--- title: "Notion Custom Agents: Automate Your Weekly Project Report" description: "Build a Notion Custom Agent that writes a dated weekly project report from your task database, then test it against an empty week and a repeat run." canonical: "https://adamtheautomator.com/notion-custom-agents-weekly-report/" --- # Notion Custom Agents: Automate Your Weekly Project Report > Build a Notion Custom Agent that writes a dated weekly project report from your task database, then test it against an empty week and a repeat run. Source: https://adamtheautomator.com/notion-custom-agents-weekly-report/ --- ATA Learning Tap to hide [ ATA Learning ](/) * [Home](/) * [Tutorials](/tutorials/) * [Instructors](/author/) * [Advertising](/advertising/) * [Recommended Resources](/resources/) * [About Adam](/about-adam/) Search for: * [](https://twitter.com/adbertram) * [](https://github.com/Adam-the-Automator) * [](https://www.linkedin.com/company/adam-the-automator-llc) * [](/feed/) ![Notion Custom Agents: Automate Your Weekly Project Report](https://adamtheautomator.com/wp-content/uploads/publisher/3f25d9c85b2b81c4811ed7ed9b290871/d4ebe25fb13196c5dffbe9c4668d57fe178690406552541f4e2f3828e040cc6f.webp) # Notion Custom Agents: Automate Your Weekly Project Report [![](https://secure.gravatar.com/avatar/d0b9d42e21e5622713f8b693aa5c0f9244d5f7dd200ed29b8398f52dee5de337?s=192&d=mm&r=g)Adam Bertram](https://adamtheautomator.com/author/adam-bertram/)26 October 202614 min. read Categories: [AI Agents](/category/ai-agents/) Tags:[Notion](/tag/notion/)[AI](/tag/ai/)[Automation](/tag/automation/) Table of Contents * [What a Custom Agent Does That a Database Automation Cannot](#what-a-custom-agent-does-that-a-database-automation-cannot) * [Prerequisites](#prerequisites) * [Build Two Databases Before You Write a Word of Instructions](#build-two-databases-before-you-write-a-word-of-instructions) * [The One Schema Decision That Matters](#the-one-schema-decision-that-matters) * [Nine Seed Rows and a Hand-Written Answer Key](#nine-seed-rows-and-a-hand-written-answer-key) * [Two Rows That Catch a Lazy Implementation](#two-rows-that-catch-a-lazy-implementation) * [Write Instructions the Agent Can Repeat Every Monday](#write-instructions-the-agent-can-repeat-every-monday) * [The At-Risk Rule Needs a Definition](#the-at-risk-rule-needs-a-definition) * [Access: Two Databases, One Read-Only](#access-two-databases-one-read-only) * [One Trigger, One Timezone](#one-trigger-one-timezone) * [The Window Is Absolute Dates, Not “Last Seven Days”](#the-window-is-absolute-dates-not-last-seven-days) * [Four Tests Before the Trigger Goes Live](#four-tests-before-the-trigger-goes-live) * [Why the Empty Week Still Gets a Page](#why-the-empty-week-still-gets-a-page) * [Know What One Weekly Run Costs](#know-what-one-weekly-run-costs) * [Six Failure Modes and Their Fixes](#six-failure-modes-and-their-fixes) * [What the First Notion Weekly Report Must Prove](#what-the-first-notion-weekly-report-must-prove) Do not put a Notion Custom Agent on a schedule until your project database has a second date property for the day a task finished. With only a due date, the report cannot separate the work you shipped from the work you promised, so it takes the due date and hands you a confident summary of your own plans. Every rule in the instruction block names a property, so the properties have to exist before the instructions do, and you can’t grade a run until you’ve written down the counts you expect from it. You finish with one dated page per week that states its own reporting window and says zero out loud when nothing happened. ## What a Custom Agent Does That a Database Automation Cannot Every Monday someone on your team opens the project database, reads the rows, and writes the update from memory, producing a different level of detail depending on who is on vacation. Notion’s [Custom Agents overview](https://www.notion.com/help/custom-agents) describes the replacement in one line: “Custom Agents live inside Notion and run on your instructions.” What separates a Custom Agent from the [Notion Agent](https://www.notion.com/help/notion-agent) in your page is autonomy, since “Custom Agents are designed to run automatically in the background based on triggers and schedules.” The permission model decides whether you can trust the output. Notion AI acts as you and inherits your access. A Custom Agent acts as itself, with its own grants and no inherited permissions from you or from whoever triggers a run. That design lets an agent report on a database some readers cannot open, and it means a careless scope exposes everything the agent can see: [Custom Agents sharing and permissions](https://www.notion.com/help/custom-agents-sharing-and-permissions) states that agent users “can retrieve information from resources they lack access to.” A [database