The Future of Software
The world is changing faster than it ever has before. And with that change comes a lot of excitement, opportunity, fear, and uncertainty. So I wanted to take some time to share my thesis on the future of software, arguably the industry sitting right at the precipice of this unprecedented change.
Jevons paradox
One of the first things I get asked is some version of this: AI is so good at writing software nowadays, doesn't that mean software will be free and engineers will be out of a job? My thoughts here are best summarized by Jevons paradox, which states:
"As technological advancements increase the efficiency with which a resource is used, total consumption of that resource often rises rather than falls."
The famous example is what happened after the invention of the ATM. Many assumed that once a machine could withdraw money automatically, bank tellers would soon become obsolete. The opposite happened: the number of bank tellers actually increased. ATMs lowered the operational cost for banks, which let them open new branches and service more customers. More branches meant more teller positions. Except this time, tellers didn't spend the majority of their day dispensing cash. They moved on to tasks like financial planning, answering questions about the bank's systems, and other work that machines couldn't do.
I believe the same will be true for software. AI will make software easier to create, which means more people will be able to build it, software will solve more problems in the world, and the market will meaningfully expand.
A more recent example makes the point even stronger. In 2021, IKEA rolled out a customer service chatbot named Billie (after the Billy bookcase, naturally). And it worked: Billie ended up handling 47% of all customer inquiries, that's roughly 3.2 million interactions a year that no human ever had to touch.
Now you might have guessed IKEA would have naturally cut the support team in half, reducing the bottom line. Instead, they did the opposite.
The team dug into the half of conversations Billie couldn't handle and noticed a pattern: customers weren't just asking about order status and returns, they wanted help with home planning and interior design. So instead of laying anyone off, IKEA retrained 8,500 call center agents into remote interior design advisers. That new design service pulled in €1.3 billion in revenue. Not a cost saving, but rather a brand new business line, built entirely out of the capacity the AI freed up.
And this is the real reason Jevons paradox holds. When you make people more productive, they don't pack up and go home early. They reinvest that freed-up capacity into doing more. The bank teller moved on to financial planning. The IKEA agent moved on to designing living rooms. Higher productivity doesn't shrink the work, it raises the ceiling on how much value can be created.
So how do you explain the "SaaS-pocalypse," where so many software stocks fell in value this past year?
I think software as an industry will expand. I'm not saying the winners will be today's incumbents. As I'll get into later, this seismic shift will benefit those who can adapt and stay flexible, so I believe the market is simply pricing that uncertainty in.
Side note: I do think the other explanations for the price drop hold less merit.
- Software charges per seat, so revenue will shrink as fewer people use the tools and more agents take over. But price is fundamentally a derivative of value. The way these companies price will change, and as long as they keep delivering value, they can keep charging for it. For example, one such new way of pricing is based on usage or even outcome delivered by the product.
- The value software provides will fall because it's now easier to build. If it's easier for you to build, it's also easier for the providers already in the market. They'll be more productive too, and a team focused on one problem will use that productivity to compound the value they deliver and ship even more than they do today.
The immutables: what won't change
"I very frequently get the question: 'What's going to change in the next 10 years?' I almost never get the question: 'What's not going to change?'"
- Jeff Bezos
In times of unprecedented change, I think this question matters more than ever. The truth is, no one knows what the future holds. I get it: it's far more exciting to debate grandiose, jarring predictions, and it's even comforting to latch onto whatever successful people say is coming, especially when it happens to benefit us. But I'd argue that focusing on and investing in the things that won't change is what helps you cut through the noise and build lasting value.
Data is paramount.
Data will always matter. If anything, its value grows, because AI is only as useful as the data it has access to. Software providers that collect and store useful data will stay valuable.
Someone needs to be responsible.
In business, doing the work is only half the equation. Being responsible for its delivery is the other half. At the end of the day, a person is still on the hook for prioritizing and completing tasks, whether they use AI or not. So if a team chooses to build something in-house with AI instead of buying an external service, they're still responsible for the outcome. The fundamental value of exchanging services therefore remains: (1) you offload the responsibility of maintaining a service, and (2) a provider focused on a single value prop will produce a more effective solution (see "Focus compounds" below).
The world is changing quickly. Staying flexible is vital.
The murkier the future, the more important it is to stay flexible. People have always been bad at predicting the future, and as the number of variables grows, it becomes far more productive to adjust and pivot quickly than to try to forecast accurately. Big technological shifts have always favored the disruptors: the internet gave birth to the software and social media giants, and the cloud revolution let new players like Salesforce reinvent entire industries. The same will happen with AI.
Focus compounds.
