The things you can't vibe-code

I sat through a panel at a travel media conference last year where an AI expert spent his presentation explaining that AI wasn't actually 'intelligent' and couldn't even generate a picture of a clock correctly. I left thinking he'd steered around the actual effects of LLM adoption in order to make a room full of people feel a little better.

Twelve months later, the distance between "I wish I could build X" and "I built X" has never been narrower. (I've built a handful of things myself this year with AI tools.) A lot of other people are doing the same thing, in the outdoor industry and beyond.

How many "I built a thing to replace feature X from Strava" projects have launched in the last few months? Dozens? Hundreds? I built my own little training plan app in an afternoon by feeding in a CSV of my Strava data. It told me I was severely undertrained for a mountain ultra because I'm not getting enough elevation in London. (Fair.) Strava apps are only one example; it’s happening across nearly everything related to the outdoors, sport, and travel (in addition to other consumer categories).

Calorie tracking apps, route planners, all-in-one travel planners, coaching platforms, segment analyzers, social fitness alternatives, recovery dashboards, AI training plan generators. These are all popping up (seemingly daily) from solo-builders and new startups. Anything map-based has a similar story. Most map-centric apps sit on roughly the same underlying OSM and trail data; what differentiates them is the interface, community, and distribution, not the data layer. Same for anything built on publicly available APIs like weather, snow conditions, fire data, and air quality. If your product's primary value is "we pulled public data and put a nice interface on it," it’s probably cloneable in a weekend.

It's not just hobbyists and vibe coding. Engineers at your favorite brand are running coding agents every day. Marketing leads are drafting subject lines and ad copy. Brand teams are running research. Enthusiasts are shipping apps over a weekend. Operations teams are building one-off internal tools that a year ago would have sat in a backlog for a year. RunSignup, the race registration platform, built out an entire vibe-coding webinar series for race directors, teaching them to build custom apps with RunSignup's public API.

Now, most of these little apps will fail as businesses, because statistically, most businesses fail. But it’s also because making something and posting it on social media is not the hard part of building a product. Distribution is hard. Scaling and sustaining infrastructure is hard. Customer support is hard. Security and privacy are hard. Keeping the lights on past an initial wave of novelty is hard. Many of the "problems" people are breathlessly solving are things the community has magnified far beyond the actual addressable market.

But some of them won't fail. Some small team will find a niche an incumbent didn't care about, build something genuinely better, get loud on TikTok or in the right Reddit community, and grow from there. Take Cal AI. Two teenagers built an AI calorie-tracking app that logs meals from a photo, hit $30M in revenue in under two years, and then got acquired by MyFitnessPal last December — by all accounts mostly for the userbase; there’s nothing particularly special about their underlying technology.

And many of them won't need to grow. Len Necefer recently shipped the Expedition Power Planner, an iOS app that models battery, solar, and power station loads for backcountry documentary shoots. I spoke to one tour operator who has been building internal tools for things like trip cost estimation so they can answer customer questions faster while on the phone. PackTrak will never make money, but I spent a few hours last weekend fixing bugs and adding features because a user from Germany emailed me some feedback. There's going to be more and more of this, not less.

When code and content can be generated in a few keystrokes, the things that matter are the things you can't generate. Things like brand. Community. Trust. Proprietary data. Real relationships with the people who use the product. Physical infrastructure you can't vibe-code around.

Network effects aren’t going away. Strava isn't competing purely on the quality or depth of its training analysis. It has 150 million users and a social graph that's nearly impossible to build from scratch, which is why anyone who's tried a Strava clone for more than a week feels the absence of their friends and drifts back. Similarly, the core Instagram experience is not particularly complicated. But the years are littered with a graveyard of over-hyped "Instagram alternatives" that promise to bring a chronological feed that "everyone wants" and no ads.

Proprietary data is more important than ever. A coding agent can build a booking engine in a long weekend. It can't fabricate a decade of conversion data, pricing history, or guest behavior analytics. You can build a trails app, but it’ll take something very special to compete against years of photos, reviews, and user activity.

The advantages for incumbents are significant. If anything, larger companies are in a position to use this moment well, either by acquiring the small, fast-growing competitors who prove out a niche or by actually using new tooling internally to ship faster and clean up the parts of their products that have been embarrassing for years. That said, in a subscription-fatigued world, a new challenger that’s 70-80% as good but at lower price point can still eat into incumbents market share.

Relationships are a kind of proprietary data of their own. I've owned findoutdoorjobs.com for almost a decade at this point. I initially bought it because I thought outdoorindustryjobs.com was such a poor experience. I could spin up a full-featured jobs site over the weekend (who knows, I still might), but that's a far cry from building the relationships with brands and hiring managers that would make it sustainable and useful.

You have to make people actually care about your brand. That means having a voice, telling stories, being a company with recognizable personality attached to it rather than a faceless PE operation. This has always mattered, but in a world where software is increasingly generic and content is being churned out by bots, the companies with a real point of view and human perspective are going to stand out more. Yes, you can make your own snow forecast app. But why should I trust you over Joel at OpenSnow?

There are interesting versions of this in adjacent industries. Karri Saarinen, CEO of Linear, is a public voice on design and product strategy, and the company's worldview is visible in the product itself. Tyler Denk, CEO of Beehiiv, writes a personal newsletter to 120k readers about his life and the business. Rafat Ali, CEO of Skift, s a loud, opinionated voice on travel, in his editorial and on LinkedIn. Each of them provides a more human lens into what the company thinks and how it operates.

Right now, I’m not sure I could say the same about most of the outdoor industry. Patagonia has a clear voice, but not many others come to mind with strong POVs or approachable leaders. This doesn’t mean that every CEO needs to be doing FDR-style fireside chats, but everyone should at least be thinking about how to be more human. I'm empathetic to how hard it is to 'speak' as a company, or as a person in a visible leadership role. It can also go sideways (as we saw with the McDonald's CEO). Recent history shows that the outdoor industry can be particularly reactive and susceptible to groupthink, leaving little room for “mistakes”. But there's still room for brands to differentiate here, and the bar feels low.

A year from now, there will probably be more of all of this. More tiny apps. More hobbyist tools. More Cal AIs. More hot takes about "everyone making their own personal software." A lot of bad products and a few genuinely useful ones.

But you can't ask Claude to "make a website to visualize outdoor recreation data" and end up with data.hereandthere.club. What's in it still depends on a pile of product and design experience, opinions, context, and decisions that don't exist in any training data. The challenge is building companies and products that still matter when anyone can build software.