dmayle

There are four lights, my friends

I suspect it was one of the content review bots. In my first Show HN post, I handwrote the comment, but using a bulleted list style suggested by an AI agent, and that got flagged. I reached out to the mods, and they told me to switch to prose, which passed muster.

Thank you. Most of the wording and voice had to be tweaked multiple times since I've started... sometimes a phrase that I started with would seem grating after a few tries, so I put a lot of work into it. I'm glad to hear that you liked it.

I actually made a similar comment about bring-your-own-tokens on a different post a number of months back, and the response I got was that it didn't make sense. I didn't realize until building something myself, but it turns out that different models behave very differently to prompts, it's not just a question of model capability. If I tried to allow people to integrate their own models, not only does the product become much more unreliable, but the whole premise, which is that you don't have to worry about it, is now gone.

That's neat, did you wire a voice model to your own backend, or do you use something off-the-shelf and seed the context and save the output?

Personally, I tend to dodge phone calls, but it's true that I use this mostly with voice. Rather than having something talk back to me, I'm using a gesture and talking to it.

How often are you taking that phone call?

Thank you. No, this is my first pricing strategy. I came to it two different ways:

1) Bottom-up: this is the price that is sustainable given that there is a model reading your task list every morning

2) Top-down: this is priced between simple todo lists that just manage your data, and more complex AI tools that do a lot more than tasks.

Does your freemium app feature AI? Does it convert well into paying customers?

Thank you for the answer.

It's not exactly what I was hoping to hear, but it's also why I don't have a free tier, and I'm not relying on promo codes. If people won't pay for it at all, I want to learn that quickly, rather than building up a large number of users on a free tier that I can't sustain. Even a small paying audience can make this maintainable.

Voice input is prominent in the app, but the gesture setup is manual for a web app. I'll try to make it clearer on the site, help users get it set up.

I take your point. It's not the first llm-assisted app I started writing, but it is the first one I'm launching, so I guess that counts.

I'm hoping the answer to that is the same reason I built it, which is that proper task management is actually hard work. I think we like to fool ourselves into believing that if we removed this friction or that friction, we would just start the habit of managing our tasks, and only then would we get stuff done.

This app is about getting off that track, and just doing things instead. You might use it differently, but I've built it so that I only barely use the app. It's almost like a note taker, I just say the things I need to do into it, and check my email in the morning.

Thanks. That's both encouraging, and sobering, at the same time.

I'd love to know more about how that played out, two questions actually: 1) did people not sign up, or did they sign up and leave? 2) for the ones who left, did they dislike the triage, or did they think it was great and just not want to pay for it?

Getting the answers to those would help me choose where to focus first.

I've used it every morning since June, with incredible delight in what it does for my days, so I'm still going to try to find users who find the same value that I do.

Thank you for the offer.

There are already a decently large number of todo apps that don't have AI. They compete on audience, and integrations, so it's hard to differentiate and hard to deliver value worth switching. This one doesn't just add an AI chat interface, it uses the strengths of a model, handling arbitrary input, and re-reads your list every single day without getting bored or overwhelmed.

It's a prompt to a model with a set of rules that were built up from my usage (and those of testers) from roughly the last four months. It takes into account a deadline, if you specified one, urgency if mentioned, the text of the task itself. It looks at the age of the task, and its activity, the project, and it gets the whole list every morning. The model chooses three, and the reasons behind those three. I know it works, because I've used it and tuned the rules. When it got it wrong, I fixed it, using evals to make sure the existing cases didn't break.

Yes, you can build something with ChatGPT if you have a $20 subscription, and I'm sure that there are people who have already done that.

If you go that route, you do need a place to store your list that your assistant can see. For this app, it is the one and only place your task list lives, with voice input, and the UI has been tailored over many iterations, to make the common use cases as simple and as obvious as possible.

Getting an email to work the way this one does will take trial and error, because this one's not a summary, it's intentional choices, with the rationale, and it's complemented by the UI to customize your day.

Handling the prompts, as they change, will take work, and if you're using scheduled prompts, you'll need a way to know whether your new prompt works. I keep an eval set because the prompt change you make today can fail in subtle ways when looking at a different set of tasks. I've seen this nearly every time I've made a prompt change.

Thank you. It's a fair criticism, and I'll take note.

For the landing page, that's good to know. I didn't catch that, and I need to fix it, though I could definitely use your help with some specifics, if you're willing. I linked straight to the trial page. Is your criticism about that page, specifically, or about the home page?

What came out here that did not come across over there? Not speaking to design elements, but what could have told you it does more?

Thank you, and I agree. I want this to do one thing, and do it very well.

