Daniel Rosehill

One of the most underappreciated products in tech in my opinion. The only criticism is that the API could be a little more developed as could some of the extra features. However: Markdown handling is the best that I've seen (and it does autorendering and not double pane mode which can be really distracting). And it scales really well - I've dumped thousands of notes into collections and the recall speed has been impressive.

Thank you so much for providing this! I'm exploring a bunch of uses for STT and TTS (my current one is using TTS to create my own personal podcast based on prompts I send to an agent). There is a place for "top tier" TTS, sure. But there's a much bigger one, for me right now, in having an affordable speech API that I know I can afford to play around with. But that's underselling it. The quality is actually good and (unlike so many speech APIs) .... no weird hurdles getting the API calls to generate ... and a clean binary to work with on the other side.

I discovered the wonders of ASR speech-to-text about six months ago. I’d tried speech-to-text years ago, but revisiting it recently really opened my eyes to how far the technology has come. That kicked off a bit of a journey looking for the right client. At the time, I was using another voice capture tool. It worked well enough, but the UI felt a bit clunky. Once you start using tools like this seriously, though, you realize how much potential they have to become a core part of your productivity workflow. I knew I had something decent—but I also knew there had to be something better out there. That’s when I came across AudioPen. Pros: Really thoughtfully designed UI. Clean, intuitive, and just gets out of the way. I use it heavily—often creating multiple notes in a row—and it handles that well. You can finish one note and immediately start the next. Webhook and API support. This was something I specifically needed, and it’s what pushed me to keep searching. AudioPen does this well, giving me flexibility to move notes into other systems. Smart rewrite functionality. One of my core use cases is converting speech into useful formats like emails or technical documentation. For that, you need good transformation prompts. AudioPen has a solid, well-written prompt library that isn’t bloated and actually works. You can still access the original transcript, which is important. Sometimes I want the clean version, sometimes I need the raw text—this gives me both. Works well on Android and across platforms. The interface makes it easy to manage notes, copy to clipboard, assign tags, download audio—it’s all just smooth. Signed up for the yearly plan and happy I did. I heard about it on Reddit and I’m glad I took the time to find something solid and reliable. Suggestions / Things to note: I’d really like the ability to add custom system prompts. The built-in ones are good, but user-defined prompts would make it even more useful. There should be an option to fully disable rewriting. Even though you can view the original transcript, sometimes I just want to record and keep it as-is. 15-minute recording limit per note. That’s fine with me, but worth mentioning. The countdown timer during recording can be a little distracting. I’ve run into the occasional sign-out due to authentication, but otherwise, it’s been stable. Overall: AudioPen does what it sets out to do: voice-to-text capture, and it does it really well. A lot of tools in this space try to do too much and end up being complicated. This one stays focused, and that’s what makes it great. If you’re looking to bring voice into your workflow without the overhead, this is one of the best tools I’ve found.

I think that the idea of an agentic IDE is really interesting and I'm fully supportive of the transformative potential in this app and in the concept generally, breaking down barriers to entry for development and making programming accessible to far more people. Equally, I think that large measures of realism are called for regarding the actual capability of AI to deliver on this promise at the moment, especially with regard to handling context, which is inherently challenging. I think that this creates a responsibility for those generating the first wave of these tools to market their products responsibly. I began absolutely loving Windsurf and I'm still a big fan of the product and see its potential as vast. However, I think that it has seriously failed in its obligation to strike this balance and to provide a fair service to its users. While the company's social media speaks of an endless train of feature additions and advancement, the actual experience of using this on a day-to-day basis, as evidenced also by the Reddit communities, is rather different to the marketing. Personally, I found the product to be beset by an almost endless train of bugs that affect the ability to get anything done with it. At the moment, there's no completions happening on Anthropic, which has been the case many more days than not during the past few months. Long-term, I now have very mixed feelings about this product and the company behind us Because I feel that my trust in their product has been to an extent taken advantage of. I hope that others move in to fill this space and are more transparent in their engagement with consumers, especially with regard to uptime and product availability. Users should not feel the need to resort to bypassing official support channels to communicate the idea that the product is not available.

Whisper has been one of the few AI technologies that has been totally life-changing. When I tried out Whisper for the first time, it had probably been a decade since I first experimented with speech-to-text software and expected that not much would have improved. Qualitatively it's such a leap over what came before it that it almost feels like a different product class. My only criticism of Whisper is that there isn't a greater variety of clients supporting it, especially in the realm of Android tools. Since discovering it, I've adopted just about every whisper technology that I can and spent most of my day voice typing now.

Overall I think that Dify AI is a nice orchestration and backend platform for creating AI agents. However, given that it doesn't really contain anything in the way of front-end, there is reason to question whether.using it makes much sense over choosing an all-code approach.

Very underrated workhorse of a large language model. Particularly useful for more instructional workflows. supports high volume prompting with ease.

Long-term I think that Gemini and Google will have the edge over competitors. The structured output builder in AI Studio in particular is very impressive and 2.5 enables extremely long continuous outputs that are valuable for lots of emerging use-cases.

The API is useful, although they continue to have some issues with citations in particular frontends. Once this is resolved, I would love to move all LLMs over to it. Long term, I think it will make much more sense for builders to use LLMs with search integrated rather than having to go through the additional tooling of search API add-ons.