A 30-Day Test: Living With a Voice-Only Planner

I built Lunelo. I spent roughly eighteen months obsessing over voice input, AI parsing, and the question of whether a planner could work without a single typed character. So when I decided to use nothing but Lunelo for thirty consecutive days — no Notion, no scratchpad, no typed task anywhere — I knew I was not running an objective test.

That bias is real and I am not going to pretend otherwise. What I can offer is something slightly different: the specific friction points, the unexpected moments of clarity, and the honest list of things I would change about my own product if I were starting the design over. A founder who dogfoods and reports only the wins is writing marketing copy. I am trying to write something more useful than that.

Here is what thirty days of voice-only planning actually looked like.

Week 1: the muscle memory was wrong

The first thing I noticed was my hands. Every time I sat down at my desk, my fingers moved toward a keyboard before I had consciously decided to do anything. Typing tasks is a reflex I had built over a decade of using Notion, Things, and a rotating cast of apps I kept meaning to consolidate. The muscle memory pointed in the wrong direction.

Speaking a task aloud felt performative at first, especially at the office. There is something mildly absurd about sitting at a desk, looking at your phone, and saying “remind me to follow up with the design contractor about the revised scope.” The words feel louder than they are. My colleague glanced over twice on day two. I explained I was testing something. She nodded in a way that suggested she did not entirely believe me.

The syntax question also took adjustment. When you type a task, you abbreviate. “design scope follow-up” is enough because future-you reads context into it. When you speak a task, the AI has more to work with — Whisper transcribes, Claude parses intent and pulls out the action, the time, and any implicit priority — but you have to learn to speak in complete enough thoughts to give it that context. “Follow up with Marco on the revised design scope, this week” works. “That Marco thing” does not.

By day five I had developed a rough speaking style: subject, verb, context, rough time. It took five days of conscious effort to get there, which is longer than I expected. The app itself did not teach me this; I figured it out through trial and error. That is something I have thought about a lot since.

The week ended with a realization: the awkwardness was not a product problem, exactly. It was a habit-change problem. But product design can either ease that transition or ignore it. I had largely ignored it.

Week 2: the unexpected privacy upgrade

By week two the awkwardness had dropped to background noise, and something I had not predicted started happening: I was planning during walks.

The context shift mattered. When planning lives in a keyboard app, it lives at a desk. You open a laptop, you review your list, you type things. The planning moment is tethered to a screen. When the capture method is your voice and a phone in your pocket, the planning moment can happen anywhere with ambient quiet — a walk to the coffee shop, the stretch between a meeting and lunch, the two-minute pause before you leave the building.

I started doing a short voice planning session every morning during a ten-minute walk. Nothing formal. I would run through what I was thinking about, what felt urgent, what I kept avoiding. Lunelo captured it. The AI stripped out the filler and gave me back a short ordered list by the time I got home. The daily focus view meant I was never staring at an overwhelming backlog — just the things that actually mattered today.

What surprised me was the secondary effect: I was on my phone less. When task capture moves off the screen, the habitual phone-check loses one of its rewards. I was not opening the app to browse or reorganize or feel productive without being productive. I opened it, spoke, closed it.

I would not call this a planned feature. It is an emergent property of the input modality. But I noticed it clearly enough in week two that I started logging it deliberately.

Week 3: the failure modes I hit

Week three is where the experiment got uncomfortable, which means it is also the most instructive part.

The loudest failure mode was the café. I work from coffee shops two or three days a week, and voice capture in a busy space is, at best, unreliable. Whisper can handle moderate background noise reasonably well, but a steam wand firing three feet away during transcription is not moderate background noise. I lost several captures mid-sentence and had to repeat myself. Once, the transcript came back coherent but wrong — a task about “the API integration spec” became “the app integration spec,” which is close enough to cause quiet confusion later.

Multi-language was the second problem. I think in Russian when I am tired. My team communication is in English. My planning, it turns out, is in whichever language I happen to be in at the moment. Lunelo is English-only by design — that was a deliberate early scope decision — and switching mid-sentence sometimes confused the parser in ways I had not fully stress-tested before using it as my only planning tool. “Remind me to call Иван about the server costs” produced an interesting result.

Numbers were a third friction point. Dictating numeric specifics — “the 14th, at 3:30, for the Q2 review” — worked most of the time, but enough times it did not that I learned to speak numbers more deliberately, with short pauses around them. This is a learnable behavior, but it should not be necessary.

The subtlest failure mode was re-triggering. If I paused mid-sentence to think, the recording occasionally stopped and submitted a partial capture. The silence threshold that determines “user is done speaking” is probably tuned for average speakers. I am not an average speaker when I am working through a complex task out loud.

Week 4: the moments it actually worked

After three weeks of adjustment, week four produced the moments I had actually been building toward. I want to be specific about what these looked like, because specificity is what separates “it worked” from marketing language.

The clearest one happened in a meeting. Someone mentioned a follow-up that needed to happen before the end of the sprint. I did not reach for a notepad or unlock my phone under the table. I held the phone below the table edge, pressed record, and said nine words. The task appeared in my list. Nobody noticed. I have never captured a meeting action item that cleanly with any other tool.

