Automating Workflows with AI Agents Across the Organization

notion · Tech · 2026 · source Agentic: yes · promotional bias: high · source-backed · extracted with prompt 0.5.4 (schema v2)

Generated from extraction.json — do not edit by hand. Every claim cites verbatim source evidence; stated means the quote directly asserts the claim, implied means it supports it indirectly.

Problem

Domain: Internal workplace automation

Notion wanted employees across functions to automate repetitive knowledge lookup, work routing, triage, and team-support workflows using existing workspace context rather than waiting on manual Slack responses or hand-created tasks.

Constraints:

• agents must live inside Notion and work across Slack, Mail, Calendar, and other tools
• repeat questions need source links and accurate workspace-context answers
• feedback and bug reports must be routed without losing information
• non-engineers as well as engineers need to build and adapt agents
• many workflows involve ongoing triggers or schedules rather than one-off chat

Architecture

Notion's Custom Agents are configured once with a workflow, trigger, or schedule, then run automatically across Notion, Slack, Mail, Calendar, and related tools using existing workspace knowledge. Internal examples include a CX Slack Q&A bot that searches knowledge bases and returns source links, a Feedback Router that monitors Slack feedback and creates/routs tasks, a Web-team triage agent that files tasks and tags owners, and domain agents for style guidance, security alerts, office logistics, and weekly work-in-flight summaries.

Diagram confidence: reconstructed (reconstructed from prose).

flowchart LR
  notion-workspace-knowledge[Notion workspace knowledge] --> custom-agent-runtime[Custom Agent runtime]
  custom-agent-runtime --> slack-channels[Slack channels]
  custom-agent-runtime --> mail-and-calendar[Mail and Calendar]
  slack-channels --> cx-q-a-agent[CX Q&A agent]
  cx-q-a-agent --> notion-workspace-knowledge
  slack-channels --> feedback-router-agent[Feedback Router agent]
  feedback-router-agent --> task-databases[Task databases]
  slack-channels --> wilbur-web-triage-agent[Wilbur web triage agent]
  wilbur-web-triage-agent --> task-databases
  custom-agent-runtime --> task-databases

Components

Data flow

An employee defines a repeatable workflow and connects it to a trigger, schedule, or Slack channel. When a question, feedback item, or incident appears, the Custom Agent searches relevant Notion knowledge bases, docs, or runbooks, decides where the work belongs, and posts answers, source links, tasks, cross-posts, or tags back into Slack/Notion so the owning team can act.

Why this architecture

Notion uses embedded, workflow-specific agents because the work is scattered across internal knowledge, Slack requests, tasks, and team routines. Keeping agents inside the workspace lets employees use existing docs and databases as context, while triggers and schedules let repeated work happen without waiting for the original builder or a central automation team.

Models used: Notion Custom Agents · Notion AI

Scale

• Custom Agents run across Slack, Mail, Calendar, and other tools.
  “They live in Notion and work across Slack, Mail, Calendar, and other tools, using your existing knowledge for context.”article, stated
• Custom Agents have spread from one person's workflow to many teams.
  “What started as one person building a Custom Agent to solve an annoying problem has turned into tons of agents running across the company.”article, stated
• Support answers that previously took hours can now be returned in minutes.
  “And sometimes, it would take a few hours for someone to reply, meaning the support agent on call couldn’t get back to the customer right away.”article, stated

  ““Now our Custom Agent replies in minutes so our team can respond to customers faster and move onto the harder questions and more strategic decisions.”article, stated
• People share new Custom Agents in Slack every week.
  “Every week in our Slack channels, you’ll see this mentality coming to life. People share new Custom Agents in Slack regularly”article, stated

Alternatives considered (roads not taken)

Manual knowledge lookup and link pasting

Rejected because: Support agents waited hours for someone to find and paste links from the knowledge base.

Manual product feedback routing

Rejected because: Routing feedback, creating tasks, and preventing loss required constant attention.

Centralize agent building with only engineers

Rejected because: Notion emphasizes a broad builder base across Marketing, Sales, IT, individual contributors, and team leads.

Patterns

tool-use-orchestration (agentic)

Custom Agents work across Notion, Slack, Mail, Calendar, task databases, and knowledge bases to answer, route, and create work.

rag-retrieval (retrieval)

Q&A agents search team knowledge bases, docs, runbooks, and enablement materials before answering with links.

async-message-orchestration (agentic)

Agents are triggered from Slack channels, run continuously, and post responses or tasks back into the appropriate channel/database.

workflow-decomposition (operations)

Employees automate narrow repeatable workflows such as Q&A, feedback routing, web-team triage, style-guide feedback, and weekly work tracking.

human-in-the-loop-gate (operations)

Agents route tasks and tag people so humans can handle harder support questions, product changes, bugs, and security decisions.

Tradeoffs

Let employees build workflow-specific agents

Chose distributed Custom Agent creation by teams and individuals over centralized automation owned only by engineers.

Rationale: The article argues that one person's agent can be adopted by many teams and that non-engineering functions are active builders.

Embed agents where work already happens

Chose Notion agents connected to Slack, Mail, Calendar, and workspace knowledge over separate automation bots detached from workspace context.

Rationale: The agents need existing knowledge for context and must act in the same channels where questions, feedback, and work arrive.

Automate routing while leaving strategic decisions to people

Chose agents that answer repetitive questions, route feedback, and file tasks over agents making every downstream support, product, or security decision.

Rationale: The automation removes busywork so teams can focus on complex questions and important decisions.

Gotchas (what bit them)

Knowledge existed but was slow to surface.

Root cause: Support agents had to find a knowledgeable person, wait for document search, and receive a pasted Slack link.

Resolution: The CX Custom Agent searches the knowledge base and enablement docs and returns answers with source links in minutes.

Product feedback could be lost or delayed.

Root cause: Routing feedback to the right team and creating follow-up tasks required constant manual attention.

Resolution: The Feedback Router monitors Slack, identifies product areas, creates tasks, and notifies relevant channels.

Manual triage consumed time that teams could spend solving the work.

Root cause: Incoming web bugs or translation issues needed someone to listen, file tasks, and tag owners.

Resolution: Wilbur listens to the web-team Slack channel, replies, files a task, and tags the right person.

Evaluation

The article evaluates Custom Agents through internal adoption stories and workflow outcomes: faster support answers, task-routing automation, cross-team reuse, and broad participation by non-engineering teams.

Metrics: support answers moved from hours to minutes · tons of agents running across the company · weekly Slack sharing of new Custom Agents · agents used by CX, Engineering, Product, Web, Marketing, Environment, Security, Recruiting, IT, and team leads

Outcomes

• Support agents receive answers with links faster.
• Feedback routing and task creation run continuously with less manual coordination.
• Multiple teams adapted agent patterns from earlier successes.
• Employees across functions began building Custom Agents for their own workflows.

The article does not give exact agent counts, but it describes tons of agents across the company, weekly sharing in Slack, and adoption across many departments.