Silicon at the Bottom, Taste at the Top


The New Era of SaaS: Taste, Not Syntax

Before AI coding agents began to proliferate, coders were like scribes in the pre-Gutenberg era—trained in the esoteric art of transforming and preserving knowledge, and paid handsomely for it. The printing press demolished the scribes’ moat. Coding agents are doing the same to syntax. This does not mean software engineering is dead, but its mechanical foundation has been commoditized, and the moat has moved. The new moat is taste, not syntax.

Scribes didn’t disappear after Gutenberg; they became authors, editors, and publishers. Something similar is happening now. The people building software in this new era aren’t engineers in the traditional sense—they’re authors. Today’s software giants won’t die from this shift; the next generation of them will simply never be born. The story of what happens when authorship gets cheap has played out before: it ends with a creator economy, and it ends with the monolith moving somewhere you don’t expect.

Phase 1: The Hyperlocal Dawn (The Hobby Phase)

Software now starts at home. I’ve spun up multiple apps for my own needs and know plenty of friends doing the same. I host mine on an old laptop that stays plugged in, accessible on the local network. It serves my family perfectly well. It’s not a full-fledged app I can use on the go, and it doesn’t need to be.

I suspect a large population of enthusiasts is quietly doing exactly this: developing and self-hosting custom apps for themselves, their families, and high-trust circles. The drivers are mundane and powerful: software shaped exactly to your needs, with no subscription fees or monthly meters running. Most are hobby apps, spun up in an evening and discarded just as casually. Gone are the days when creating software required significant investment; today, a bespoke, fully functional tool is just one prompt and a Raspberry Pi away. LLMs have removed the last barrier to entry: the software itself.

Phase 2: The Software Creator Economy (The Influencer Phase)

Hyperlocal tools inevitably graduate. When your custom app works beautifully for your circle, word spreads—and what spreads isn’t the code, it’s the taste embedded in it.

The next logical step is a software creator economy: multiple apps for the same use case, with people choosing based on the sensibility of the creator, the way they choose Substack writers or YouTube channels today. Picture a fitness influencer who doesn’t write a line of code but ships a workout tracker tailored precisely to her training philosophy—her progressions, her deload logic, and her community’s shared leaderboards. Her subscribers don’t want the objectively “best” fitness app; they want her voice and her community’s shared context. Or think of your local barber, who no longer needs Square for bookings because a bespoke tool that matches exactly how his shop runs costs nothing to create.

The camera industry is a close precedent. Canon, Kodak, and Sony made capture devices cheap and ubiquitous, much as AI labs are democratizing software creation today. Yet the creator economy for video wasn’t born from cameras—it was born from YouTube: a distribution and monetization layer the camera vendors never dreamed of building. Cameras let anyone film; YouTube made some filmers into businesses. Coding agents let anyone build, but only a fraction will ever build for an audience: the hobbyists of Phase 1 are the creator class of Phase 2, and everyone else is their audience. The real question is: who will build the YouTube of software?

Because here is what the fitness influencer does not want to think about: uptime, security patches, payment processing, data migrations, App Store reviews, and what happens to 5,000 paying subscribers’ data when she loses interest and moves on. Coding agents nail functionality; yet the moment you productize, the unglamorous backend engineering becomes the entire game. Curation-as-moat implicitly puts the curator on the hook for operations, and no creator will accept that deal. Either a platform absorbs those obligations wholesale, or the software creator economy hits a quality ceiling and stays a curiosity.

AWS, GCP, and Azure technically solve these problems, but they are built for engineers, and their complexity is disqualifying for this audience. There is an opening for a low-complexity, no-hassle, turnkey hosting and monetization platform—Discord-like in its accessibility, YouTube-like in its economics. AI labs that control their own physical infrastructure hold a massive structural advantage in capturing this space. While they must still earn market trust on security and reliability, building the orchestration layer on top of their bare metal is ultimately a software problem—and solving software problems is exactly what these labs do best. I believe this is the next major revenue unlock for the industry, and if executed well, it makes these platforms the new gateway through which we interact with software.

The Infinite Software Tax

This creator economy has a physical cost, and the market is not pricing it in.

