I Can't Code, but We Built Four Tools: Notes on Vibe Coding
The short version: knowing how to code is no longer the barrier to making a tool. Being able to explain clearly is. I can't write any code myself. All four tools on INTENTION Made started with us at INTENTION (識聚有限公司) talking and describing what we wanted, having AI write it, then revising one version after another until we used them every day ourselves.
This piece covers three things: what vibe coding actually is, how we do it, and where I think it's going next.
1. What vibe coding is, and where it stands now
The name is new. An AI researcher tossed out the term online on February 2, 2025: go entirely with the feel of it and forget the code even exists [1]. In November of that year, Collins Dictionary named it Word of the Year, defined as "using natural language to tell AI to write code" [2].
A lot of people use it. In March 2025, in the then-current batch of a well-known startup accelerator, a quarter of the startups had 95% of their code written by AI [3]. In a 2025 developer survey, 84% were using or planning to use AI to write code [4]. In 2026, 73% of people using AI tools use them every day [5].
But it isn't a cure-all. In the same 2025 survey, 72% of developers said they don't vibe code at work [4]. Of the code AI writes, 45% carries security vulnerabilities [6]. A randomized controlled trial found that experienced engineers using AI were actually 19% slower, yet afterward felt they had been 20% faster [7].
My reading: vibe coding has made "making something" very cheap. It hasn't made "making something right" cheap.
2. How we do it (all four tools came about this way)
The moment we got stuck | The tool we built |
|---|---|
Finally found a reference image, saved it, and got a blurry thumbnail | Fully: paste a link, and images and videos save at original size |
Nobody wrote up the meeting, and no one could remember who said they'd do what | Hearby: tap it before the meeting and you have notes when it ends, every line timestamped (Mac and Windows) |
I could say the idea out loud, but my typing couldn't keep up | Talky: type by talking, and when you're done it pastes at your cursor |
The portfolio was too big to send | Slimly: shrinks it enough to send, with almost no visible difference in quality |
There are only five steps, and the order matters:
First, write "that moment" as one sentence. For example: "That one is a thumbnail. This one is the file." If you can't get it into one sentence, you usually haven't worked out what you're trying to solve.
Say clearly what you want, and also what you don't. Fully had it written down from the start: no grabbing paid content, no touching protected videos. The "slim" in Slimly only ever refers to files, never to bodies. What you don't want matters as much as what you do.
Use it yourself first. Every one of these tools is something we use daily. You only find out what's wrong by using it.
If it's wrong, redo it. For Talky's first interface, we just borrowed Hearby's look. Only after using it did we notice that while it was "organizing," you couldn't tell what it was doing or how much longer it would take. In September we rebuilt the interface so that at every moment it can answer: what's happening now, how far along it is, and how much longer.
Once it works well, make all of it public. All four tools are free and open source, with the code on GitHub (github.com/intentionltd888). Open source means fully open, not halfway: we say "your data isn't uploaded," so the code is right there for anyone to check.
3. How I see it
First, the barrier has moved from "can write code" to "can explain clearly." I often tell new team members that understanding logic isn't the same as knowing how to code. It's three moves: breaking down (splitting a big job into small steps you can check), describing (putting the standard in your head into words), and checking (knowing what "good" looks like). AI can only get as far as you can explain clearly. A prompt isn't a magic spell. It's a brief.
Second, small teams will punch further and further above their weight. In 2026, one software company with 146 people added US$100 million in revenue in a single month [8], and another reached US$100 million in annual revenue with 50 people [9]. Being small is no longer a reason you can't build things.
Third, most of what gets built will be thrown away, and that's a good thing. A research firm predicts that by 2028, 60% of apps built with vibe coding will be retired because they don't deliver lasting value [10]. I agree, and I think that's how it should be. Most tools are disposable: use it, then toss it. The ones that stay are the few that actually solve your own problem. Our four tools have stayed because we use them every day.
Fourth, we aren't a software company, and we don't plan to compete with software companies on who's more useful. We build tools for only two reasons: we got genuinely stuck, or it seemed fun. Making something with a point of view that we actually want to use matters more to us than building the most powerful features.
