Notion Custom Agents after 4 weeks: What really works

Four weeks ago, I set up four custom agents in Notion and wrote about them. At the end of the article was a promise: "I'll see how it works in practice. I'll report back."

Here is the report. Honest, without a glossy conclusion.

What has become of the four original agents

A quick reminder: I had set up four custom agents on launch day. A transcript auto-analysis, a morning briefing, a tally form generator and a slack assistant.

Four weeks later, all four are still there. But not all of them are used with the same intensity.

Morning Briefing: the daily companion

The Morning Briefing agent has turned out to be the most useful of the four. Every morning there is a structured overview in Slack: What's on the agenda today? What is overdue? That sounds simple. And it is. But that's exactly why it works so well.

My morning now looks like this: First coffee, cell phone in hand, open Slack. The briefing shows me everything important at a glance. Later, when I'm sitting at my laptop, I click directly from the Slack message into the linked Notion entries.

The best thing about it: the agent has actually uncovered things that I would otherwise have overlooked. I work with two date fields in my task database, "Planned for" and "Deadline". My normal notion view filters on "Planned for". For two or three tasks, I had accidentally only set the deadline, not the planning date. They were invisible in my standard view. The agent found them anyway.

Without it, I would have simply forgotten about these tasks.

Transcript auto-analysis: the knowledge engine

This agent has become an integral part of my workflow. I often watch YouTube videos on business topics, marketing psychology, SEO, GEO, AI. The problem is that many of these videos are 15, 20 or 30 minutes long. They end up on my "watch later" list, which gets longer and longer as the months go by until I clear it up at some point.

My workflow today: In the evening on the sofa, I copy the URLs to the clipboard and have them transcribed by TurboScribe. I export the transcripts as Markdown and import them into my own Notion database. As soon as a transcript is in there, the auto-analysis starts.

The result: summary, key messages and concrete business ideas that I can use directly.

In four weeks, over 60 videos have run through this workflow. Topics ranged from copywriting psychology and GEO strategies to Claude Code tutorials. This resulted in specific tasks and content ideas. Sometimes, out of seven ideas generated, not a single one is usable because it doesn't fit my business. But there were also cases where I was able to use five out of seven ideas directly.

The "watch later" problem is solved. The videos are no longer ignored, but systematically converted into knowledge and tasks.

Tally Form Generator and Slack Assistant: On standby

In brief: Tally is a form tool that can be used to create surveys, feedback forms or registration forms. It is easy to use, flexible and integrates well with other tools. The Tally agent creates such forms automatically based on a description in Notion, without me having to work in Tally itself.

I hardly used the Tally agent during the four weeks. Not because it was bad, but simply because I had no use case. No form needed, no agent started. It stays in the workspace and is activated when needed.

It's similar with the Slack Assistant: It converts unstructured Slack messages into Notion entries. Handy when you're on the go and want to quickly capture an idea. In winter, I'm usually sitting at my laptop, so I don't need this as often. In summer, when I'm out and about more, I'm sure I'll use it more often.

Neither agent has failed. They are simply waiting for their moment.

Four new agents in four weeks

Between launch day and today, I not only used the existing agents, but also built new ones. This didn't happen according to a plan, but whenever I thought that a recurring task could be automated.

Meeting Analyst

The Meeting Analyst analyzes note entries in my database. When I transcribe a meeting (either via the internal Notion function or via a Fireflies transcript), the agent kicks in and derives tasks and content ideas.

I use it very often, actually for most meetings. It delivers solid results and saves me having to work through minutes manually.

Social Media Publisher

This agent is automatically activated as soon as a blog article is set to "Published" in my content database. It then independently creates social media drafts for LinkedIn, Facebook, Mastodon and BlueSky, creates them as separate entries in the content database and notifies me in Slack.

I have created a lot of blog content in the last few weeks. This agent has therefore been in constant use and has relieved me of a lot of manual work. The quality of the posts is good because the agent has access to my brand compass and knows the style of my brand.

This is an important point, which I will come back to in a moment.

TCCI Skill Generator

I am a member of the TYPO3 Education & Certification Committee. Among other things, we create the syllabus for the Integrator certification. The Skill Generator creates suggestions for new learning objectives from URLs (mostly from the official TYPO3 documentation).

The results are not perfect. In a field as specific as TYPO3, a lot of experience and contextual knowledge must be incorporated that the agent does not have. But that's okay. We use it as a supplier of ideas. It generates several suggestions per topic, and we pick out the cherries: the description of suggestion 1, a goal of suggestion 2, an adjustment here, an addition there.

The advantage: we never start with a blank sheet. The agent provides us with a starting point, which we then revise as a team. This speeds up the work noticeably, even if human review is always necessary at the end.