The more aligned and focused you are on a particular problem, the better you'll be at solving it. This sounds obvious, but I think people today are so frazzled by the possibilities of AI that they end up paralyzed by choice. What does this mean for software? Those who focus on solving a specific problem for a specific person will win out as the best solution. And tying it back to my earlier point about more people building software than ever: to be the best at solving a specific problem, you have to be a focused, verticalized service.
How will lasting impact be built? What are the new moats of software?
Now that the code itself is easier to build, what becomes a software business's moat? Here are the areas where I think differentiation and stickiness will live.
Data that's difficult to obtain
Anyone, whether a person or an agent, is only as effective as the quality of data they can access. So data that's hard to get, such as first-party data, becomes more important.
Business process
Data is what a company knows. Business process is how it acts on that knowledge. Over time, a good platform accumulates the operational logic of the businesses running on it: automations, workflows, and now, with AI, prompts and skill files. The best products recognize this and double down to make sure these processes run better on their platform than anywhere else.
Brand and taste
Even when you're selling to businesses, you're still selling to a person. People operate not just on logic but, more often than we admit, on emotion. As software gets easier to build, feature differentiation shrinks, and the decision comes down to how the customer feels when every competitor looks the same on paper.
Domain knowledge and expertise
Industries that demand deep domain expertise, or that have barriers to entry (like regulation), make it harder for new providers to break in.
Infrastructure and problems of scale
Problems of scale and maintaining critical infrastructure are still genuinely hard, especially as demand for agents grows. Solving for these infrastructural issues will remain valuable, as it not only requires deep domain knowledge, but also the high-level decision making of what architectural tradeoffs make the most sense for your particular customer demographic.
People
Perhaps the most underrated, and what I believe is the most important of them all. The quality of the value you provide is ultimately most correlated with the quality of your team. This becomes even more true with AI. The technology amplifies the output of the individual, which means the best people will have even higher leverage to do the best work.
The services industry
You may have heard the stat that for every $1 spent on software, $6 is spent on services. People believe AI is about to start disrupting the services industry, but what does that actually mean?
Services and software: the convergence
To start, software and services being intertwined isn't a new idea. Oracle and IBM make massive revenue from both the software they sell and the professional services they provide to help you use that software.
- People have even pointed out that because these legacy systems are so complicated and convoluted, the vendors can charge more for services, since customers are forced to lean on domain experts (but that's a whole other discussion).
That said, services and software have largely stayed separate as industries. I mostly attribute this to the high cost of managing both. It was expensive for software companies to also run a services arm, and expensive for service-based companies (like consulting firms) to learn how to use and build technology. But with AI improving efficiency and, most importantly, lowering the barrier to acquiring new knowledge, I think that line is about to blur.
Customers will expect software companies to also provide the implementation and deployment. And they'll expect services firms to build custom software solutions for them. The result? Businesses that do both.
Introducing: the Palantir model
I call this merging of software and services the "Palantir model."
Palantir's business model has always been a bit of a black box to most people. In essence, it's this: technical consultants who build a solution for you on top of their core data platform (Foundry).
Because Palantir offers both the strategy for how to solve a problem (service) and the means to solve it (software), their value proposition becomes simple: directly solving business challenges, end to end.
They popularized the role of the "Forward Deployed Engineer" (FDE): someone with a technical background who interfaces directly with customers to both scope out the work and build and deploy the solution.
To me, FDEs are just consultants rebranded. Since it's getting easier to learn and use software to solve problems, the consultants of the future will be expected to do exactly that.
So, bringing it all together, what do I think the future of services and software looks like? Businesses that can do it all, end to end: identify the pain point, come up with a solution, build it, and deploy it.
The consultant role will evolve into something closer to an FDE, and the value prop of these businesses will simply be: "give me a problem and I'll fix it."
Final note
These are my opinions as of June 1st, 2026. Recognizing that we're in a fast-changing world, I fully expect them to change just as fast as everything plays out. I'll leave you with a few bold claims I have about the future:
- Company sizes will shrink. Market share will be dominated by millions of small to mid-sized companies, each solving a very specific and unique problem.
- Software as an industry will grow 10 to 100x, thanks to how much more accessible it becomes to solve the world's problems.
- Service-based businesses and traditional software-as-a-product businesses will merge closer to one concept, becoming verticalized end to end "problem-solving" businesses
- The security industry will expand massively. Two reasons: (1) teams are moving faster to grab market share in this new world, leaving themselves more vulnerable to attacks, and (2) "security by anonymity" (avoiding attacks simply because you aren't well-known enough to be targeted) is far less valid now that agents make attackers more productive.