automation](https://www.notion.com/help/database-automations) fires on an event and performs a fixed action. A weekly report needs more judgment: comparing completion against due dates, noticing that a blocked task is also four days late, and writing the finding into a paragraph. The page has to be composed, which is why the data underneath it has to be rigid. When identical output matters more than judgment, a Worker or a database automation is the right pick, since [the three tiers are not interchangeable](https://adamtheautomator.com/notion-automation-agents-vs-workers/). ## Prerequisites Confirm all of this before you open the **Agents** section, because most of it fails later without an error message. Custom Agents need a Business or Enterprise plan ([Notion pricing](https://www.notion.com/pricing)), and a downgrade to Free or Plus switches every existing agent off without deleting it ([downgrading your plan](https://www.notion.com/help/plan-downgrade)). Business includes 1,500 Standard plus 500 Select credits per member per month, and an agent spends its owner’s Select credits by default, even when teammates run it. Everyone can create agents by default, and Enterprise admins narrow that under **Settings** > **Notion AI** > **Agents** > **Control who can create agents** ([sharing and permissions](https://www.notion.com/help/custom-agents-sharing-and-permissions)). That page also carries the ownership rule: after 7 days with no owner, the agent stops running. Plan on a desktop or web session, because the mobile apps can’t build or edit an agent ([Notion’s Custom Agents help](https://www.notion.com/help/custom-agents)). You need edit access to the source database to add properties, plus a separate destination database, because this build grants the agent **Can view** on the tracker and **Can edit** on the reports database. An agent with **Can edit** on its source can write back to it. Notion’s help pages don’t publish a build number for Custom Agents, so check the panel names below against what your workspace shows. ## Build Two Databases Before You Write a Word of Instructions One tasks database beats a Projects database joined to a Tasks database. Weekly questions are task-level (what closed, what is late, what is stuck), and every extra source is another scope to maintain and another thing the agent reads on each run. ### The One Schema Decision That Matters `Completed` has to be its own Date property, separate from `Due`. Remove it and “completed this week” becomes unanswerable, so the agent substitutes the due date and reports promises as deliveries with no error anywhere. The other properties exist so the agent has something to count or sort. | Property | Type | Why the agent needs it | | --- | --- | --- | | `Task` | Title | The one-line name that appears in the report | | `Status` | Status | Options: `Not started`, `In progress`, `Blocked`, `Done` | | `Owner` | Person | Answers “who is stuck” and feeds the missing-owner count | | `Priority` | Select | `P0`, `P1`, `P2`, used by the at-risk rule | | `Due` | Date | Time off, no `Include time`. The commitment date | | `Completed` | Date | Time off. When the task actually finished | | `Blocker` | Text | Free text, used only when `Status` is `Blocked` | Notion’s [date property](https://www.notion.com/help/database-properties) supports `End date`, `Include time`, and `Date format & timezone`, and none of those replaces a second date column. Leave `Due` and `Completed` as plain dates so day counts stay whole numbers. The destination database needs nine properties: `Report` (Title), `Week start` (Date, Monday), `Week end` (Date, Sunday), `Completed`, `Overdue`, `Blocked`, `At risk`, `Next week` (Number each), and a `Status` that runs `Draft` then `Published`. Those five number properties are what let you chart a quarter of reports, one series per section the agent already writes, and Notion’s own [project reporting walkthrough](https://www.notion.com/help/guides/automate-project-updates-and-reporting-with-custom-agents) keeps the same split between project data and report pages. The diagram below traces one run: read the tracker, compute the window, write one page. ![Diagram: a nine-row Projects Tracker with status pills feeds a Custom Agent configured with instructions, tools and access, and a recurring Monday 08:00 America/Chicago trigger; a bracket marks its reporting window from Mon Oct 19 00:00 to Sun Oct 25 23:59:59, and an arrow leads to a Weekly Reports page titled Weekly Report 2026-W43.](https://adamtheautomator.com/wp-content/uploads/publisher/7c3cff05af2e736904bb7c989c944b45f87f0d8310696d31102afd889c443686.png) ## Nine Seed Rows and a Hand-Written Answer Key An agent with no test data is an agent you can’t grade. Seed nine rows covering Monday Oct 19 to Sunday Oct 25, 2026, run the agent on Monday Oct 26 at 08:00 America/Chicago, and write