5 hours a week in that example. Time your own inbox for a few days before deciding what this is worth.

You're done when

It saves an unsent draft instead of sending one.
The drafts sound like you and propose times you are actually free.
The daily brief matches your real calendar. Nothing is invented.
It stops and asks on anything about money, health, or family.
You have watched it run at least once and read every draft.

What could go wrong Make it safe to run

02/5

05 · Travel Concierge

Travel Concierge

This agent watches your inbox for booking confirmations, pulls them into one clean itinerary, and cross-references your elite statuses so you never miss a lounge, an upgrade, or a fast-track line. It is for the traveler piecing together confirmation emails across six threads the night before a trip.

Organizes your trip. It never books or cancels anything.

Trigger Event-Based, when a booking confirmation arrives

Connectors Gmail, Google Docs, Google Calendar

Expected build time 45 to 75 min

Connectors, and what each one does here

Gmail. Reads the confirmation emails as they land. It reads only; it never replies to an airline or hotel.
Google Docs. Writes one itinerary document per trip that you open on the way to the airport.
Google Calendar. Adds each flight, check-in, and reservation to your calendar as an event.

Knowledge Source, and why

Attach a Loyalty Programs doc as a Knowledge Source, listing each airline and hotel where you hold status and the tier. That is what lets the agent tell you a specific perk you have earned rather than a generic "check for upgrades." Keep membership numbers out of it if you would rather; the tier alone is enough to flag eligibility.

Skills to create, and why

Build an Itinerary Format Skill through Write Skill Instructions that fixes the shape of every trip doc: flights in order, hotel with address and check-in time, confirmation codes in one place. Build a second Perk Match Skill that states how to compare a booking against your status tiers and only flag a perk the tier actually earns.

Setup steps in order

Create the agent and paste the instructions block below into "Add instructions for the agent."
Add the Gmail, Google Docs, and Google Calendar Connectors.
Write your Loyalty Programs doc and attach it as a Knowledge Source.
Create the Itinerary Format and Perk Match Skills.
Run it manually on one real confirmation email and check every detail against the source.
Set the Event-Based Trigger to watch for confirmations once that run is clean.

Make it yours, then paste it in

Title each trip doc with

Perks to flag

Nothing you type here leaves your device. It only rewrites the block below.

Paste this into "Add instructions for the agent"

Paste this

You organize my travel. You read and summarize only. You never book, change, or cancel anything.

When a booking confirmation arrives in my inbox (flight, hotel, or car), read it and pull the confirmation code, dates, times, flight numbers, and addresses. Keep one Google Doc per trip titled with the destination and the dates, following the Itinerary Format Skill, and add each leg to my calendar as an event.

Cross-reference the airline and hotel against my Loyalty Programs doc. Flag where I am eligible for an upgrade, lounge access, or an expedited security line, and name the status that qualifies me. Use only the facts in the confirmation and the Loyalty Programs doc. If a detail is missing, write "not in the confirmation" rather than guessing. Do not use em dashes.

Never reply to the airline or hotel. Never book, change, or cancel a reservation. Never enter a payment detail. If a confirmation looks like a change or a cancellation, stop and use Ask Question to show me the email.

Open Gumloop and paste this block in. Stuck partway through? Email hello@bnedai.com.

What to check before you turn it on

Run it on one real confirmation and check every code, time, and address against the source email.
Confirm it wrote "not in the confirmation" where the email was missing a detail, instead of guessing.
Confirm each flagged perk actually matches a status in your Loyalty Programs doc.
Confirm it replied to no one and changed no reservation.

Time math (illustrative, swap in your own counts)

If you take 6 trips a year and building each itinerary from scattered confirmations takes 40 minutes, that is 240 minutes a year. If reviewing the drafted itinerary takes 10 minutes a trip, the agent gives back about 3 hours a year in that example, and you stop missing perks you already paid for. Count your own trips before trusting that shape.

You're done when

Every code, time, and address in the itinerary traces back to a confirmation email.
It wrote "not in the confirmation" instead of guessing a missing detail.
Each flagged perk matches a real status in your Loyalty Programs doc.
It replied to no one and changed no reservation.
You have watched it run on at least one real trip.

