Chaen.ai Work with me

AI news

Narrative Slide AI Agent: What I Found Trying It Out

🗓 ⏱ 5 min read

中文日本語

Narrative Slide AI Agent: What I Found Trying It Out

Narrative is a new slide AI agent. Here is what I found after trying it: the editing loop, running it through Claude Code, and where a tool like this fits in…

Narrative, a slide-making AI agent, has launched. It was built by Fukuyoshi-san, who also helps out my own company, and I tried it right after it shipped. This piece covers three things: what stood out when I used it, the criteria I apply before letting an AI agent anywhere near our decks, and how a tool like this actually lands in the Japanese market. For exact specs, check the official information — what I can add is the buyer’s view of what to look at before deciding.

What stood out when I tried the Narrative slide AI agent

The first thing I felt was that the UI and UX are genuinely well built. For a deck tool, that is the whole game. Generating text is something any general-purpose chat tool can do. Slides are different, because the visual output is the deliverable. A human will always touch the result afterward, so the speed of that editing loop decides whether you keep using the tool a month later.

In real work, the painful part of AI-generated decks is not slide one. It is slide twenty and beyond: aligning heading hierarchy, nudging a diagram back into place, cutting the slide you crammed three messages into. If those small operations feel bad, you end up using the AI for a rough draft and then rebuilding everything by hand, and your total time barely improves. The part that felt good here was exactly that layer.

There is a second point worth naming. Tools like this get good through an accumulation of small judgment calls, and those calls show up in the first ten minutes of use. I rate Fukuyoshi-san’s engineering highly from watching him work, and what you notice immediately in a product like this is usually the detail work.

The real reason our decks got better was borrowing someone else’s design

Honestly, the thing that transformed my company’s decks was not a tool. It was that I borrowed the design of Fukuyoshi-san’s decks as a reference. That surprised me at the time. Deck design looks like a talent question, and it turns out a large share of it is a pattern question. Patterns can be borrowed.

My process was simple:

  • Pick exactly one deck you genuinely think is good.
  • Study the structure, not the content: how many messages per slide, how big the gap is between heading and body type, how many colors are in play, how much white space is left alone.
  • Pour your own content into that same structure.
  • Cut whatever extra information crept in while you were pouring.

The fourth step is the hard one. Everyone wants to add more to their own deck. The reference deck looks clean precisely because someone didn’t add. This is where the division of labor with an AI agent becomes clear: the agent makes production against a pattern dramatically faster, while deciding the pattern stays a human job. Invert that order and you just get scattered output and more cleanup.

Four things I check before adopting a slide AI agent

There are more of these tools every month, so here is the checklist I actually use.

  • Is the output shippable as-is, or is it a draft? For internal sharing, a draft is fine. For a client proposal, the bar jumps.
  • How fast is the edit loop? This is the UX question, and it determines whether the tool survives past the first week.
  • Does it run inside a subscription we already pay for? Whether new spend is required changes the difficulty of adoption more than any feature does.
  • Can one person try it alone? Meaning: can you validate it without first getting the whole team to agree?

For Narrative, what I can speak to right now is the second, third, and fourth. The feel is good. The AI behind it can connect to Claude Code. It is free if you are using it on your own. The first one depends entirely on what kind of decks you make, so the fastest answer is to rebuild one real, current deck and judge it yourself.

Why connecting to Claude Code matters in practice

Being able to wire the backend to Claude Code is not a hobbyist detail. It is an adoption detail. When a company tries to add an AI tool, what stalls it is usually not capability — it is the cost conversation. The moment the question becomes “how much per month, out of whose budget, for how many seats,” the evaluation itself stops moving.

If the thing runs inside a subscription you already hold, you can test it without opening a new budget line. If your engineers already live in Claude Code every day, deck work rides the same path they are already on. Not adding another account to manage is a small benefit that quietly matters.

The honest caveat: this setup pays off in organizations where somebody is already using Claude Code. If nobody is, you need a different evaluation. Check which of those two you are before you form an opinion — otherwise you will judge the tool on the wrong axis.

How slide AI agents land in Japan, and what that means for overseas teams

One piece of context for readers outside Japan: in Japanese companies, the deck is often where a decision physically lives. Internal explanation, reporting upward, proposals to partners — things agreed verbally frequently do not move until they exist as slides. That makes deck speed and deck quality a direct input into decision speed. Which is why handing this work to an AI agent has more leverage here than people assume.

On how tools spread, “free for one person” is doing real work. In Japan it is common for adoption to run bottom-up: an individual tries something, it is good, it spreads inside the department, and only later does it get formally contracted. A free single-user entry point is less a feature than a distribution channel.

For overseas AI companies who want to reach Japan, I think these are the three things that get checked first. Does the layout hold when you put Japanese text into it? Does it run inside a subscription the customer already pays for? Is there an entry point an individual can try without asking permission? Being able to state those three clearly, in Japanese, moves the needle more than a feature list does.

How to use this at your own company

Two actions, nothing more. First, take one current proposal deck, rebuild it with a tool like this, and lay it side by side with your old version. Better or worse becomes obvious within minutes when they are next to each other. Second, pick one deck internally and declare it the reference for “good,” then share it. If you let an AI build decks before the pattern is settled, the output scatters and your editing time goes up.

Pattern first, automation second. Hold that order and almost any tool in this category will pay off.

FAQ

Is Narrative free to use?

It is free if you use it on your own. I tried it right after it shipped. I have not checked the terms for team use, so look at the official information for that.

What AI runs behind Narrative?

It can connect to Claude Code, which means you can run it inside a Claude Code subscription you already pay for. I have not verified any other setup.

How do I make my company's decks look better?

For me, the turning point was borrowing the design of Fukuyoshi-san's decks. Picking one deck you genuinely admire and copying its structure is faster than it sounds.

Based on a post on XOriginal post ↗

Share this article

XFacebookLINEはてブ

About the author

Masahiro Chaen
Masahiro Chaen Founder & CEO, DigiRise Inc.

CEO of DigiRise Inc. Has supported 600+ companies and 50,000+ people with AI training, consulting and development. GMO AI & Crypto advisor, Google Gemini AI advisor. 200K+ followers on X.