Code is the Idea, Not its Execution
In 2005, Derek Sivers published a tiny framework that became a common way to think about the value of an idea. He argued that ideas are just a multiplier of execution. An app/business/SaaS idea on its own, without execution is basically worthless.
Fortunately for my sanity, Sivers' meme eventually caught on. People now, mostly, understand that "I've got a great idea for an app" isn't a valuable business, and doesn't justify investment, be that from VCs or technical cofounders. No one will pay you for your brilliant idea unless you've got execution, or a track record of execution, to back it up.
Sivers was right. But twenty years later, AI has fundamentally changed what we should include in the "execution" bucket. Working software is no longer part of execution. It's now just the codeification of the idea.
Working Software Used to be Part Of Execution
For decades, building software was hard, slow, and expensive (and building good software still can be). You needed engineers, months of development time, and/or significant capital just to get a working product into users' hands.
The software itself was a good chunk of the execution. Having the idea for "Uber but for dog walking" was trivial. Actually building a reliable, scalable, viable, well-designed app that handled payments, scheduling, GPS tracking, and notifications? That was the hard part. That was where the value lived.
This is why SaaS became a powerful business model. The pitch was compelling: we've done the hard work of building this complex software so you don't have to. Pay us monthly and you get the benefit of a product that you can't build yourself, for a fraction of the cost. You can even customise it or build on it to bring your "AirBnB for plant watering" to life.
The economics made sense. The cost of building software was high, so centralising that cost across many customers created enormous value. The SaaS company bore the engineering burden once; thousands of customers shared the benefit.
But Now... Software is Just an Idea in code
AI-driven software development has compressed the cost and time of building software dramatically. You can argue about by how much, and the quality question is real, but the direction is undeniable. What used to require a team of five engineers and six months can now be accomplished by one or two people in weeks. And even more worrying for software developers, the people building may not even need to be developers. In some cases, a functional prototype can be built in an afternoon.
The result is that software development is migrating from Sivers' "execution" column into his "idea" column. Building the app is becoming as cheap as having the idea for it (and a Claude Code subscription). A brilliant idea with a working prototype is still, in Sivers' framework, not much more than the original idea and cost of the burned tokens.
When the Cost of Production Collapses
This isn't the first time production costs have collapsed. Every time it happens, the same thing follows: value migrates away from production and toward adjacent layers.
The printing press destroyed the scribes' monopoly on book production. It didn't make books worthless, but it did shift value to authorship, editing, curation, and distribution. The people who could identify what was worth printing and get it to the readers captured the new value. It also changed the unit economics on what could end up committed to the paper.
Internet publishing made distribution free. As Clay Shirky put it,
"Publishing is not evolving. Publishing is going away. Because the word 'publishing' means a cadre of professionals who are taking on the incredible difficulty and complexity and expense of making something public. That's not a job anymore. That's a button."
When everyone could publish, value shifted to aggregators who solved discovery, trusted voices who solved credibility, and curators who solved the paralyzing paradox of choice.
Music recording followed the same arc. When home studios made recording cheap, the industry's value migrated from developing musicians and selling recordings to selling experiences and brands or controlling distribution through platforms like Spotify.
Joel Spolsky articulated the economic principle behind this pattern back in 2002:
"Demand for a product increases when the prices of its complements decrease."
Post is great and goes into a lot of other details. He talks about smart companies deliberately who purposefully commoditise their complements. When one layer of a value chain becomes cheap and modular, the adjacent layer captures the profit. Clayton Christensen called this the "Conservation of Attractive Profits" . Value doesn't disappear when a layer commoditises, it migrates.
I think we can see that AI is commoditising software production. The value is migrating. The question is, where to?
(side note, to think about this for your industry/business, I highly recommend exploring the ideas of Wardley Mapping)
The Traditional SaaS Squeeze
Traditional SaaS businesses are caught in a tricky spot, squeezed from all sides.
On one side, the cost of building bespoke software is collapsing.