What could go wrong Make it safe to run

03/5

06 · Home Project Overseer

Home Project Overseer

This agent handles the paperwork of a renovation or repair. It reads incoming contractor emails, logs timelines, permit statuses, and invoice amounts into one sheet, and flags any invoice that comes in over the original bid. It is for anyone trying to keep a contractor's emails, dates, and numbers straight.

Tracks the paperwork. It never pays or approves anything.

Trigger Event-Based, on a new email from a contractor

Connectors Gmail, Google Sheets, Google Drive

Expected build time 45 to 75 min

Connectors, and what each one does here

Gmail. Reads incoming emails from your contractors. It reads only; it never replies.
Google Sheets. Logs every timeline update, permit status, and invoice amount into one tracker row by row.
Google Drive. Holds the original bids and contracts so the agent can check an invoice against what was agreed.

Knowledge Source, and why

Point the agent at your Home Renovation Drive folder as a Knowledge Source: the signed bids, the contracts, the permit paperwork. An agent with knowledge attached gets a Search Company Brain tool that returns the exact bid line with a citation, which is what lets it catch an invoice that runs over.

Skills to create, and why

Build an Invoice Extract Skill through Write Skill Instructions that names exactly which fields to pull from a contractor email and how to log them: date, contractor, category, amount, note. Build a second Bid Match Skill that states how to find the matching line in the original bid and what counts as "over," so the flag is consistent and not a judgment call the agent makes differently each time.

Setup steps in order

Create the agent and paste the instructions block below into "Add instructions for the agent."
Add the Gmail, Google Sheets, and Google Drive Connectors.
Create the tracker sheet and put your signed bids in the Home Renovation folder.
Attach that folder as a Knowledge Source.
Create the Invoice Extract and Bid Match Skills.
Run it manually on one real contractor email and check every logged number against the source.
Set the Event-Based Trigger once that run is clean.

Make it yours, then paste it in

Tracker sheet name

Flag an invoice when

Nothing you type here leaves your device. It only rewrites the block below.

Paste this into "Add instructions for the agent"

Paste this

You manage the paperwork for my home project. You read, extract, and log only. You never pay or approve anything.

When an email arrives from one of my contractors, read it and pull any timeline update, permit status, and invoice amount. Log each into the Home Renovation Tracker Google Sheet: date, contractor, category, amount, and a one-line note. Search my Home Renovation Drive folder for the original bid before you log an invoice.

If an invoice comes in higher than the documented bid for that line, flag it in the sheet and use Ask Question to show me both numbers. Use only the amounts in the email and the bid in the folder. Do not estimate a number that is not written down. Do not use em dashes.

Never reply to a contractor. Never approve a change order. Never pay an invoice or enter a payment detail. At the end of each week, post one summary: what is on schedule, what slipped, and what is waiting on me.

Open Gumloop and paste this block in. Stuck partway through? Email hello@bnedai.com.

What to check before you turn it on

Run it on one real contractor email and check every logged number against the source.
Send a test invoice higher than a bid in your folder and confirm it flags it and uses Ask Question.
Confirm it logged only amounts that were written down, with nothing estimated.
Confirm it replied to no contractor and approved nothing.

Time math (illustrative, swap in your own counts)

If a renovation runs 12 weeks and you spend 30 minutes a week chasing emails and updating a spreadsheet by hand, that is 360 minutes over the project. If reviewing the agent's log takes 5 minutes a week, the agent gives back about 5 hours across the job in that example, and one caught overbilling can pay for far more than that. Use your own project.

You're done when

Every logged number traces back to a contractor email.
It flags an invoice that runs over the bid and uses Ask Question.
It logged only amounts that were written down. Nothing is estimated.
It replied to no contractor and approved nothing.
You have watched it run on at least one real email.

