I couldn’t agree more.
Here’s a graph from my article on writing about music: How to Write About Classical Music Without (Necessarily) Becoming a Critic | Strings Magazine
"As one AI candidly admitted to me: “I apologize for my potential limitations—while I strive for accuracy, I can sound authoritative even when wrong and may fabricate convincing details. I recommend verifying important claims, using me as a research tool rather than a definitive source, and always fact-checking specific historical information independently.”
We’re already stuck in the future.
Thanks for the discussion -
LV
Something new is starting right now with a well-known app. It’s not about “Adding an AI to an app”. It’s actually about an AI completely mastering the functionality of an app. It knows the manual by heart and has 20 years of experience in using the app, so to say. Without going into detail, applied to Scrivener, this could mean the following.
The AI does not cook. The AI does the washing of the dishes.
The human takes over the creative act of writing. But everything else, everything that has to do with the “appearance” of the text is done by the AI.
Of course, this requires that the developers of an app agree to provide the AI with all the specifications for their app.
I don’t know how I feel about this yet, but I can well imagine that many people will like this approach. We will see ![]()
It is so important that folk understand that LLMs do not – and indeed cannot – regurgitate facts. The response they produce may be factual, but the LLM has no means to determine whether or not the data returned is in fact factual – in part or in toto.
Therefore an LLM stating “I strive for accuracy” is nonsense. It cannot strive for accuracy because it cannot determine fact from fiction.
Is the AI a professional designer? Because professional page design tools actually have a great deal of automation, but without an experienced hand at the wheel that just lets you make a mess faster.
Have you seen how AI-functionality makes your Facebook ads way uglier than you could ever think off? No, AI is no Designer yet.
The thing is, every aspect of professional book production is hard: the writing, the editing, the design, the page layout…
People who want AI to “do the dishes” usually mean the parts they themselves can’t/don’t want to do. But just because you can’t do it, that doesn’t mean AI can. (Although it may mean you don’t have the ability to evaluate the AI’s output.) AI “art” is just as terrible as AI “writing.”
Folks, I have no idea if and how this will work in the end. Maybe not at all. I just made a random example.
Scrivener has so many functions that even an advanced user doesn’t know all of them.
How do you create a table? Mm, AI, insert a table with 4 columns here. The AI does it. No, I want the margins thicker …
AI, can I change this and that in the settings? The AI does it.
These kinds of things that have nothing to do with writing but can waste a lot of time.
Now I kind of feel like illustrators do about AI…
Agentic UI is definately becoming more useful. In fact, given the very finite and deterministic space that UI and apps provide, this will be meat and potatoes for AI UI agents, I think the tech is certainly more than there, it just needs the first steps in integration into the OS where it can have a maximal impact. The reason we don’t have it, like self-driving cars is the technical burden of legacy. If all cars were self driving, the tech has been ready for many years. But legacy on the roads mean human drivers, cyclists, and difficult to predict pedestrians. In the OS it is decades of technical debt where OS elements and features are poorly “labelled” and/or inaccessible. Apple should be several years ahead, in that their accessiblity features, file-level metadata in the filesystem and applescript integration across the OS should make agents much more robust and useful. But apple abandoned or never reliably supported these, it it great at implementing, but terrible at maintaining. As we don’t or can’t build from the ground up, then “AI” will often end up as a gimmick to try to distinguish app A from its competitors.
Hello,
No updates have been available for Windows since 2023, even though the world of writing has evolved with the arrival of artificial intelligence.
Are you planning a future update?
With the integration of artificial intelligence and the implementation of an API key (OpenAi or Claude Anthropic), for example?
Thanks.
Ludovic
Grammarly and the like have sufficient AI capability and integrate seamlessly with Scrivener, outside of their standard spelling and grammar checkers. Then there’s a host of LLMs you can link to and call on without leaving your Scrivener editor, if you like.
Apple Intelligence is available to Scrivener on Mac via menu options to limited improved operating system functionality.
What value/capability purpose-built Co-Pilot+PCs will ultimately bring to the market is still a bit of a hit and miss at this time with much of the capability still passing through various Insider stages right now. For instance the ability to describe the content of a photograph is interesting, but would somewhat dumb down user creativity.
The question is, what would you expect L&L should do at this time? Scrivener is not in any way behind the trend, except they don’t pretend to have AI capability that is the be all and end all for writers.
That’s like stating the world of digestion has evolved after people started eating soap.
Some of us would argue that it is for the worse. ![]()
I would expect that you’ll see on the Windows side what we’ve already done on the Mac side: allow users to decide for themselves whether to use the tools that the operating system provides.
Where I would find IA tools useful for Scrivener:
- Understand and do the proper search and replace without the need for me to make sure my regular expression is well crafted. Show the before / after for control, of course, before applying the change. Use case example: change all straight double or single quote to the proper curvy ones.
- Make some smart statistics about dialogues. How much dialogue vs prose. And more (see next)
- Highlight the dialogue lines with one color per character, and of course deduce who is speaking from the context when there is no attribution. Add a flag when unsure. Provide the statistic of how much screen time each character has.
- ‘Understand’ the author’s punctuation choices and point to some inconsistencies.
- Provide statistics about words but not those who are useless to know their frequency.
- Make a map or timeline with the themes that are conveyed in the draft and how many references they have.
- Suggest synopsis for empty cards (let the writer pick one out of three, following the logic of the other synopses).
- Take as input the rules the writer decided to follow for the manuscript (like “character A never says xxx but instead will xxx”, or “the POV always follow character A except when xxx”) and point out possible inconsistencies.
Where I would not see IA works for me in Scrivener:
- (this is a NO) Generate some content in my stead.
All of that, and ideally, locally processed.
There is a crop of online writing systems that employ AI. I suspect that this is where we’ll see this sort of innovation. Indeed, some of what you have described is already available.
I agree it would be ideal to have the AI run locally. Such an app would employ an offline LLM of one flavor or another.
I believe most of Apple’s current language tools run locally (nothing leaves your machine), so this feels technically feasible using smaller models without eating all your local RAM.