A company that once had no choice but to buy a SaaS tool, as self build was too expensive, can now have AI build something tailored to their exact workflow. Not a toy prototype, but a functional, deployable application that does precisely what they need, without the configuration overhead, without the features they'll never use, without the compromises inherent in a one-size-fits-most product. And without the costly monthly (often per seat) cost.
On the other side, the overhead of being generic hasn't gone away.
SaaS products still need to serve multiple customers' use cases with a single product. That means configuration systems, integration layers, permission models, onboarding flows, and the endless work of making one solution fit many different contexts. This is the "configuration/complexity tax", and customers bear it in the form of workarounds, inflexibility, unused features, bugs, clunky UX, slowness, and time spent adapting their workflows to someone else's assumptions.
The SaaS value proposition used to be: "We built this complex thing so you don't have to." When building complex things becomes cheap, that proposition hollows out. Why pay for and adapt a generic solution when you can have something built to your exact specifications for a lower total cost of ownership?
This doesn't mean all SaaS dies overnight, of course not. But it means the shipped software itself, the code, the features, the UI, is no longer the moat it once was. Satya Nadella has said plainly that
"the era of SaaS as we know it is coming to an end."
Foundation Capital wrote about "Taking stock of the SAASpocalypse".
The software was part of the execution. Now it's just the codification of the idea.
What Execution Actually Means Now
If software is now part of the idea, what's the execution? Sivers' formula still holds, but we need to stop thinking about software as execution. Execution now means the hard, expensive, time-consuming things that make the idea actually work for businesses and users:
Distribution. Can you reach and acquire customers efficiently? Having a great product alone has never been sufficient. When everyone can have a great product, distribution becomes the primary battleground.
Compliance and trust infrastructure. This is perhaps the most underappreciated aspect of execution. A startup can vibe-code a competitor to your SaaS product in a weekend. It's a lot harder to vibe-code your way through a SOC 2 audit. The internet is rife with vibe-coded security horror stories. Your customers aren't just paying for your code, they're paying for the assurance that comes with your compliance posture, your security questionnaires, your vendor risk assessments, your insurance. This takes years of investment. It is a genuine moat, especially in B2B.
Network effects. Does your product get better with more users? A social platform, a marketplace, a collaborative tool etc. These create compounding value that cheap code alone can't replicate. You might be able to clone Figma's core features in a weekend. You cannot clone its community of designers sharing components, plugins, and templates without much more effort.
Data moats. Do you have proprietary data that makes your product better? AI can generate software, but the ten years of niche/industry specific transaction data that makes your fraud detection model accurate, or the millions of user interactions that trained your recommendation engine; that is much harder to hallucinate your way into.
Brand and trust. When anyone can build software, trust becomes the scarce resource. Who do you trust with your data? Who do you trust to be around in two years? Who do you trust to handle the edge case that breaks at 2am? Who do you trust to not use YOUR data against you or sell it to the highest bidder? Brand is accumulated trust, and that has real value.
Switching costs and integration depth. Being embedded in a customer's workflow, as a deeply integrated partner with their other systems, holding their historical data, trained into their team's muscle memory. Though this can be a shallow moat.
None of these things can be prompted into existence. They are built slowly, expensively, and deliberately. They are the execution.
What This Means for Builders
If you're building (or thinking about building) a SaaS product, ask yourself: if a competitor could replicate your software in a week using AI, what would you have left? If the answer is "not much," you don't have a business, you have an idea in code.
And like your non-technical friends who used to come to you with "a great idea for an app," don't expect investors or early employees to be as excited by your working code as you are. Working code + working business is where it's at. The code is table stakes now.
Kevin Kelly wrote about 1,000 True Fans — the idea that you only need a thousand deeply committed customers to build a sustainable business. In a world of cheap generic software, this cuts deeper than ever. The winning strategy isn't "build a SaaS that sort-of works for everyone." It's "understand a specific audience so deeply that you build exactly what they need, and own the relationship."
The value isn't in the software. It's in solving the valuable problem, knowing what to build, who to build it for, and having the trust, compliance, distribution, and relationships to deliver it.
Ideas were cheap. Now software is too.