What could go wrong Make it safe to run

04/5

07 · Dining & Delivery Manager

Dining & Delivery Manager

This agent keeps a running log of your takeout, reservations, and loyalty balances by reading the receipts that already land in your inbox. It is for the person who has no idea what they spent on delivery last month and would like to.

Logs what you already spent. It never orders or books.

Trigger Event-Based, when a receipt or reservation arrives

Connectors Gmail, Notion or Google Sheets

Expected build time 30 to 60 min

Connectors, and what each one does here

Gmail. Reads delivery receipts and reservation confirmations as they arrive. It reads only.
Notion or Google Sheets. Holds your dining log. Pick whichever you already live in; both are verified Connectors.

Skills to create, and why

Build a Receipt Parse Skill through Write Skill Instructions that names how to read each service you use (UberEats, Grubhub, DoorDash format their receipts differently) and which fields to record. This is the reusable procedure that keeps every row consistent no matter which app sent the receipt. It is the clearest case on this page of something that belongs in a Skill rather than the instructions box.

Setup steps in order

Create the agent and paste the instructions block below into "Add instructions for the agent."
Add the Gmail Connector, then Notion or Google Sheets.
Create your dining log with the columns you want to track.
Create the Receipt Parse Skill covering each service you order from.
Run it manually on a handful of real receipts and check every total against the source.
Set the Event-Based Trigger once the totals match.

Make it yours, then paste it in

Log it in

Pick one

What each row should record

Nothing you type here leaves your device. It only rewrites the block below.

Paste this into "Add instructions for the agent"

Paste this

You keep my dining log. You read and record only. You never place an order or make a reservation.

When a food delivery receipt (UberEats, Grubhub, DoorDash) or a restaurant reservation confirmation arrives, read it and pull the restaurant, the date, the items, and the total. Record each into my dining log: restaurant, date, order, total, and reservation time if there is one. If a loyalty reload or points receipt arrives, update the balance in the same log.

Use only the numbers on the receipt. Do not estimate a tip or a total that is not printed. Do not use em dashes. Never order food. Never book, change, or cancel a reservation. Never enter a payment detail. At the end of the month, give me one summary: total spend, most-ordered spot, and any reservation coming up.

Open Gumloop and paste this block in. Stuck partway through? Email hello@bnedai.com.

What to check before you turn it on

Run it on a handful of real receipts and check every total against the source email.
Confirm it recorded only printed numbers, with no estimated tip or total.
Confirm it placed no order and made no reservation.
Read one loyalty-reload row and confirm the balance is right.

Time math (illustrative, swap in your own counts)

The other builds save you minutes. This one hands you a number you did not have before. If you order delivery 20 times a month, the log turns a vague sense of overspending into one figure you can act on, for the cost of the receipts already sitting in your inbox. Run it a month, then decide whether to keep it.

You're done when

Every total in the log matches the receipt it came from.
It recorded only printed numbers. No tip or total is estimated.
It placed no order and made no reservation.
Loyalty balances update correctly from a reload receipt.
You have watched it run on several real receipts.

What could go wrong Make it safe to run

05/5

08 · Event Coordinator

Event Coordinator

This agent runs the logistics of one gathering, a birthday dinner or a weekend getaway. It reads incoming RSVPs, logs who is coming and any dietary notes, and drafts each confirmation reply so you just review and send. It is for the host doing RSVP math in their head.

Wrangles the RSVPs. It never sends without your yes.

Trigger Create With AI, watching for RSVP replies

Connectors Gmail, Google Sheets

Expected build time 45 to 75 min

Connectors, and what each one does here

Gmail. Reads the RSVP replies and saves each confirmation as an unsent draft. It never sends.
Google Sheets. Holds the master guest list: name, attendance, plus-ones, dietary notes.

Skills to create, and why

Build an RSVP Log Skill through Write Skill Instructions fixing how a reply maps to a row: how to read "count me in, plus my partner," where the dietary note goes, how to mark a maybe. Build a second Confirmation Draft Skill holding the shape and tone of the reply so every guest gets the same warm, complete note.

