View / The ease of vibe coding could upend the economics of scarcity and scale

Recent remarkable and rapid improvements in vibe coding — using AI systems to write programs — are upending Silicon Valley’s balance of power, away from talented developers and towards startup founders with a good idea.

But they will also remake the economics of scale and the corporate processes built around husbanding and prioritizing scarce tech resources.

As my colleague Reed Albergotti wrote this week, vibe coding shifts us from a model where a small number of developers build software for millions — or billions — of users to one where millions — or billions — or people can write software for themselves, or just a few other users. If development is dirt cheap, products don’t need to have enormous value or massive reach. Where does that leave companies that have optimized themselves around leveraging until-now scarce tech talent to build products at scale?

An epoch ago — back in the 2010s — one of my duties was to lead tech development for the Reuters’ sprawling 2,500-person newsroom, where we had a dedicated team of developers building tools for news gathering and publishing. The work wasn’t glamorous. It consisted mostly of pouring over detailed product road maps, making tedious business cases to prioritize one feature over another, and painstakingly gathering user requirements. And all that made sense: for all of Reuters’ scale, there were only so many developers available, and we had to make sure we made the best use of a limited number of technologists’ time to benefit the most people.

That’s at the core of what any company that builds digital products does. They’ve built entire — sometimes-stifling, sometimes creativity killing — “product management” bureaucracies around how to ensure not a second of developer time is wasted. Those imperatives also drove businesses to prioritize building products that would deliver the most scale or the most profits.

What happens when the cost of building digital products collapses? When coding skills are no longer concentrated in the hands of a few, but are widely distributed to people who have never heard of JavaScript, Python or JSON? How will companies have to restructure and reorganize to take advantage of this brave new world?

Back in the before times, at Reuters, we did not build tools on a whim.

Now at Semafor, where we have a 40-person newsroom, one other journalist and I are busy vibe coding a set of tools for our colleagues to use. We have a copy editing bot that checks stories against our house style; a widget that suggests headlines; a program that can read the story you’re working on and find related content across the web, in multiple languages; a tool that will take a story and extract what it asserts as facts, what the writer’s analysis or assumptions are, and how the issue or event is framed; and even one that predicts angry tweets. 

Neither of us are technologists. We spent less than half a day, with another colleague — who also isn’t a technologist — strategizing over the architecture of our tools so we don’t step on each other’s toes. Other than that, we just code. In our spare time. In between our day jobs.

Yesterday I made two new tools. One on a whim. (My colleague made another two.) I spent zero time gathering requirements, didn’t make a road map, and decided on my own which features might be cool to have. I’m not sure how many of my colleagues will find it useful, but I don’t need the whole newsroom to love it; even if it turns out that it’s only really valuable to just two or three people, it may turn out to have been a great use of my time.

The point is, we may no longer need to make products that scale. Building these bots was so simple that they could pay back the minimal investment of my time by helping just a handful of people. It was cheaper, faster, more efficient to build something, and then see who might want to use it (or scrap it if no one liked it) than embarking on the much more ponderous process I was used to — that all large organizations are used to.

And that means that vibe coding — as with any technology that suddenly shifts a good from scarcity to plenty — has the potential to upturn much of the core calculus, thinking and economics that underpin how and what companies currently do and build. It may suddenly become much more economically feasible to make highly customized products for thousands of people, not millions; it may be cheaper to build, test and discard tools that it is to do painstaking market research first. The idea that you need to chase scale may fade.

To be sure, there’s plenty that can go wrong in this scenario, not least security flaws, poorly coordinated architecture and badly designed systems. Vibe coding, like many other generative AI capabilities, is already surprisingly good, but will need to improve to be truly ready for prime time. And there will continue to be a need to oversee and manage — and roll out, and market, and monetize — products. Making things is only one small part of what companies do.

But a core part of what many companies have done — hoard and apportion tech talent and time — may be in for a real shakeup, and soon.