For years, I’ve had a secret ambition tucked away somewhere near the back of my brain. It was to write a simple note-taking app—one that wouldn’t be overwhelmed with features and that would reflect my own mental filing system. In part, this yen stemmed from my dissatisfaction with existing notetakers. But I also saw the project as an adventure in software development that could only make me a smarter technology user.
Just one thing stopped me: The formidable technical knowledge required even just to get started. I’m not an utter programming neophyte, but my skills largely atrophied after I graduated from high school and never extended much beyond writing buggy games. Almost everything I’d need to know about modern coding I’d have to learn from scratch.
Or so it seemed.
Recently, however, a new wave of AI-infused tools with names such as Replit, Bolt, and Lovable has enabled a phenomenon that OpenAI cofounder Andrej Karpathy has dubbed vibecoding. It doesn’t involve coding an app yourself. Instead, you use a chatbot-like interface to tell an AI collaborator what you envision, and let it do the heavy lifting. You’re more product manager than programmer, and while a certain aptitude for technical matters is helpful, the barrier to building something is dramatically lower than in the past.
Using a Replit feature called Agent, I put together my dream notes app in a week, finding the process so addictive that I often tinkered into the wee hours. I gave my brainchild a name (Doolee, as in “duly noted”) and used ChatGPT to design a logomark (a pencil twisted into a lowercase d). Mostly, though, I simply told the Agent what I wanted, including features that occurred to me as I was overseeing the project. The web-based result runs on my iPhone, iPad, and Mac, and doesn’t feel much different than a native-app version might have. It requires a little more fit and finish before I can declare it complete, but I’m already smitten with it.
As a product team of one building an app with an intended user base of one, I aimed only to please myself. I’ve always loved sticky notes as a metaphor for note management—they’re informal, quick, and flexible. So I asked Replit’s Agent to make my app look like a searchable wall of them. It took just a few minutes to rough out a minimum-viable-product version. From there, I just kept tweaking and adding more capabilities, drawing inspiration from my favorite features in other notetakers I’ve used over the decades, from a 1990s DOS program called Info Select to Evernote to the one I’ve been using recently, Bear.
I had the Agent program features such as a search bar right at the top, a hashtag browser, and lists for task management and other purposes. I made it turn URLs into little cards that display page titles and source sites. I got it to sync notes back and forth between devices, including in scenarios where the app might not have access to the internet and would need to sync later. Even a week ago, I wouldn’t have guessed I could will something so professional-looking into existence.
What’s it like collaborating with a software engineer that happens to be a piece of software itself? Throughout the development effort, the Replit Agent almost always grasped my requests without me having to spell out every detail. Its first drafts of new features—written using web technologies, such as TypeScript and React, that are far beyond my ken—were often solid. When they weren’t, I provided feedback to nudge it in the right direction. It came off as calm and persistent, and often heaped praise on my feature requests (“That’s a fantastic idea!”) in a manner that was somehow synthetic and charming.
But as our collaboration progressed, it became clearer that the Agent doesn’t really think like a human. It couldn’t use the app it was constructing; verifying that everything worked was part of my job. At every step, the AI appeared to be puzzling out the project, as if it hadn’t been involved all along. Fortunately, it was a quick study.
I also learned not to trust the Agent too much. Whenever it finished debugging a problem area, it declared that work to have been a success, which it often wasn’t, especially at first. Weirder still, at one point, the Agent helpfully proposed adding a feature that would turn audio recordings into text. When I took it up on the offer, I saw no evidence that it followed through.
Even if the Agent proved overconfident and obtuse at times, the end result is an app I could never have produced on my own. Even if I’d hired a competent human programmer, I doubt that I’d have ended up with something that made me so happy so quickly.
Speaking of paying programmers: The basic free Replit plan might whet your appetite to the service’s possibilities, but you’ll probably need to spring for one of the paid tiers to tackle serious projects. I maxed out the $20-per-month one I signed up for pretty quickly and ended up investing almost $300 in producing Doolee. I will also be paying Replit fees to host my app, though they shouldn’t add up to a fortune as long as I’m the only user. Given how long I’ve craved building something like this, I don’t find the cost unreasonable.
Along with learning something about the highs and lows of AI-centric product development, I came away from this venture even more attuned to the ways productivity software in its conventional form can bog us down. With off-the-shelf apps, we’re at the mercy of design decisions we had nothing to do with. Most products are trying to please everybody, which leads to feature bloat. Anything with much history—Microsoft Word turns 42 this year—is likely to be particularly cluttered with cruft.
The tech industry’s conventional wisdom says that users typically ignore a huge percentage of the features in the software they use. (The exact figure cited varies, but Microsoft CEO Satya Nadella told me that Office users tend to utilize just 5% of its features.) The only way around this conundrum would be to create your own apps, built with only the features you want, implemented as you see fit. Until tools such as Replit came along, that would have been a pipe dream for most of us. Now it’s an everyday reality, albeit one that’s still slightly mind-bending. I can’t wait to see where it goes—and I hope to use my Doolee app for years to come.
You’ve been reading Plugged In, Fast Company‘s weekly tech newsletter from me, global technology editor Harry McCracken. If a friend or colleague forwarded this edition to you—or if you’re reading it on FastCompany.com—you can check out previous issues and sign up to get it yourself every Friday morning. I love hearing from you: Ping me at hmccracken@fastcompany.com with your feedback and ideas for future newsletters. I’m also on Bluesky, Mastodon, and Threads, and you can follow Plugged In on Flipboard.
Accedi per aggiungere un commento
Altri post in questo gruppo
President Donald Trump on Friday said is signing an executive order to
Your favorite iPhone could soon become much pricier, thanks to tariffs.
Most of us know the general (albeit simplified) story: Russian physiologist Ivan Pavlov used a stimulus—like a metronome—around the dogs he was studying, and soon, the hounds would start to saliva
AI tools are everywhere, changing the way we work, communicate, and even create. But which tools are actually useful? And how can users integrate
The way Bran Ferren sees it, the future of warfare depends as much on creativity as it does on raw firepower.
The former head of research and development at Walt Disney Imagineering—the
The nonstop cavalcade of announcements in the AI world has created a kind of reality distortion field. There is so much bu
