I am frequently baffled how Chinese labs (Like Alibaba Qwen) tend to make better 'marketing' material than a lot of what OAI/Anthropic does, especially Anthropic.
The Qwen3.8-Max release trailer was one of the better examples of it.[1] Meanwhile Anthropic releases tend to be very dense and maybe a 20-30 second teaser of an abstract beige flower. I get it, they aim a _ton_ towards people who release their release reports and system cards, but at what point do you start marketing to the everyday people if your goal really is "Beneficial/Benevolent AI for everyone."
---
I suppose the Chinese sound more optimistic than that?
In my corner of the world we were hit with the expectation that we are pushing up to 10 things forward at the same time since agents can operate in parallel.
I also realize the juxtaposition of that video coming out of a Chinese company and the slave labor used across the country. Still a guy can dream.
Google Search didn't change since 1997. It's just textbox.
Chat interface probably should stay the same.
My biggest issue is how primitive text editor is. For a trillion-dollar company, they could make it work better with their beloved markdown format. Tripe-backquote will work when I'm typing new message, but it won't work when I'm editing message. It's also unnecessarily smart, so I often have issues editing single-backquoted strings. Ideally I'd prefer to keep all markdown markup intact and just add proper styling on the fly, as I type it, so I can actually edit markup if necessary. For example I can type ``` in ChatGPT and it instantly switches to code editor. But what if I wanted to type ```bash? It doesn't let me. Also I hate that there's no way to make <Enter> insert newlines. Most of my messages are multiline and I sometimes press <Enter> just to send unfinished message and waste tokens.
But yeah, outside of that little rant, I think that textedit UI is pretty much essential and will probably stay relevant for a long time.
Google Search changed tremendously since 1997. Yes, it's still a textbox. But commands changed and disappeared all the time.
"Google Search didn't change' is only true if you used it as a 'here's a keyword, give me random results' box.
Yes, like 99.99% other users I guess.
But then where does the money come from?
All of the main AI companies have speech-to-text. You can interface with them solely by talking to them, which is how every sci-fi technology ever worked.
That people don’t do use that as much isn’t that chat is antiquated but it has qualities of its own. Being forced to think a little bit before writing a prompt comes out cleaner and more efficient. Much like planning a presentation before doing it.
Summarisation is one of the biggest features that makes the AI providers apps killer apps in their own right. The ability to ingest arbitrary unstructured data, digest it, and return it in an more familiar format is a massive time-save and huge productivity boost.
That’s one key part, the other is that everyone is a developer now, so you’d better make software that’s customizable in some way. Opening the code ensures this pretty well (a good plugin system is even better but you can do both).
> Open Source it.
I don't think every piece of software is open source today, nor do I think it should be. Data security is a valid and critical concern, but that doesn't automatically call for open sourcing, right? __In the age of agents__, going open source also means your work will be copied and rebranded within an hour or two.
I agree that note-taking and summarization shouldn't be hot topics today. Solutions existed before GPT-3, and they weren't that bad. Still, I believe AI apps should sell a better experience for specific use cases, just like non-AI software. An ecosystem helps, but if the whole pitch is around plugins, you are essentially selling a platform—and that platform is unlikely to attract good builders.
Open source is a great way to get the attention of technical users, especially the ones who use LLMs for other tasks, already.
Open source is a great way to get the attention of technical users, who use LLMs for other tasks, already.
That argument has become me after the x-craft series of vibecoded apps, which appear to be functional to the level I used the 'closed-sourced' tools.
Second, I also think that (outside of search), we'll look at the chatting interface of today's AI tools in 5 years and laugh at how antiquated they seem.
Finally, Obsidian (not to mention Apple Notes) is free. So to be competitive here, your value prop has to be an absolute home run. I'm never going to root against fellow entrepreneurs, but this just seems like a more pricy, maybe more stylish Obsidian (without all the plugin ecosystem? and you also need a browser extension?). Obviously, hindsight is 20-20, but if I were to release a productivity app today, I would do a few things:
1) Open Source it. Especially in the age of agents, people want to make sure that no shenanigans are being done with sensitive data, especially when it involves private notes.
2) Dual license it. Folks have done this since Red Hat, and I think dual licensing for personal & professional use (or have a revenue cutoff like Unreal or Unity) is the move.
3) Create a plugin ecosystem around it. I think the great thing about agents is that even non-technical folks can come up with stuff and vibe-code a plugin. Incentivize people to do this and build a community.