your expected counts first. Notion’s [agent best practices](https://www.notion.com/help/best-practices-for-creating-and-optimizing-a-custom-agent) expect several test runs before a schedule goes live. | Task | Status | Owner | Priority | Due | Completed | Blocker | | --- | --- | --- | --- | --- | --- | --- | | Migrate billing webhook to v2 | Done | Dana | P0 | Oct 21 | Oct 20 | | | Add CSV export to reports | Done | Priya | P1 | Oct 24 | Oct 23 | | | Fix duplicate invoice emails | Done | Dana | P1 | Oct 18 | Oct 19 | | | SSO session timeout bug | Blocked | Marcus | P0 | Oct 23 | | Waiting on Okta tenant config from IT | | Q4 pricing page copy | In progress | Dana | P1 | Oct 22 | | | | Audit S3 bucket policies | Not started | Priya | P0 | Oct 15 | | | | Vendor security review | Blocked | Marcus | P1 | Nov 6 | | Legal review pending | | Rewrite onboarding checklist | In progress | Priya | P2 | Oct 30 | | | | Deprecate legacy /v1 endpoints | Done | Dana | P1 | Oct 10 | Oct 9 | | Nothing sits at risk in those nine rows, because the only open task due inside seven days is the onboarding checklist on Oct 30, a P2 with an owner and no blocker text. The expected counts, written down before the first run: 3 completed, 3 overdue, 2 blocked, 1 next week, and 0 at risk. The three that closed are the billing webhook migration, the CSV export, and the duplicate invoice emails. That last one finished Oct 19 against an Oct 18 due date, so it counts as late and the report says so. The three overdue rows carry day counts: the S3 audit at 11 days, the Q4 pricing copy at 4, and the SSO timeout at 3, and both blocked rows (that SSO timeout and the vendor security review) belong in the blocked section of the report. ### Two Rows That Catch a Lazy Implementation “Deprecate legacy /v1 endpoints” finished Oct 9, outside the window, so an agent that ignores the date filter reports it anyway. “Vendor security review” is blocked with a due date of Nov 6, eleven days past the run date, so it lands in the wrong section unless the instructions separate blocked from late. Both rows stay in the seed because they tell you which mistake you shipped. * * * _**Pro Tip: Keep the seed rows in a duplicate of the database and test against a copy of the agent. Testing against live rows is how a real tracker ends up with a fake task called “test”.**_ * * * ## Write Instructions the Agent Can Repeat Every Monday The same [best-practices guidance](https://www.notion.com/help/best-practices-for-creating-and-optimizing-a-custom-agent) asks for the outcome first, then a real example of the output with the edge cases named. A weekly report needs all of that plus a rule for the week that already has a page. If you’ve already built [a Notion Custom Agent that turns task descriptions into checklists](https://adamtheautomator.com/notion-custom-agents-task-checklists/), this block follows the same shape: one precise job with every property named, and one fixed output. If the agent doesn’t exist yet, open **Agents** in the sidebar, click **+**, and choose **Create blank** so the instructions field starts empty. **Create with AI chat** will draft instructions for you, which defeats the point of writing them by hand here. Name it `Weekly Report`; the access step attaches `Projects Tracker` and `Weekly Reports` to it. Then paste this into the **Instructions** field and swap in your team and database names. ```text You produce the weekly project report for the Platform team. Audience: the engineering manager who runs the Monday standup, then the rest of the team. Keep the sections in the numbered order below, and keep each one under five bullets with one line per task. No prose summaries. Source: Projects Tracker (read only, never edit it). Destination: Weekly Reports. One page per week, no duplicates. 1. Window: previous Monday 00:00:00 through Sunday 23:59:59, America/Chicago. Compute the calendar dates from the run date and write them in the title and the header. 2. Completed: Status is Done AND Completed falls inside the window. Append "(late)" when Completed is later than Due. 3. Overdue: Status is not Done AND Due is before today. Sort by days overdue, highest first, and state the number of days for each row. 4. Blocked: Status is Blocked. Include Owner and Blocker text. A task that is blocked and overdue appears in both sections, with "also blocked" on the overdue line. 5. At risk: Status is not Done, Due falls in the next seven days, and at least one of these is true: Priority is P0 or P1, Owner is empty, Blocker is not empty. Write "0" when nothing qualifies. 6. Next week: Status is not Done and Due falls inside next Monday through Sunday. 7. Data quality: count tasks missing Owner, tasks missing Due, and tasks with Status Done and an empty Completed field. Report all three counts. 