4. My predictions
Our company keeps a prediction ledger. Each entry states its due date, its probability, and what counts as right, and we settle it when it comes due. Our in-house trend researcher estimates the probabilities. The ones related to this piece:
Big companies will build this kind of feature into their own products. Prediction: by March 2027, at least two of the three major AI services launch "voice input plus automatic polishing" that works in any app. The ledger puts it at 30%. If it comes true, tools like Talky will be up against the giants directly.
But small tools like these won't disappear. Prediction: by the end of 2027, at least one voice-input or meeting-notes startup closes another funding round or gets acquired. The ledger puts it at 93%. In 2026 alone, one voice-input startup raised US$280 million at a US$2 billion valuation [11], and one meeting-notes startup was valued at US$1.5 billion [12].
My call: within two years, "can you code" will no longer be the barrier to building tools. "Can you explain it clearly, and can you check the result" will be. At that point the contest isn't about who can code. It's about who knows what they want.
5. For creators who want to try
Don't start by thinking "I'm going to build an app." Take a sheet of paper, write down the three small things that annoyed you most this week, pick the smallest one, and try to say it clearly in one sentence. That sentence is the start of your first tool.
Once it's built, remember to use it yourself for a month.
The tools we've built are all on INTENTION Made, free to download: https://www.intention.ltd/made
Sources
Andrej Karpathy, post on X, 2025-02-02. https://x.com/karpathy/status/1886192184808149383
Collins Dictionary, Collins Word of the Year 2025, 2025-11-06. https://blog.collinsdictionary.com/language-lovers/collins-word-of-the-year-2025-ai-meets-authenticity-as-society-shifts/
TechCrunch, A quarter of startups in YC's current cohort have codebases that are almost entirely AI-generated, 2025-03-06. https://techcrunch.com/2025/03/06/a-quarter-of-startups-in-ycs-current-cohort-have-codebases-that-are-almost-entirely-ai-generated
Stack Overflow Developer Survey 2025 (AI section), 2025-07. https://survey.stackoverflow.co/2025/ai
Stack Overflow, The results of the 2026 Developer Survey are here, 2026-10-06. https://stackoverflow.blog/2026/10/06/the-results-of-the-2026-developer-survey-are-here
Veracode, 2025 GenAI Code Security Report, 2025-07-30. https://www.veracode.com/press-release/ai-generated-code-poses-major-security-risks-in-nearly-half-of-all-development-tasks-veracode-research-reveals/
METR, Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity, 2025-07-10. https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/
TechCrunch, Lovable says it added $100M in revenue last month alone with just 146 employees, 2026-03-11. https://techcrunch.com/2026/03/11/lovable-says-it-added-100m-in-revenue-last-month-alone-with-just-146-employees/
SaaStr, How Gamma hit $100M ARR with a team of 50, 2026-07-29. https://www.saastr.com/how-gamma-hit-100m-arr-with-a-team-of-50-grant-lees-4-lessons-at-saastr-ai-2026/
Gartner (Nitish Tyagi), Govern Vibe Coding for Citizen Developers With Self-Service Platforms (licensed reprint), 2026-06-29. https://tray.ai/guides/gartner-govern-vibe-coding/
TechCrunch, Wispr raises $280M at $2B valuation, 2026-08-17. https://techcrunch.com/2026/08/17/wispr-raises-280m-at-2b-valuation-as-it-looks-beyond-dictation/
TechCrunch, Granola raises $125M, hits $1.5B valuation, 2026-03-25. https://techcrunch.com/2026/03/25/granola-raises-125m-hits-1-5b-valuation-as-it-expands-from-meeting-notetaker-to-enterprise-ai-app/
About the author
陳韋傑 (Aloha jie) is the founder of INTENTION (識聚有限公司, INTENTION CREATIVE CO., LTD.) and a director. INTENTION is a creative strategy and visual production company in Taipei, founded in 2022, working on album visuals, music videos, brand films and social content. Production is handled by its production arm NOUS ASIA (冉冉創意股份有限公司). INTENTION Made is where the team turns the friction in its own creative workflow into free tools.
留言