Setup steps in order

Create the agent and paste the instructions block below into "Add instructions for the agent."
Add the Gmail and Google Sheets Connectors.
Create the master guest list sheet with your columns.
Create the RSVP Log and Confirmation Draft Skills, with your venue and parking details in the instructions.
Run it manually on a few real RSVP replies and read every drafted confirmation.
Set the Trigger to watch for replies once the drafts read right.

Make it yours, then paste it in

Guest list sheet name

Confirmation should include

Nothing you type here leaves your device. It only rewrites the block below.

Paste this into "Add instructions for the agent"

Paste this

You help me run one event. You read RSVPs and draft confirmations. You never send.

When an RSVP reply arrives for my event, read it and pull the guest's name, whether they are coming, their plus-ones, and any dietary note. Log each into the master guest list Google Sheet. Draft a short confirmation reply that thanks them and gives the venue address and parking details, and save it as a draft for me to review.

Use only what the guest wrote and the details in this instruction. Do not invent a dietary restriction or a plus-one. Do not use em dashes. Never send the confirmation yourself. Never remove a guest from the list. If an RSVP is unclear about who is coming, use Ask Question and show me the message.

Open Gumloop and paste this block in. Stuck partway through? Email hello@bnedai.com.

What to check before you turn it on

Run it on a few real RSVP replies and read every drafted confirmation.
Confirm each row matches what the guest actually wrote, with no invented plus-one or dietary note.
Confirm the confirmations saved as drafts and nothing sent.
Send an ambiguous RSVP and confirm it uses Ask Question instead of guessing.

Time math (illustrative, swap in your own counts)

If you host one gathering with 30 invitees, and logging each RSVP and writing each confirmation by hand takes 4 minutes, that is 120 minutes over the run-up. If reviewing a drafted confirmation takes 1 minute, 30 drafts cost you 30 minutes. The agent gives back about 90 minutes on that event, and no RSVP slips through a crowded inbox. Count your own guest list.

You're done when

It saves each confirmation as an unsent draft instead of sending one.
Each row matches what the guest wrote. No plus-one or dietary note is invented.
The confirmation carries the right venue and parking details.
It stops and asks on an ambiguous RSVP instead of guessing.
You have watched it run on a few real replies.

What could go wrong Make it safe to run

09 · App Rules & setup

What should the agent never be allowed to do on its own?

An agent needs your yes before anything it cannot take back: sending an email, deleting a message, changing a calendar event, paying an invoice, or removing someone from a list. Everything short of that line can run unattended once you have watched it work a few times. Automate the routine, and keep the irreversible on your side of the line.

First, authorize the connectors

Connecting your personal Gmail and Google Calendar to Gumloop runs through a direct Google sign-in. You never generate an API key, configure a Google Cloud project, or manage scopes by hand. Gumloop uses the standard OAuth flow to hold the connection.

Open the Connectors page from your Gumloop dashboard.
Add the Gmail connector. You are sent to Google's own sign-in screen. Pick your account and grant the permissions. Gumloop stores the credential.
Add the Google Calendar connector the same way.

Then lock it down with App Rules

Once connected, the agent can technically send an email or delete an event. Writing "never send without asking" in the instructions is not enough. Instructions are requests, and a model can ignore a request. App Rules are enforcement. They block the call at the connector, before it runs, even if the agent tries.

On your Chief of Staff

Set an App Rule on the Gmail connector that blocks Send Email, Delete Email, and Trash Email. That leaves the agent with Create Draft and nothing else, so it can read your inbox and write replies but never send or erase one.

On your calendar

Set an App Rule on the Google Calendar connector that blocks Delete Event. The agent can read your availability and propose new events, and it can never erase one that is already there.

Pair the Google sign-in with those two App Rules and the agent can read and draft on your behalf, and it can never permanently change your inbox or your calendar without you clicking the last button.

The four rules that cover the rest

Rule 1. Draft, never send.

Write it into the agent's instructions in plain words: produce a draft, put it where I will see it, stop there. Then make the App Rule match by blocking Send at the connector. An instruction is a default; the App Rule in rule 3 is the lock.

Rule 2. When the rule runs out, ask.