8. Title exactly: Weekly Report [ISO week of the window's Monday] ([Mon date] to [Sun date], [year]). Take the ISO week from the Monday that starts the window, never from the run date, which sits in the following ISO week. Put the window, the run timestamp with timezone, and the source database at the top of the page. 9. If every section is empty, still create the page. Write "Status: No activity this week." and "None." under each section. Never skip a week, never leave a section blank, never invent a row. If a section cannot be computed, name the property that is missing. 10. If a page with this week's title already exists, update it. Do not create a second page. ``` Rule 3 turns a list into a decision aid by sorting and counting days late, and rule 9 guarantees a page exists every week. A manual re-run on a Tuesday creates a second page for that week unless the instructions forbid it. ### The At-Risk Rule Needs a Definition Notion has no at-risk concept, so an agent told to flag risk invents a threshold each week unless you write one down. The rule above comes from this build, since [Notion’s agent best practices](https://www.notion.com/help/best-practices-for-creating-and-optimizing-a-custom-agent) set no thresholds: an open task due inside seven days fires when it is P0 or P1, when nobody owns it, or when the blocker field has text. With the seeded rows it returns zero, which is the correct answer when everything risky is already overdue and listed as such. ## Access: Two Databases, One Read-Only Open the agent’s **Settings**, go to **Tools and access**, and add both databases by name: `Projects Tracker` at **Can view**, `Weekly Reports` at **Can edit**. Then find the **Web access** toggle and turn it off ([the Custom Agents help page](https://www.notion.com/help/custom-agents)), since a report on internal work has no reason to read the internet. Notion states the default plainly on its [sharing and permissions page](https://www.notion.com/help/custom-agents-sharing-and-permissions): “By default, new agents have no access to all pages in the workspace.” Linking a page inside **Instructions** doesn’t add it to **Tools and access** either, which is the mistake that produces a first run with half the sections empty. The grant is explicit and nothing is inherited. Access revocation runs one way. Anyone who owns the project database can open **Share**, find the agent, and remove it, even without access to the agent itself ([sharing and permissions](https://www.notion.com/help/custom-agents-sharing-and-permissions)). Notion’s help pages don’t describe what the next run does with a revoked source, and its [best practices](https://www.notion.com/help/best-practices-for-creating-and-optimizing-a-custom-agent) say a failed run typically points to a permissions gap, so read **Activity** before you trust the page. A report that comes back with zero completed, zero overdue, and zero blocked is indistinguishable from a quiet week. ## One Trigger, One Timezone Custom Agents can carry several triggers at once, and the recurring one is right for predictable work. Open **Settings** > **Triggers**, add a `Recurring` trigger, then choose weekly and set the time with a timezone: Monday at 08:00 America/Chicago, for example. Notion’s [Custom Agents documentation](https://www.notion.com/help/custom-agents) is worth following literally at that step: “Review the next scheduled run time to confirm the schedule is correct.” That one line of text is the cheapest timezone check you get. A saved agent with its triggers toggled off won’t run, so confirm the toggle before you walk away. The trigger timezone is chosen at the trigger, per schedule, and it isn’t inherited from whoever reads the report later. ### The Window Is Absolute Dates, Not “Last Seven Days” Three timezones stack here: the schedule timezone, the timezone stored with each date value, and the display timezone of whoever opens the page. Notion’s date picker exposes `Date format & timezone` for the second, and each person’s account timezone sits under **Settings** > **Preferences** > **Language & Time** ([account settings](https://www.notion.com/help/account-settings)). Write the window as real dates in one named timezone: “Monday Oct 19 00:00:00 through Sunday Oct 25 23:59:59, America/Chicago.” “The last seven days” slides with the run date, so a manual re-run on Wednesday produces a different window than Monday’s scheduled run and a second, disagreeing report for the same week. * * * _**Warning: A window that ends Sunday 23:59 and a run that starts Monday at 08:00 sit near both daylight saving transitions each year. America/Chicago shifts by an hour on those weekends, so the calendar window stays correct, at 169 elapsed hours at the November change and 167 at the March change, while the run keeps its 08:00 wall-clock time and its UTC instant moves an hour later in November and an hour earlier in March. For a Monday morning report that is acceptable once you write the window as dates instead of hours.