As to simple? If I wanted to joke, I'd say that my codebase would disagree, but in reality I see this as somewhat of a knife's edge. On the one side, you have the actual usability, and I've fought over every inch of the UI flows to try and keep them streamlined, useful. But on the other side, there is actual utility, and there are custom shapes for tasks I haven't addressed yet like scheduled tasks (e.g. watering the garden), repeating but varied tasks (e.g. school homework), and I need to make them fit in with the kinds of tasks people already put on a task list, like going to appointments, or personal projects like learning a guitar, work projects where a document must be finished before a presentation.

But the question is one I actually want to hear, and the answer is a resounding yes. I told my testers, when prompted about ideas (calendar integration, rich formatting), that I would have to weigh the utility for my users against the cost in complexity and maintenance, and it puts a silly smile on my face as I think about doing that work.

I'll take the opportunity to ask back to you. What does best in this area mean to you? What else could I be doing?

This is the best use case, you share all 500 things, and we manage them.

We do some organization for you (project creation, or sorting into projects you've already created), and each morning we go over all the open items, using the information you shared, any deadlines, urgency, and if you're waiting on a certain date to start. Then we rank them, and send you the three that come up, with the why, so you can reason over just those three, and correct us if we got it wrong.

And as for aspirational tasks? I had a "clean my office, this week" task that popped up as the third on my list, when the other two were taking me multiple days to do (actual tasks, not projects, just lots of trial and error so they took days), and after ignoring the task for two days, the app asked me if it was really important, and it was right, so I rescheduled it for after launch.

This app is meant to help you get things done, but it can't make you do them. So if you procrastinate, all it can do is read that signal, and offer what help it can (alternatives, reprioritization).

If it works for you, then that's great. I'd rather you be a happy customer of your own work, than an unhappy customer of mine. I built this for myself, first, and as I was using it, I had to tune it as new things came up (weekends, vacations, deadlines, etc.) just like you did.

I'd question your "spending time to learn someone else's system", here, though. The UI flows are built around using the app as little as possible. For me, I use voice input with a gesture, so bring out my phone, double-tap, say something, then put it away. I read the email once a day, and tap done 1-3 times a day (either from the email, or directly from the app). That's about all I have to do with it most days.

I'm really happy to hear that there's a hit, at least on two fronts, so I'll take the win.

In terms of the cost, I understand. I knew going into it that offering it as a subscription changes who will use it, and how many people will use it.

I didn't want to do ad-supported because distracting you just while you're thinking about what or how to do something ruins the usability, and ad revenue is not something that could cover the ongoing AI cost for this.

On the other hand, for someone who really liked it, I'd be really interested in your feedback. If you go through the signup, and then immediately cancel in billing, you'll still get the two weeks for free, and then you can see whether there's enough value for the price to be reasonable. You can export your data as many times as you like, and after you've cancelled, you can delete your account, and its data.

I agree with wanting to own your own data. That's why export was one of the first things I built (JSON and plain text), and why I built rules around user data (no access at all unless a user explicitly grants, and then only temporarily, and only for solving their problem), and tools that respect those rules.

In order to have a system doing the weighing, and the choices, there needs to be something that looks at the list, however, and end-to-end encryption would stop that cold. That's why it had to be on terms that meant no training off of your tasks, and no sales of your data, period. (It's in the terms) . There is audio, but that's so much of a privacy issue that we don't ship it to the servers, ever. It's stored locally on your device, so you can export it, or play it back if it's useful.

There is no self-hosted version of this that I could build and maintain. Other than the trust issue, what changes would a self-hosted version bring for you? Local models only?

That's fair. I wanted to share the story, which boils down to: I didn't want a todo list, I wanted software that did all the painful work of a task management system. It does for me what I did for myself when other task systems broke down: reads the list, emails me with three for the day. I built it for myself, and using it convinced me to offer it for sale. I tried to make it as reliable as possible, as fast as possible, and was unwilling to compromise on trust (no training, no use/sale of your data, no looking at your data without explicit recorded consent).

I used AI to help build it, but this isn't a one-shot app: I did research, UI flows (ascii first, then designs and mocks), code critiques, schema rewrites, performance analysis and tuning.

...and I get the idea that you can vibe-code a very polished looking version of something, and I think that's going to change software development in ways we haven't thought of yet. I think open-source becomes viable with an audience of only one person now, the true long tail.

However, just like you can already see older open source projects, the immediate value pales when compared to long term activity, and one-off projects die fast. (And I've built one-off task management tools for myself before).

I think the real value here is in having something that does all of the work, and without you having to spend your time on it. Just as I'm using Vercel so that I don't have to manage build systems and servers, and Neon so that I don't have to manage a database, I am using this so I don't have to manage my task list.

