Getting Closer to a "Clone AI" with Notion Custom Agents

Feb 17
3 min read

Why "AI Agents"?

In late 2025, Notion released a beta of its Custom Agents feature. The biggest difference from traditional AI assistants is that they work autonomously in the background. Instead of being called up in a chat, the AI does the work on its own based on triggers and schedules.

This is revolutionary for small teams. Envital is currently running multiple projects in parallel with domestic and international partners. Meeting note summaries, task transcription, research organization — these "should-do-but-can't-get-to" tasks had been piling up.

7 Agents Built in Just 2–3 Hours

At the February 2026 Team Monthly, I shared the results of this hands-on experiment with the team.

Here are some of the 7 agents I created:

Morning Review/Sorting Agent — Classifies and organizes voice-input memos, distributing them to tasks and notes
Weekly Review Agent — Automatically aggregates progress by account and generates a report page
DB Structure Change Detection Agent — Detects changes to database schemas and notifies the team of affected areas

The key is that instructions can be written as Notion pages. In other words, you can define how the AI behaves the same way you'd write any ordinary work document.

The Path to a "Clone AI"

Inspired by a case of an entrepreneur using a clone AI (for automating Twitter posts, etc.), Envital is gradually increasing the level of AI autonomy.

Step 1: Custom Prompts (Manual Execution)

AI called explicitly by the user, like /cv (cleaning up voice input) or /ct (transcribing tasks). Even at this stage, the time savings on routine work are more than enough to justify it.

For example, after a meeting, just typing /ct in a Notion meeting notes page automatically transfers action items as pages into the Tasks DB — with assignees, deadlines, and related projects filled in automatically.

Step 2: Custom Agents (Automated Execution)

On a daily or weekly schedule trigger, the AI automatically generates reviews and reports. Humans just review and revise the output.

For instance, the morning review agent launches every day at 7:00 AM JST, categorizes the previous day's voice-input memos, and generates a report page prefixed with "[AI Proposal]." By the time you open your laptop in the morning, the organized memos are already waiting for you.

Step 3: Agent-to-Agent Collaboration (Concept Stage)

Multiple agents work together to carry out information gathering → analysis → reporting → task generation in one seamless flow, learning from your feedback and habits to self-correct their own prompts. At that point, it truly starts to feel like "another version of yourself."

Small Teams Benefit Most from AI Agents

AI adoption at large companies is slowed down by security reviews and IT department approvals. Small teams, on the other hand, can decide today and start today.

Every Envital team member has committed to developing AI agent skills. The first step: "meeting note summarization → task creation" — the kind of work everyone finds tedious.

The key is not to aim for perfection from the start. Begin with custom prompts, then graduate the ones that work well to custom agents. Improve the instructions weekly. This feedback loop is what reliably increases AI accuracy over time.

Tips for Getting Started

Pick one task you repeat regularly — meeting note organization, email drafting, research summaries, etc.
Write instructions as if delegating to a person — eliminate ambiguity, and clearly define inputs and outputs
Don't aim for perfection; run the feedback loop — improve instructions weekly to increase accuracy
Make it a team effort — take one person's success and roll it out to other team members

AI agents aren't "magic." But with the willingness to keep refining your instructions, you'll steadily move closer to having another version of yourself.

At Envital, alongside business development in the mobility and energy sectors, we are actively working on improving operational efficiency through AI. We'll continue sharing our hands-on experiences with these initiatives.