**_ * * * ## Four Tests Before the Trigger Goes Live Run the agent by hand before the trigger goes live. The **Activity** tab logs what triggered a run and what the agent did at each step, plus any error text when one fails, which is the fastest way to read a bad run ([Notion’s Custom Agents help](https://www.notion.com/help/custom-agents)). Start with the baseline: seed the nine rows, run once, and compare against your written counts of 3 completed, 3 overdue, 2 blocked, 1 next week, and 0 at risk. Then duplicate the database with no rows inside the window, point a test copy of the agent at the duplicate, and run; the page has to exist with every section reading “None.” For idempotency, run twice in the same week and confirm one page was updated rather than two created. For the schema break, rename `Completed` to `Finished` in a copy of the database, run, and confirm the failure appears in the data-quality line instead of producing zeros with no explanation. ### Why the Empty Week Still Gets a Page A skipped week leaves a hole in the database that looks identical to a broken agent, which destroys the only signal you have. The page below is the exact output an empty week should produce with the instruction block above. ```text Weekly Report 2026-W43 (Oct 19 to Oct 25, 2026) Generated: Mon Oct 26, 2026 08:00 America/Chicago Source: Projects Tracker (Tasks) Window: Mon Oct 19 00:00:00 through Sun Oct 25 23:59:59, America/Chicago Status: No activity this week. Completed this week (0) - None. No task has a Completed date between Oct 19 and Oct 25, 2026. Overdue (0) - None. No open task has a Due date before Oct 26, 2026. Blocked (0) - None. No task is in Blocked status. At risk (0) - None. No open task qualifies under the at-risk rule. Next week (0) - None. No task is due between Oct 26 and Nov 1, 2026. Data quality: 0 missing owner, 0 missing due date, 0 Done without Completed. ``` The data-quality line is the part most builds leave out, and it separates “quiet week” from “broken property.” A report saying zero completed and zero integrity problems is a fact. A report saying zero completed alongside a stack of tasks missing a completion date is a bug you can act on. The timeline below shows the difference between the two outcomes across three weeks. ![Comparison of two three-week timelines: skipping the empty week leaves Week 41 and Week 43 pages with a gap marked broken or quiet?, while always writing a page yields a continuous Week 41, Week 42, Week 43 run with Week 42 annotated 0 completed, 0 data issues. Calendar ticks read Oct 5, Oct 12, Oct 19, Oct 26.](https://adamtheautomator.com/wp-content/uploads/publisher/911128ffe7e6d256d2efd2f83c068a5fe430131900af231638e085bf54b71bff.png) ## Know What One Weekly Run Costs Notion’s [credit documentation](https://www.notion.com/help/manage-your-ai-credits) puts a medium task (writing a work report, for instance) around 10 credits and a complex one around 25 on Auto, so one weekly run over a few dozen rows lands in that range. Run count drives the monthly total; the report itself is cheap. | Quantity | Value | | --- | --- | | Runs per month at a weekly schedule | about 4.33 | | Credits per run (Auto, medium to complex) | 10 to 25 | | Monthly Select credit draw | about 43 to 108 | | Included Select credits per member per month (Business plan) | 500 | | Share of one person’s monthly Select pool | about 9% to 22% | The same agent on a daily schedule is roughly 30 runs a month, or 300 to 750 credits, enough on its own to exhaust a 500-credit Select allowance and push the agent onto workspace credits or into a pause. Weekly is both the cheaper setting and the correct one for a report nobody reads daily. The pause behavior is documented: “If the Custom Agent owner’s Select credits run out and no Notion credits are available, the agent will pause” ([how credits work for Custom Agents](https://www.notion.com/help/how-ai-credits-work-for-custom-agents)). An admin can fund the agent from a workspace budget instead of one person’s pool, which is the right setup for a report the whole team depends on. After the first month, read the per-agent breakdown in the [Notion credits dashboard](https://adamtheautomator.com/notion-agent-api-cost-tracking/) rather than trusting any estimate, including this one. ## Six Failure Modes and Their Fixes Nearly every failure below produces a page that looks fine at a glance, so each fix starts with making the failure visible somewhere a human will read it. The one exception is the failure that produces no page at all. | Failure | What you see | Fix | | --- | --- | --- | | Agent removed from the database share | Empty sections, run marked successful | Re-add under **Tools and access**; the data-quality line exposes the gap | | Property