The second moment was a walking lunch. Forty minutes of outdoor time, three voice notes, fifteen tasks organized and prioritized by the time I sat back down. I arrived at my desk feeling lighter than when I had left, which is not something I can say often about the planning process.

The third was post-call wind-down. After a long strategy call, I used to open Notion and spend twenty minutes writing up the things I needed to do. In week four, I spent four minutes speaking those same things, then closed the app. The AI had done the parsing. The reduction in friction at the end of an already draining call was noticeable in a way that mattered.

I also found that the absence of streaks and karma systems made a real difference during a difficult week mid-month. I missed two days of structured planning. The app did not penalize me. I came back to it without the usual layer of guilt. That is not an accident — it is a deliberate design position I believe in — but experiencing it as a user rather than a designer reinforced why it matters.

What I would change about my own product after 30 days

The hardest section to write, and the one I think earns the rest of this piece any credibility.

The microphone indicator during recording is too subtle. When I was in noisy environments, I sometimes could not tell whether the app was actively capturing or had already stopped. A more visible, higher-contrast recording state — not intrusive, but clear — would have prevented several failed captures. I know why the current design is quiet; we did not want the UI to feel anxious. But quiet and invisible are different things, and we landed closer to invisible than I realized.

The silence threshold that submits a recording needs to be user-configurable, or at minimum adaptive. A setting that lets a slower thinker hold a longer pause before submission would fix the re-triggering problem I described in week three. This is a small change with a significant effect on the experience for people who speak the way I apparently speak when I am problem-solving out loud.

There is also no graceful path for when a capture fails in a noisy environment. Right now, you get a transcript that is either wrong or incomplete, and it lands in your list with no flag. I want a low-confidence indicator — something that says “this transcription might not be right, review before it goes on your list.” We have the signal; we are not surfacing it.

Finally, the onboarding does not teach the speaking syntax. It shows the app; it does not help you find your voice. I spent five days figuring out how to speak tasks effectively. A 90-second guided first capture — something that walks a new user through one complete task from voice to list — would shorten that curve considerably.

Would I recommend it to someone who is not me?

Yes, with a specific profile in mind. If you are someone who spends time in physical motion during the day — commuting on foot, walking between meetings, driving without passengers — Lunelo fits naturally into that motion. If you find yourself capturing the same thought three times because you wrote it down and then forgot where, voice capture with AI structuring solves that specific problem cleanly. If you are curious about the ADHD-friendly planner use case, the no-shame, no-streak design is genuine and it shows in daily use.

I would not recommend it as a replacement for a deeply customized Notion setup if that setup is genuinely working for you. I would also not recommend it as a primary tool if the majority of your planning happens in loud open offices or shared spaces where speaking aloud is socially inconvenient. And if you need robust calendar sync, multi-team task assignment, or project hierarchy, Lunelo is not that — it is a personal daily layer, and it makes no apology for the scope.

The best planner app is not a universal answer. It is a fit question. Lunelo fits a specific kind of person planning a specific kind of day. I know that because I am one of them, and also because thirty days of being my own test subject made the edges of that fit very clear.

Frequently asked

Is Lunelo actually free, or is free just the hook? The free tier includes voice input, AI task parsing, and the today and week views — the core planning loop costs nothing. Premium adds weekly insights, browsable history, and visual themes. The premium features are real additions, not artificial limitations on the free experience.

Does it work without an internet connection? The app is local-first, meaning your task list persists and remains accessible offline. Voice transcription requires a connection because it goes through Whisper. AI parsing requires a connection because it goes through Claude. The reading and editing of existing tasks works offline.

How does the AI know what to do with what I say? Your voice is transcribed by Whisper, then Claude reads the transcript and extracts the task, the time reference, and any relevant context. The structured result goes into your list. You can edit it manually afterward if the parse was off.

What happens to my voice recordings? They are processed for transcription and then not stored. The structured task data lives locally on your device. The privacy model was a deliberate design decision from the beginning.

Can I use it in languages other than English? Not currently. English is the only supported language. Multi-language support is on the roadmap, but I am not going to give a timeline I cannot stand behind.

Bottom line

Thirty days with a product you built is a strange experience. You see the seams you knew were there and a few you had convinced yourself were not. You also find the moments that make the whole thing feel worth it — the meeting capture that took nine words, the walk that ended with a clear head instead of a longer list.

Lunelo is not complete. No honest version of a v1 product is. The microphone indicator needs work. The silence threshold needs a setting. Noisy environments expose the limits of the input model. Those are real problems and I am not going to wave them away.

What I can say after thirty days is that the core idea holds. Voice-first planning is not a novelty. For the right person, in the right environment, it removes more friction than it creates. I still use it as my primary planner. That is either the highest possible endorsement or the deepest possible conflict of interest, and I will let you decide which.

If any of this sounds like the kind of planning tool you have been looking for, the trial is free and the onboarding takes about two minutes. You can start at lunelo.app.