You have to be very careful, especially with important topics such as an official learning path for TYPO3 integrators.

Weekly review and content ideas

This agent is the newest in the team. Every Friday evening, it automatically evaluates the completed tasks and published content of the last seven days and derives content ideas from this.

After three weeks, the conclusion is: it's okay, but still too early to make a final judgment. The next few weeks will show whether the ideas generated are really useful or whether the agent too often comes up with suggestions that don't fit.

The agent that failed

An experience report would not be honest if I only wrote about successes.

I had built another agent to keep an eye on the official TYPO3 changelog. The idea: The agent checks new entries to see if they are relevant for TYPO3 integrators and informs me about them.

Sounds sensible. But it didn't work.

The agent was often wrong in its assessment of what was relevant and what was not. Evaluating the changelog correctly requires a deep understanding of how integrators work with TYPO3 on a day-to-day basis. Which changes affect them directly? Which are only of interest to core developers? The agent was unable to make this distinction reliably.

In the end, I was quicker if I simply looked at the changelog myself. My experience from almost 20 years of TYPO3 work is not so easy to pack into an agent instruction.

The agent is now deactivated and will probably be deleted.

What has really changed

The most exciting question after four weeks is not which agents are working. It's whether something has changed in the way I work.

The answer: yes. And not in the way I would have expected.

I'm not just saving time on existing tasks. I'm doing things that I wouldn't have done before. More content, more blog articles, more ideas used from analyzed videos. The transcript auto-analysis alone has led to me systematically drawing knowledge from external sources instead of letting it rot on a "watch later" list.

That's the real added value: not saving time on what I already have, but opening up new possibilities that weren't realistic before.

However, I need to get back into the habit of doing this: Actively paying attention to which recurring processes are suitable for an agent. The initial fascination of seeing automation potential everywhere faded after a few weeks. That is normal. But consciously looking at your own workflows is the key to getting the most out of custom agents.

The rule of thumb is simple: what things do you do regularly that always require the same steps? These are your candidates.

Why some agents deliver better than others

Anyone who works with AI knows the principle: if the basics are bad, nothing good can come of it. This also applies to custom agents.

With my content agents (social media publisher, weekly review), the quality of the results is significantly better than with the TCCI Skill Generator. The reason is simple: my content agents have access to my brand compass. This describes exactly how I communicate, what style I use and what wording I avoid. This allows the agent to write text as if it came from me. The rework is minimal.

The Skill Generator lacks this foundation. Preparing TYPO3 specialist knowledge in a form that an agent can use reliably would be enormously time-consuming. So we in the TCCI team use it for what it does well: Delivering ideas that we then refine with our expertise.

This is not a groundbreaking insight. But it does show where the leverage lies: If you want custom agents to deliver good content, you first need clarity about your own style and positioning. A brand compass (or something similar) is not a nice-to-have, but the basis for useful results.

Costs: still an open chapter

Custom Agents are only available in the Notion Business Plan. The agents are currently still in a free test phase. From May 2026, a credit system will be introduced, the exact prices of which I do not yet know.

To be honest, I can't yet assess the cost-benefit ratio. What I can say: When the credit system arrives, I will take a close look at which agents are worth their price. Some will stay. For others, I will consider switching them off.

The barrier to entry that nobody talks about

If you're reading this article and thinking "I want that too", I have an important point to make.

Custom agents only work within a well-structured Notion workspace. And "well-structured" means: database-based. Databases for projects, tasks, content, notes, resources, contacts. With clean properties, relations and views.

It takes time to set up. Not days, but weeks to months. Depending on how complex your business is.

An agent who creates your morning briefing needs a task database with date fields and status properties. An agent who generates social media posts needs a content database with platform fields, campaign relations and a clear structure.

Without this foundation, even the best agent is useless.

This is the uncomfortable truth: custom agents are not "just set up". They are the tip of an iceberg, and the iceberg is a well thought-out Notion workspace.

Conclusion after four weeks

Custom agents are not a miracle cure. Some work brilliantly on a day-to-day basis (Morning Briefing, Transcript Auto-Analysis), some are waiting to be used (Tally, Slack Assistant), some provide good starting points that need human rework (TCCI Skill Generator), and some fail because the task is too complex to automate (Changelog Agent).

What has changed: I do more, not just faster. The agents have enabled workflows that were not realistic before. Analyzed 60 videos in four weeks and turned them into usable knowledge. A structured briefing every morning without lifting a finger. Social media drafts that are created automatically as soon as a blog article goes live.

The most honest sentence after four weeks: The fascination of the new has faded. What remains is the real benefit. And that is considerable with the right agents.