renamed (`Completed` to `Finished`) | Completed count drops to zero, no error anywhere | Name exact properties in **Instructions**; re-run by hand after schema edits ([date property docs](https://www.notion.com/help/database-properties)) | | No `Completed` property at all | Completed list fills with tasks whose due date fell in the week | Add the property before the first scheduled run | | Dates stored with `Include time` in another timezone | A Friday task lands in the wrong week, or in two weeks | Fix the window to absolute dates in one named timezone | | Manual run on top of an active schedule | Two pages for one week, one of them stale | Idempotency rule; check **Triggers** before a manual run | | No page at all | Silence, with the last run recorded in **Activity** ([Custom Agents help](https://www.notion.com/help/custom-agents)) | Check trigger toggles, then **Credits**, then plan tier and ownership | Open the agent, go to **Activity**, find the failing run, and read what triggered it and what it did at each step. Configuration changes are reversible through version history ([Custom Agents help](https://www.notion.com/help/custom-agents)), so a bad instruction edit doesn’t have to be rebuilt from scratch. ## What the First Notion Weekly Report Must Prove Six checks on that first generated page, each of which fails independently: 1. The header dates are the Monday and Sunday you intended, in the timezone you named. 2. The completed count matches a view you filter by hand: `Status` is `Done` and `Completed` inside the same range. 3. One overdue row’s day count is right. An off-by-one there usually means a date-only value met a date-with-time value. 4. Every blocked row carries blocker text. 5. Each task appears once, except the deliberate “also blocked” annotation. 6. One page exists for the week, and a second run updates it. The agent’s owner funds it and owns its instructions, and Notion documents that ownership transfers through **Settings** > **Members** > **Recently Left** when someone departs ([sharing and permissions](https://www.notion.com/help/custom-agents-sharing-and-permissions)). Enterprise workspaces also get an [audit log](https://www.notion.com/help/audit-log#custom-agent-events) for configuration changes and an Agent Directory where triggers can be disabled outright. Before the first real week, run the whole build against the nine seeded rows and put the generated page next to your written counts. Five matching counts plus an empty week that still produces a page means the agent is ready for its Monday trigger. If either check fails, the data-quality line names the property to fix. Share this article [Share on X](https://twitter.com/intent/tweet?url=https%3A%2F%2Fadamtheautomator.com%2Fnotion-custom-agents-weekly-report%2F&text=Notion%20Custom%20Agents%3A%20Automate%20Your%20Weekly%20Project%20Report)[Share on Facebook](https://www.facebook.com/sharer/sharer.php?u=https%3A%2F%2Fadamtheautomator.com%2Fnotion-custom-agents-weekly-report%2F)[Share on LinkedIn](https://www.linkedin.com/sharing/share-offsite/?url=https%3A%2F%2Fadamtheautomator.com%2Fnotion-custom-agents-weekly-report%2F) ## Related Posts ![](https://adamtheautomator.com/wp-content/uploads/publisher/3f25d9c85b2b81c28cc6c39755dac3c7/423d6296450a064f65dba039d6aec0339ac2ba3b86b2f0ef3706c81ff1e57223.webp) ### [Notion Automation: Pick the Right Tier Every Time](/notion-automation-agents-vs-workers/) Pick the right Notion automation tier: compare database automations, Custom Agents, and Workers, and split ticket routing to cut credit costs. ![](https://adamtheautomator.com/wp-content/uploads/publisher/3f25d9c85b2b8114a4eaea746fdde119/3e64cba088977a23c019e580855960ecff3338dc2fbb3b87595bece2d6b6a02e.webp) ### [Notion Custom Agents: Automate Support Request Routing](/notion-custom-agents-support-routing/) Build a Notion Custom Agent that reads each new support request, recommends an urgency, and assigns an owner from a routing directory you maintain. ![](https://adamtheautomator.com/wp-content/uploads/publisher/3f25d9c85b2b817b9916e1d7c0ce0d01/9553685a308ab52c50530c8cc96e715bb8f2194da3d00a46fb09a3e8836a559a.webp) ### [Notion Workers: Stop Agents Guessing Customer Lookups](/notion-worker-customer-lookup/) Build a read-only lookupCustomer tool in a Notion Worker, then attach it to a Custom Agent so it answers with a plan and status instead of a guess. ## Categories * [IT Ops](/category/it-ops/) * [Cloud](/category/cloud/) * [DevOps](/category/devops/) * [Home Ops](/category/home-ops/) * [Information Security](/category/infosec/) * [Software Development](/category/software-development/) ## Site * [Home](/) * [Tutorials](/tutorials/) * [Instructors](/author/) * [Advertising](/advertising/) * [Recommended Resources](/resources/) * [About Adam](/about-adam/) Copyright 2026© ATA Learning | [Privacy Policy](/privacy/)