I get the concern. I don't think the characterization is true at all (I've used todoist, I've used asana), but I do care about privacy, and I do care about owning my data, which is why it is at least as easy to take out (if not easier), then it is to put in.

If you use it, your list goes to Anthropic to be ranked, under commercial terms that exclude training on it, and I never look at it (I've built support tools to help customers, but they won't access user data unless the user opens a support case, and then explicitly grants access for a temporary time). It isn't open source, but you can export everything at any time as plain text or JSON.

I also get the company concerns... When I was at Google, there were strict rules, and I used buganizer (among other internal systems), but even at Google there were external companies that were trusted with this kind of data once they'd been reviewed (for example, I used asana while I was at Google).

I'm a software engineer from the east coast (who has lived in Boston, NY, the bay area, UK, France), and I built a task system called 3 & Done. It's the place you trust to store everything you need to do. It accepts voice input for rapid entry, as well as text (it stores the audio locally so you can listen to jog your memory) and emails you every morning with the three most important things for you to do that day, and why, based on what you've said about the task: deadlines, urgency, sitting too long.

I've been at software engineering for about 44 years now... Grew up with it, which was unusual when I started, and I was always good about getting into the zone and doing things. The career started small, small company, then back office, the dot com startups, then smaller companies again after the bust, a non-profit, and eventually YouTube (Google at the time, but just barely) and Meta (just Facebook when I was there).

When I hit the startups, and later big tech, it was no longer enough to just be an engineer, to have drive, to be good, you had to be organized, you had to follow through, get stuff done, and that's where the task systems started. Twenty-five over the years (and yes, I've actually counted), from Moleskine notebooks with Getting Things Done, evernote (remember evernote?), bug trackers, todoist, asana, google keep, among others.

When I managed on the Facebook News Feed, I got into gantt charts (for me and my team), spreadsheets, OmniPlan... I remember apologizing to my director for being so disorganized (because I felt I was making a mess of it), and he told me that I was the most organized engineer he knew.

That was because I was doing all of the work in the background, the trimming, the sorting, the reviews, but it was just so much work. Which was why I changed from system to system. Each time it built up, and I had to spend more time looking at things I couldn't or didn't get done, when I was rushed and couldn't put time into today's review, it would pile up, and then using it was so much work that I would start taking mini-notes, just emailing myself a few things to do the next day.

That's actually the thesis of the app (but not where it started), that if I could just have a few things to do emailed to me each day, I could get it done.

This app actually started when I described my previous startup failures into a Claude web session, and it repeated back to me what I had just said. My previous failures were because I spent too much time building the stuff I needed, and not building what I was supposed to be building.

I decided that the app would email me three things to do every day (which matches best practices for attention), but that the day was always yours to decide. We won't overload you, but you can add more, if you're feeling up to it, or cut back, and we never shame and try our best not to nag (I often only finish two). If we get it wrong, you can tell us, and we'll re-rank it once a day, or you can set it manually, and if you do, we never override that choice. The web app is live, and native for both platforms is coming. In the meantime, on Android, you can use gestures to quick launch into the voice input so saving a task is only a gesture away (e.g. double-tap the back of a Pixel, double-press the power button on Motorola).

Now let's talk about the dirty parts. I used AI to help me build this. Anthropic models mostly, and the brains of the app come from Anthropic models (I like the quality, I like the privacy, absolutely no training on customer data, and the stance of the company). This app is not some one-shot wonder where I asked frontier models to build a task system and then I put it on a server to show you.

I have prompts for the various features (e.g. ranking, generating suggestions, task extractions), I have eval sets of task lists (not anyone's real tasks, but constructed from patterns in my tasks, or the tasks of testers who allowed me to fix the problems they were having), with the three I expect to come out. I store the eval outputs, and re-test whenever I add a new rule to a prompt, or whenever testing a new model version. The great thing about having this eval set is that it catches when I get it wrong. When I introspect a failed run, sometimes it's because I need to tune a prompt, but sometimes it's because I missed something in my own ranking.

I did the same thing with my infrastructure as well, so this time, instead of setting up servers, and building login systems, and managing databases, I built on Vercel (git push to serving Next.js), Neon (hosted postgres).

I started getting emails mailed to me by mid-June, but started getting usable suggestions on June 26, and I've been using it every morning since. I wanted it to be easy to get started, and easy to use, so the input handles mess (you can add multiple tasks in the same run-on sentence, and a friend who helped me test it just pasted his entire obsidian todo list in wholesale to get started).

I wanted to make sure that anyone could try it out, so I built a trial feature that uses the real in-app ranking with a list of items you provide.