Turn on the Ask Question Ability, one of the agent's built-in Abilities. It pauses the run and hands you a structured choice instead of letting the agent guess. Use it at every fork you did not write a rule for, like an invoice that beats a bid or an RSVP that is unclear about who is coming.

Rule 3. Hard-block what must never happen.

App Rules are enforcement policies rather than guidance, and they stop the call the agent tries to make. Deleting an email, erasing a calendar event, entering a payment detail: if you would not want to undo it later, it belongs in an App Rule and not in a paragraph of instructions.

Rule 4. Approve tools one at a time.

Gumloop documents four approval mechanisms: App-Level Settings, Per-Tool Custom Controls, App Rules, and the Ask Question Ability. Connect only the tools this agent's job requires, then set approval on the individual tools that send, delete, or spend, rather than approving an entire app in one click.

The difference between a Skill and an App Rule is the one that trips people up. A Skill is a reusable set of instructions that teaches an agent how to do a specific task, and what it produces is a suggestion the agent can override. An App Rule cannot be overridden. Write a careful Skill, feel covered, and you have installed a preference rather than a control.

See what usually goes wrong

10 · When it breaks

What usually goes wrong on a first build?

The failure that shows up first on almost every build is an agent that produces something technically correct in a voice that is not yours, because it was never given a real example of your writing to work from. Scan the symptoms below for the one you are seeing; the cause and the fix sit underneath each.

"It doesn't sound like me. It reads like a form letter."

Cause: the agent was given a description of your voice instead of examples of it. Fix: build a voice Skill through the Upload Files path using three to five messages you actually sent, not messages written as samples. If a correction you make in chat produces the right voice, tell the agent to turn that into a Skill so the fix persists.

Applies toChief of Staff Event Coordinator

"It sent something instead of drafting it."

Cause: the instructions described the intent without stating the prohibition, so nothing stopped the send. Fix: write the restriction as its own sentence in "Add instructions for the agent," in the form "Never send." Then back it up in the product, not in prose. Set an App Rule on the Gmail connector that blocks the Send action outright. A Skill can only ask; an App Rule stops the call.

Applies toChief of Staff Event Coordinator

"It made something up that isn't true."

Cause: it was asked for a detail it had no source for and produced a plausible answer instead of an admission. Fix: attach real material as a Knowledge Source, since an agent with knowledge attached searches it and answers with citations. Then instruct it to write "not verified" or "not in the confirmation" when it cannot confirm something, and check the output for that phrase. A first run with no gaps anywhere usually means the gaps were filled rather than absent.

Applies toTravel Concierge Home Project Overseer Dining Manager

"Nothing happens when it's supposed to run."

Cause: the Trigger condition is wrong for the job, or a connected account's authorization expired. Fix: open the Triggers panel first and confirm you picked the right one of the three types, since a Scheduled Trigger will never fire on a new email and an Event-Based Trigger will never fire at 7:30am on its own. If the Trigger is right, re-authorize the Connector.

Applies toAll five

"I set it up and now I'm afraid to touch it."

Cause: it went straight to a live schedule without a watched test period, so you have no evidence about how it behaves. Fix: pause the Trigger, run it manually three to five times on real inputs, and read every output completely.

Applies toAll five

"It sort of works, but I can't tell if it's right."

Cause: the first build was scoped too large to verify in one sitting, so no single run tells you whether it worked. Fix: shrink the job until one manual run fits in an afternoon and you can read every output it produced. One task, one output. Scale comes from a second agent, not a bigger first one.

Applies toAll five

"It's logging the same thing twice, or pulling old information."

Cause: the lookback window is too wide, or nothing tells the agent what it has already handled. Fix: put an explicit date boundary in the instructions, in the form "only receipts from the last seven days," and write a duplicate rule naming the field that identifies a record as the same one. For travel that field is usually the confirmation code; for dining it is the order number.

Applies toTravel Concierge Home Project Overseer Dining Manager

Something else? Check the FAQ

11 · Questions

Frequently asked questions

How do I start using AI in my personal life?