A single centralized app serving a billion users is a marvel of efficiency. Thousands of engineers spend years amortizing every byte and every cycle: shared caches, deduplicated storage, request batching, ruthless profiling. A million boutique apps serving a thousand users each deliver the same aggregate utility with none of that optimization. Each of them carries its own database, its own serving stack, its own redundant copies of the same solved problems. Fragmentation is, by construction, computationally wasteful—and we are about to fragment on purpose, at civilizational scale.

Call it the infinite software tax. Every app authored into existence claims its slice of compute, memory, storage, and network. Serving a CRUD app for a fitness community is a modest CPU and DRAM workload. But hosting a million of them multiplies that physical overhead by a million, with utilization far worse than any hyperscaler would tolerate internally. Even generously assuming platforms consolidate these apps onto shared multi-tenant infrastructure, the per-app overhead—isolation, state, idle capacity for spiky long-tail traffic—compounds into demand that simply did not exist when software was consolidated into a few hundred monoliths.

The market is transfixed by GPUs—and to be fair, authored apps will call models at runtime, so inference demand is real. But that is precisely the demand the market has already priced. The unpriced tax falls on everything serving these apps: general-purpose cores, commodity memory, flash, and interconnect. If the software creator economy materializes even partially, the second-order trade isn’t the model builders. It’s the boring silicon underneath the long tail.

Who Survives the Era of Infinite Software

None of this spells doom for existing consumer software companies. Several of their traditional moats remain entirely intact—and the smart ones will find that this new world hands them opportunities, not just competition.

Network effects remain as durable as ever. A workout app where your friends already hang out beats a marginally better one without them, and no coding agent can prompt a community into existence. Companies that own where people gather hold an asset that cheap software cannot replicate—and the biggest prize of the coming era, the platform layer where creator communities live, is a natural extension of exactly that strength.

Trust compounds the advantage. People will reasonably hesitate to put their finances in an app an influencer prompted into existence last Tuesday. Established brands have spent years earning credibility on security, reliability, and support, and in a world of a million unknown authors, that credibility becomes more valuable, not less. The companies that hold it are the obvious candidates to rent it out—becoming the infrastructure creators build on, the way YouTube lends its payment and content systems to filmmakers who could never build them alone.

Incumbents also get the same tools as everyone else. Nothing stops Strava from using agents to ship a richer, more personalized experience, and incumbents that lean into this will delight their users. There’s an honest division of labor here: incumbents excel at serving the broad middle at scale, while authored apps serve the specific philosophies that don’t average across a hundred million users. Personalization by the platform is a feature; personalization by an author is an identity. Both have room to win.

B2B plays out differently because the enterprise customer is becoming a builder. A business paying per-seat rent for a thin workflow tool can now author its own version over a weekend, shaped exactly to its process—build-versus-buy inverts for a large class of point solutions. What endures is software whose moat was never the code. Systems of record like Salesforce and Workday survive on compliance, data gravity, and institutional accountability. Businesses will still pay for accountability, but increasingly not for functionality alone, because functionality is exactly what just became free.

Today’s giants, consumer and enterprise alike, aren’t going anywhere. What changes is the pipeline behind them: the greenfield demand that would have consolidated into the next Salesforce or the next Instagram now gets absorbed—by authored apps on one side, and by software that businesses simply build for themselves on the other. And every one of those in-house tools, replicated a thousand times across a thousand companies, pays the infinite software tax in CPUs, memory, and storage—and runs an inference meter on top.

The Monolith Doesn’t Die—It Moves Up the Stack

Followed to its end, the YouTube analogy predicts something few people expect. YouTube didn’t fragment video into millions of self-hosted servers. It centralized distribution into one of the largest monoliths in history, sitting atop an infinitely fragmented sea of creators.

The same shape awaits software. Today’s billion-user apps will persist on the strength of their communities and their trust—but they are the last of their kind. The next great consumer software company will not be an app; new demand now fragments into authored, hyper-personalized, human-curated software before it can ever consolidate into a monolith. The monolith forms one layer down instead, in hosting, monetization, and trust. The giants of the last era sold software to everyone. The giants of the next era will sell the ability to make software to everyone, and quietly bill the world for the hardware underneath it.

The era of infinite software has a moat, and it isn’t syntax. It’s taste at the top of the stack, and silicon at the bottom.