If you want to try the service, there's a free 14-day trial (card required), and then it's $99/year or $9.99/month after that (plus sales tax). We warn you seven days before your card is charged and if you cancel, you keep access through the term (either the trial, or our last paid period). Online, through settings -> billing, no email, no call.

I really hope you'll try it out. If you do, please tell me if we did a good job picking out your day.

https://3anddoneapp.com/try

I watched this video by Marques Brownlee on the Oppo Find N6 (foldable).. https://youtu.be/eFUB_jL_XcM

It apparantly is a magic creaseless breakthrough... I'm eyeing one, and might put it on the Christmas wishlist...

Nope... What you're talking about is not privacy sensitivity, but trust.

For you, you don't trust the provider of the software, so you mean being able to put a barrier around your life, and the delivered software... you think that by running in a browser, you're preventing that code from being connected to anything else... However, anything that happens in that code is directly connected to the provider, whom you don't trust.

I'm referring to the opposite... having something connected to your life, but having a barrier disconnecting it from the outside world... It involves trusting the provider of the software (maybe it's open source), but even if you don't, you can install an app, cut the network, use it locally, then delete it before restoring the network, and you have a guarantee that what you did with it was private... It's possible to build that into a web app, but most websites don't work offline, where most apps (though not all), do work offline... even if it requires an initial network session to establish authorization

A ton of features work only or better on app:

  - anything privacy sensitive
  - anything processing intense (using optimized binaries)
  - widgets
  - camera and audio based features *can* work, but are more finicky (or outright broken on ios)
  - local storage (exists in web app, short lifetimes on web site)
  - performance/latency sensitive work
  - (I'm less familiar, but probably local sensors, like gps and acceleration)

It goes a step further... previously, card agreements (between merchants and the networks), required the prices to be the same between cash and credit (with the well known gas carve-out), but Durbin made that tying illegal, so now retailers are free to charge different prices for cash, credit, and debit.

The article has one thing mistaken, because it says that Durbin lowered costs for transactions, but credit owners got to keep their perks... That's not technically true (I worked at a supermarket when debit rails first went into effect, and I worked in payments when Durbin went into effect).

There are no benefits to credit users who use the debit rails, and the merchants would really rather you use the debit rails, because it is much cheaper for them. Durbin was mostly a win for the merchants, not a win for the customers.

However, if you take that to believe that the merchants lowered prices overall because they were paying less for transactions, than you might try to read into it that credit users kept their perks, while cash and debit users paid.

The true story, however, is that it's an equilibrium... When the costs go down, the saved money goes somewhere in between the two (supply and demand), and as long as there is competition, the savings are shared.

However, the real problem is that credit companies are allowed to invest interchange fees in perks at all. Credit card companies decided to take their low-risk pool, and offer them incentives, splitting the money they saved between themselves and their users, and using it as a way to pull more low-risk users. The more that happens, the more expensive it becomes for credit companies that serve mid-to-high-risk users... and since we can't stop offering credit to those users as well, those companies push for and get increases to interchange fees to cover the additional cost... which creates more room for benefits for the low-risk users, and the cycle begins anew. It's a vicious cycle that can't be fixed by changing amounts on the existing fee schedule... The only possible fixes would be in either disallowing these kinds of perks, or splitting the rail charges, and specifically charging less interchange for low-risk users (which dries out the benefit pool)

I have seen a few occasional sites that are even worse than this one.. Not only don't you get a reject all, you get two classes of settings, the default "types", and "trusted partners", and for trusted partners, about a third of them have what's called "legitimate interest" turned on, which means "we legitimately think we can share your data with this people after leading you to believe we haven't, and that's ok", and there are a hundred or more partners you have to scroll through and disable this for

You're right... I read the title too quickly... I'll have to look at Senko vs Softformer later...

Fun... This is something I actually care about...

I used to keep a version of whisperx around, because I think it's important to have not just transcription, but also timing and speaker identification (e.g. for subtitles)... It depends on pyannote, though, which has some wierd licensing (and is tougher to script the installs because of it), so I wanted to look at something that both had better transcription, and supported diarization (the speaker and timing). I decided on parakeet for the transcription with softformer (the diarization), but most of the available engines for it don't include softformer.

I coded up an OpenAI compatible server for parakeet-rs ( https://github.com/altunenes/parakeet-rs ) (which does support softformer) and I've been using it with OpenWhispr (a desktop app for transcription that handles all sorts of neat thing).

I'm doing CPU-only transcription (because I use my GPUs for other stuff and haven't gotten around to adding in the GPU-path), but it's incredibly empowering to be able to have local transcriptions at will.