The Moat You Sell Is Disappearing
Software services after software price collapse
The best of your engineers might be today 40% more productive with AI coding tools. At least that’s what this study says. Given it was published in October 2025 - in today’s timeline, roughly a decade ago - it is already outdated. My assessment would be that it is more than that, at least for the best performers and a certain range of software products. (Now, how much of an individual productivity gain translates to the organization is an entirely different story!)
This creates an interesting dynamic for software service companies — companies that build custom software for their clients.
Right now, companies that adopt AI aggressively feel like they have an edge. And they do — today. But this is a transitory advantage. Whether it takes one, two, or five years, the market will eventually catch up. AI proficiency will become the new baseline, not a differentiator.
An opportunity and a threat
Software service companies are living through this transition in two very different ways.
Some see it as an existential threat — and not without reason. Customers can now create way more software themselves — software that they used to subcontract to vendors. The barrier to entry for building a functional application has collapsed. If your client’s PM with one engineering colleague can spin up a working prototype over a few days, they are starting to question whether they need an external partner at all.
Sure, doing a limited POC and a real, production project, with real load, security constraints, and ongoing maintenance, are two very, very different things. Many of these ad-hoc in-house products will eventually fail to deliver the promised value.
And yet, it would be dishonest not to notice that building relatively typical software has become way more accessible. And let’s be straight — not every custom-built product was a complex, multi-year endeavor. You can check it out, for instance, on Clutch by browsing the typical price ranges. There are plenty of companies living off $10-100k contracts.
Others see it as a tremendous opportunity. With outstanding productivity gains, they can harvest disproportionate margins — doing more with less, delivering faster, and capturing the spread between the old cost structure and the new one.
Here is the problem, though. Your client knows this. They don’t need to master AI or even use it effectively themselves. The mere existence of AI-augmented development gives them leverage to push for lower prices. A recent CIO analysis found that enterprise customers are already demanding their share of AI productivity gains. The logic is straightforward: if developers complete tasks 40% faster, someone is capturing that surplus. The client wants to know why it isn’t them. And the traditional saying goes: the customer is always right…
A question of price
Where does this put software service companies that typically operated in a time-and-materials manner? On the one hand, customers expect faster delivery (fewer man-days). On the other hand, the agency that truly masters AI-augmented software delivery cannot reap the efficiency benefits (as they charge per day).
The first instinct is to move to fixed pricing. You capture the efficiency, the client gets a predictable cost, everybody wins. Except the prices will fall down too - in the same way that expected man-days for a project would fall down (though arguably the price arbitrage will be there for a limited time, and companies that are ahead-of-the-curve should be able to capture part of it while it lasts).
The smarter companies are moving further along the spectrum — from time-and-materials through fixed price to value-based pricing.
What is value-based pricing? At its essence, you charge for what it’s worth to the buyer, not what it costs you to produce. It might be capturing a fraction of the savings that delivering specific software produces for the customer. Or it might be a price set based on new opportunities uncovered for the customer.
McKinsey already generates 25% of its global fees from outcomes-based arrangements. You may be skeptical about McKinsey. I am not surprised. One thing is certain, though - they know how to make money, whether we like it or not. And when they overhaul how they charge, it is at least prudent to pay attention.
To quote Business Insider: “The result is a renewed industry focus on implementation and value over mere activity.” Part of me shouts: At long fucking last!
Unfortunately, value-based pricing is an entirely different and complicated game. And on top of that, moving to value-based pricing for a general software vendor is going to be even harder. Except for very specific, narrow technical cases (e.g., cloud cost optimization), a lot of value is entangled in the customer’s business. It is going to be very difficult to move towards that pricing for software service companies that operate exclusively at the technology layer.
Whichever way you go, the same pressure applies — building general-purpose software has become more accessible, and software service companies will face pressure.
Where does the moat go?
Forbes predicts that within three to five years, a software development productivity boost will exceed 80%, meaning half the labor currently needed will suffice. Now, I am very much on the side of Jevons Paradox: a software Cambrian explosion will drive the cost down, which will in turn drive the demand up.
However, there is an intermediate problem. Most enterprise customers won’t be ready to absorb a radically higher volume of delivered software. They grew their processes in a reality where building software was the bottleneck. Now, the bottleneck has shifted, and many enterprises are just not ready for consuming software at a faster pace.
Whether the demand explodes or not (and when!), the immediate reality is the same. An industry built on billing for labor will feel every point of upcoming compression. The agencies that thrive won’t be the ones that just write the most code.
The work won’t disappear — but the margins for general software-service companies will. And who wants to end up on a commodity-price treadmill?
So what’s defensible? Three things, as far as I can tell.
First, go where the barriers are structural. Regulated industries — finance, healthcare, defence. Hardware integration. Environments where certification, compliance, and safety aren’t nice-to-haves but legal requirements. AI can write the code, but it can’t navigate a regulatory audit, own a liability, or sign off on a medical device. At least — not yet.
Second, go deep into a niche. Not necessarily a regulated one — but one where accumulated expertise takes years to build. Real-time trading systems. Logistics optimization. Industrial automation. Developer’s efficiency. Insurtech. The agency that has spent a decade understanding how a specific domain actually works will beat the one with better engineers every time. This was always a good bet. AI makes it even more relevant because when the code price tag collapses, the only thing left to charge for is knowing what the code should do better than your customer.
Third — and this is where I think the real shift happens — close the gap between software and business. The “code monkey” role — take a spec, produce code — is the one most directly displaced by AI. Outside of highly specialized areas like critical infrastructure or security, the direction is convergence. Software engineering stops being a separate discipline and becomes part of how business operates. The agency of the future doesn’t just build what you spec. It helps you figure out what to build. At some point, that’s not a software agency anymore. It’s a consulting firm that happens to ship code. Coincidentally, value-based pricing is way more achievable for vendors being closer to the business (or even better — closer to an actual P&L).
One thing seems to be certain - the days of straightforward selling of pure, general technical experience are gone. Act accordingly.