The Diagnosis
SaaS (software as a service), the categorical Silicon Valley darling, is about to go through a giant upheaval. Dev jobs are either being offshored or companies are hiring via contract only, and the glory days of getting paid 6 figs for an overpaid CRUD engineering gig will be a thing of the past. As a recent example, Salesforce notoriously decided not to hire any Software Engineers this year. See article here.
Companies are now starting to realize their only real moat was sales connections or tech execution. While these were once formidable advantages, AI-powered tools are making both increasingly accessible to newcomers with minimal investment. What once required a team of skilled engineers can increasingly be accomplished by a smaller team using AI tools. This is leading to widespread labor commoditization - when any feature can be easily copied and implemented using AI, the value of individual engineering talent naturally decreases.
Even PG is getting blown away.
I talked to the CEO of a moderately big tech co who said they'd replaced Figma with Replit. This surprised me because I don't even think of them as being in the same business. But he said Replit is so good at generating apps that they just go straight to prototype now.
— Paul Graham (@paulg) January 18, 2025
The Decline
Unfortunately, there is no sign of this slowing down. If you check out this FRED graph here it shows the number of software development job postings on Indeed -- presumably with the blow up due to the pandemic, but now the numbers are below pre-pandemic levels.
This is a mark of not only the menial software engineering role becoming less valuable, but the tech sector as a whole. It turns out after all the years of middle managers telling their manager that the bottleneck was an engineering shortage, the real issue was actual value creating ideas and the risk appetite to execute them. While market factors such as the economic impact of COVID-19 and shifts in Federal Reserve interest rates have also played a role, they actually bolster this case, as the number of tech jobs has not fully recovered. This suggests that the underlying issue is not just cyclical, but structural, driven by the demand to increase shareholder value and utilize AI to do so - reducing the bloated engineering workforce.
We're seeing this proven by a new wave of highly efficient companies: Replit builds a complex development platform with under 50 people (source), and Midjourney revolutionized AI image generation with a team of less than 15 (source). These companies aren't outliers - they're early examples of how AI-powered tools are enabling small teams to deliver outsized impact. Y Combinator has noted that AI is allowing startups to operate more efficiently with smaller teams, as automation and AI tools reduce the need for large engineering departments (source).
It will probably get worse, or at the very least the bar for software engineering will rise in order to maintain employment at legacy companies. It seems clear if you are a specialized engineer and want to continue working in tech lucratively, you need to start building in your spare time. The risk of not building and learning outside of the typical work domain is your specialized engineering domain will just be cannibalized by generic AI prompts and the current job will slowly become a relic of the past. Similar to the days of using morse code to send a message -- sure, in 2025 you can write bespoke code but wouldn't it be easier to just prompt features into existence instead? See Replit, Cursor, v0.
As AI tools increasingly handle the details of code generation, the role of software engineers is shifting towards higher-level architectural knowledge. Understanding how to design efficient, scalable systems will become even more critical. While AI can generate code snippets, it is the architectural decisions that drive overall system efficiency and effectiveness. Engineers who can integrate AI-generated components into cohesive, well-architected systems will be invaluable, as they ensure that the technology not only functions but excels in real-world applications.
Domain specific software engineering roles are the most at risk. Frontend, backend, and devops specialization where one domain engineer may not have to know about the implementation or skills to contribute in another domain will soon see their value diminish -- not because their skills are not needed, but because with the speed of execution enabled by AI, teams will find there is not enough actionable, business value generating work to go around for any one specific domain.
The true "full stack" engineer will be the new normal.
However, it's worth noting that ML and AI engineers may still find opportunities as domain-specific experts. As AI and machine learning technologies continue to evolve, these engineers are crucial for developing and refining models. Yet, with the increasing commodification of large language models (LLMs), the focus is shifting. The real value is less on the ML innovation itself and more on the service layer that applies these models effectively. Companies are looking for ways to integrate AI into their products and services, making the application and deployment of AI models a key area of growth.
Building a Business
The issue is that in 2025, if you want to start a tech company, you need to have a predictable amount of staying power. Currently, there is no guarantee that your product will last longer than a couple years (even if you gain PMF) before being commoditized - especially if you're in b2b SaaS.
You used to be able to leverage talented engineers to build a product that was unique and had a moat solely based on execution. However, the replicative nature of AI has flipped execution moats on its head, and companies are already starting to realize that they can leverage productivity gains from AI to build all of their b2b 3rd party tooling inhouse for a fraction of the cost. As long as it has been done before, with just the right amount of compute, AI can quickly replicate work that has already been done to the right prompter. Execution moats are largely dead.
This of course does not mean there will not be new winners. But it is harder than ever to predict how long a business will last even if PMF is found.
Bright Spots
Despite AI's disruptive impact on traditional software work, we're entering an exciting era where software's benefits can finally extend beyond pure digital products. Now in 2025, as software development costs plummet due to AI, we're seeing a renaissance in hardware innovation and real-world services.
The most promising opportunities lie in businesses that combine software with physical products or tangible services - areas that have historically been underserved by pure software solutions. Think:
- Hardware innovation (rockets, robotics, VR devices)
- Infrastructure and logistics (autonomous vehicles, drones)
- Healthcare and biotech (medical devices, diagnostic tools)
- Physical services augmented by software
- Defense and security systems
These sectors benefit from natural moats that pure software struggles to replicate: physical infrastructure, regulatory requirements, and complex supply chains. While software remains crucial to these businesses, it's no longer the primary differentiator - it's an enabler for solving real-world problems outside of the world of bits.
This realization has led me to explore opportunities in the physical storage industry - a decidedly unsexy business that's been around for decades. While it won't make TechCrunch headlines or attract VC bidding wars, storage is a stable, recession-resistant industry that's largely ignored by tech talent. Unlike many digital services that risk AI commoditization, people will always need physical space to store their belongings - no amount of artificial intelligence can digitize your furniture or holiday decorations. (I'll be exploring this more in a future post)