Why “AI everywhere” made me build the opposite.
Every other day I see a new app shipped overnight: AI agents, vibe-coded dashboards, “just add ChatGPT” everything. I respect the speed. But it also made something obvious to me. The next wave of tools won’t win because they ship the fastest. They’ll win because people trust them enough to live inside them daily.
So I’m building Clean Core, a calm, local-first project workspace for small studios and solo creators, and I’m intentionally not leading with AI. Not because I’m anti-AI. Because my differentiation is on a different axis entirely: restraint, reliability, and craft.
A lot of modern “productivity” software has turned into endless features, endless notifications, endless integrations, and endless “smart” suggestions. It looks impressive in a demo, but it’s exhausting to actually use. Vibe-coding makes this worse in one specific way. It’s easier than ever to ship something that looks finished without building a tool someone can depend on day after day.
My plan to stand out in this wave is simple. First, restraint is the feature. Clean Core is designed to be a workspace you can open every day without feeling mentally taxed. Fewer moving parts. Fewer decisions per click. Fewer setups before you can actually work. I’m optimizing for clarity over cleverness.
Second, local-first is a trust contract. I’m building for the reality of small teams and creators where work happens at weird hours, internet is not always reliable, privacy matters, and “the tool is down” is unacceptable. Your projects should open instantly, work offline, and feel stable. That is the baseline.
Third, it’s designed by use, not by trend. I didn’t start Clean Core because it’s a good market. I started it because I needed it. I’ve used hundreds of tools, and the difference between “nice idea” software and “daily driver” software is that the builder actually lives in it. If I wouldn’t use a feature for eight hours a day, it doesn’t ship.
This isn’t vapor, either. Clean Core is already usable end-to-end as a product foundation. It has a dashboard and projects system, a project details modal and core UI flows, archive, settings, help, and stable navigation with a responsive layout. It’s not an idea. It’s a tool that’s becoming sharper through use.
Just as important is what I’m not doing. I’m not bolting an AI assistant onto a generic to-do list and calling it innovation. I’m not chasing a thousand integrations before the core experience is rock solid. I’m not optimizing for hype. I’m optimizing for a tool that feels inevitable once you try it. Speed is easy. Taste is hard. Trust is hardest.
AI is useful and I’m not pretending otherwise. But I think a lot of products use AI as the identity instead of using it as a tool. If AI shows up in Clean Core, it’ll be because it helps the user do something concrete, not because the market expects an “AI” label on the homepage.
I want to pressure-test this with people who actually ship. If you were a solo creator or a 1–5 person studio, what would make you trust a brand-new project/workspace tool enough to switch? Is it offline reliability, simplicity, import paths, pricing, something else? If this resonates and you want to follow along, I’m collecting early-access emails and I’m happy to DM a link or a demo clip.
Totally get the point about reliability and trust. It’s easy to ship something fast with AI, but keeping a system stable day after day is another challenge. How are folks balancing ship‑fast with ship‑solid in their AI projects?
When looking at a tool, I am searching for "what problem does it solve for m?e" If I can't find enough convincing evidence from a website, maybe a lack of reviews or actual testimonials of real usage. I may be turned away; I factor in one's reputation before I commit to purchase or use.
Another implicit meaning of this article is a return to one's roots. We should recognize AI had accerlerated the process of developement process of building new software. But the core of the people didn't change. We still need something useful to us, AI products that are not simply innovative for the sake of innovation. AI bring us a lot of effiencicy and we can get the experience that we cannot get before the AI come. But people's core needs haven't changed.
yeah i've used so many tools that demo well but feel exhausting after a week.
also, for me it's offline reliability + fast startup. sounds like you're nailing that part
i really like the way u worded thisb
I think people may eventually hit AI fatigue, especially with tools that use AI where it doesn't actually add value but is just a nice-to-have. It's not about slapping AI on every product; it's about making products useful to customers
"Designed by use, not by trend" — this hits home.
I'm building a feedback exchange app for indie developers, and I started it
for the exact same reason: I needed it myself. Tried Reddit, Twitter, Discord
for feedback on my side projects. Sometimes got lucky, mostly got silence or
generic "looks cool" comments.
Your point about "daily driver" vs "nice idea" software is so true. The gap is
huge and mostly invisible — it's in the small frictions that add up over
time.
For your question about what makes someone trust a new tool: for me it's "does
this respect my time?" If I open an app and it asks me to connect 5 accounts,
watch a tutorial, or set up integrations before I can do anything — I'm
already looking for the exit. Your "fewer decisions per click" philosophy is
exactly right.
Will keep an eye on Clean Core. The local-first + restraint combo is rare.
I feel sometimes people think more in terms of tech than starting with the product. It reminds me of what Steve Jobs said about starting with people and going backwards to the tech
I think it is also a matter of how you use the AI. For me building with AI it is all about the planning phase. To make the plan and ground work in a way so that the build will be constant and the code does not end up as one pool of spaghetti.
"Restraint is the feature" — this is the hardest sell in software, because it's invisible. Nobody screenshots the feature you didn't add.
I'm building an AI-powered tech news aggregator, and the temptation to add every "smart" feature is constant. What I've learned: users don't want AI that does more. They want AI that helps them do less — less searching, less context-switching, less deciding what to read. The restraint isn't anti-tech; it's pro-user.
Your "designed by use, not by trend" point resonates. The gap between "demo-impressive" and "daily-driver" is enormous, and it's almost entirely about the boring stuff — load time, reliability, not breaking your flow.
Two things I'd push back on gently:
The switching cost question is real but incomplete. You asked what would make someone switch. But the harder question is: what makes someone stay after week 2? The honeymoon phase ends fast. Import paths get you in the door. Habit formation keeps you there. How are you thinking about that transition?
"Local-first" needs a clear answer for data portability. People who care about local-first also care about not being locked in. If I put my projects in Clean Core and decide to leave in 6 months, what does that look like? Plain files? Export to standard formats? This is the trust signal that matters most to the audience you're targeting.
Curious — what's your current daily usage pattern with it? How many hours/day are you actually living in it?
To answer your question directly: for me to switch into a new workspace tool, it has to nail the first 30 seconds. Open, see my stuff, start working. No onboarding wizard, no "connect your accounts" step.
I ran a multi-tenant SaaS for grocery retailers for years. The tools that stuck for our team were always the ones with the fewest decisions per session. Not the most powerful ones. Your "fewer decisions per click" line is exactly right.
On the local-first piece - one thing worth thinking about early is how you handle the transition when someone does eventually want to collaborate. ItsKondrat raised this too. CRDTs (like Automerge or Yjs) let you stay local-first while adding sync later without rewriting the data layer. Might be worth baking that assumption in now even if collaboration ships later.
Curious what your tech stack looks like under the hood. The "designed by use, not by trend" philosophy tends to produce very different architecture choices than "what's popular on Hacker News this week."
This is a refreshing take. It's interesting that you're going against the flow. I agree that AI can be a useful tool that can bring value, but it doesnt necessarily have to be a selling point.
Being a newly self-employed remote worker, I would definitely be interested in trying Clean Core, and I'd be happy to provide some UX reviews if that could help you.
The "local-first is a trust contract" framing is spot on. I've been building side projects with AI tools (vibe coding) on weekends, and the biggest friction isn't the building — it's the context-switching between 15 different "productivity" apps that each want my attention.
Would love to try Clean Core. What's your thinking on collaboration? One thing I struggle with as a solo builder is that local-first often means "works great alone, breaks when you need to share." Is that in scope or intentionally out?