Pick one repeatable task you already do the same way every week, like triaging your inbox or organizing a trip, and build one agent for that task. Give it a trigger that starts it, connections to the apps involved, and a written job description. Watch it run on real work before you let it run on a schedule. You keep the judgment calls; the routine around them is what the agent takes.

What should I automate first at home?

Automate the rule-based, repeatable admin you already do the same way each time, because you can tell inside a single run whether the agent got it right. Common first builds: a personal Chief of Staff that triages your inbox and drafts replies, a travel concierge that builds your itinerary, a home project overseer that tracks contractor invoices, a dining manager that logs receipts, and an event coordinator that wrangles RSVPs.

What is a personal AI agent made of?

Six kinds of parts: a Trigger that starts it, like 7:30am daily to check your inbox, Connectors that link it to your apps like Gmail, Google Calendar, Notion, and Google Sheets, Skills that teach it your procedures and voice, Knowledge Sources it searches like a personal preferences doc or a home renovation folder, Subagents it can hand distinct jobs to, and Abilities like Web Search and Ask Question. It is an assembly of parts you can open and read, not a model that already knows your life.

When should I build a Skill instead of just writing instructions?

Build a Skill when the procedure is long enough that you would otherwise paste a full document into the instructions box, or when a second agent will need the same procedure, like your writing voice or your receipt-parsing rules. Keep it in the instructions when the rule governs everything this one agent does and fits in a few lines. Gumloop's own test: if you are about to paste a long procedure into the instructions, make a Skill.

Will an AI agent email people without me?

Not if you build it the way this guide shows. Every build here that touches a person drafts and stops. The agent saves an unsent draft, you read it, you edit it, and you hit send. Put the prohibition in the instructions in plain words, then back it with an App Rule that blocks the Send action at the connector. A Skill only asks; an App Rule stops the call.

Is it safe to connect my personal Gmail and Calendar to an AI agent?

Yes, when you lock it down. Gumloop connects to Google through standard OAuth, so you never hand over a password or generate an API key. Then you set App Rules that physically block the actions you never want: block Send, Delete, and Trash on Gmail so the agent can only create drafts, and block Delete Event on Calendar so it can only read your availability. Instructions are requests; App Rules are enforcement. See App Rules & setup.

Is Gumloop free?

No. Gumloop runs on a 14-day trial, after which it moves to a paid plan. There is no permanent free tier. Code 1STAGENT takes 20% off your first month.

Why do personal automations stall after a few weeks?

Usually because the first build was scoped too large to verify in one sitting, so you were never sure it was right and quietly stopped trusting it. The fix is one task, one output, one run you can read start to finish. Build the smallest useful version, watch it work, then build the next from what you learned.

Next step: pick your first build

12 · Next steps

Where do I go from here?

Pick one of the five recipes and build it this week. The Personal Chief of Staff is the one to start with, and everything you learn building it carries to the other four.

Start building on Gumloop Code 1STAGENT takes 20% off your first month.

Free live session

Retake the assessment

Mastering AI in Your Personal Life

What should I automate first?

Want these five builds in your inbox?

What is a personal agent actually made of?

Trigger

Connector

Skill

Knowledge Source

Subagent

Ability

Walking the builder screen

When do I create a Skill, use a Subagent, or just write it into the instructions?

Worked example: Personal Chief of Staff, instructions

Worked example: Dining log, a Skill

Worked example: Travel Concierge, a Subagent

Five agents to build first

Personal Chief of Staff

Travel Concierge

Home Project Overseer

Dining & Delivery Manager

Event Coordinator

What should the agent never be allowed to do on its own?

First, authorize the connectors

Then lock it down with App Rules

On your Chief of Staff

On your calendar

The four rules that cover the rest

Rule 1. Draft, never send.

Rule 2. When the rule runs out, ask.

Rule 3. Hard-block what must never happen.

Rule 4. Approve tools one at a time.

What usually goes wrong on a first build?

Frequently asked questions

Where do I go from here?

Agent Lab

AI Readiness Index

BNEDai Insights