Is software development completely done for? In a world where AI models can generate code on demand, should founders still build SaaS startups? Right now is paradoxically the worst time ever to fundraise for a SaaS company — and the best time ever to build software. Andreas Klinger, an engineer-turned-investor with over 100 investments, breaks down what changed, where the opportunities still exist, and why some of those opportunities are billion-dollar use cases if tackled correctly.
The Old World Is Gone
The classic SaaS struggle used to be: your product versus Gmail, spreadsheets, and the user's sheer laziness to adopt anything new. In most cases, the Gmail-and-spreadsheet hack was good enough — and that was already a tough bar to clear.
Now there's a new competitor in the mix: ChatGPT, Claude, and AI coding tools. Most people can get the core functionality they need from a quick prompt. If that's not enough, tools like Cursor or Claude Code let them build exactly what they need on the fly. This is the landscape you're competing in as a founder today — and it's a brutal one.
Two examples drive this home. Toby Lütke, CEO of Shopify, got an MRI scan and was told to install some Windows software to view it. Instead of bothering, he asked Claude to build him a viewer. Done. Christopher Janz, a well-known SaaS investor, used to chase startups building niche tools so he could invest in them. Now he builds those tools himself on weekends, just for fun.
The Ceiling Debate
Whenever this topic comes up, people split into two camps.
The skeptics argue that AI gets you 90% of the way there — but 90% raised to the power of 10 decisions is a horrible, unusable mess. They point out that models have already ingested all publicly available data, so what's left to improve?
The optimists counter: reinforcement learning, multimodal models for complex reasoning, and latent-space computation that could be orders of magnitude faster. The capabilities ceiling hasn't been reached yet — not by a long shot.
Three Paths for Founders
If you're starting something today, you need to make a clear decision about which path you're on, because each demands a different strategy.
Path 1: Build for fun. Your goal is learning and joy. Peter Steinberger, a retiree from Vienna, vibe-coded his project OpenClaw into existence — and it went completely viral. He's the archetype of someone who builds because the act of creating sparks joy. Similarly, Kitze, a prolific indie hacker, builds fully monetized apps at breakneck speed. His app Sto (a speech-to-text tool combining multiple AI models) sells for a one-time $50 payment — no subscriptions, no recurring revenue drama. He's even started a "tinkerer club" with over a thousand members whose shared goal is to rebuild the software world around them.
Path 2: Build a company that makes money. If you want to fundraise, SaaS is still viable — but only if you go deep. Software alone isn't worth much anymore. What enterprises actually pay for is domain knowledge you've productized, plus the hundreds of trade-off decisions baked into your product, wrapped in trust built through marketing and sales. The old IBM ad said it best: nobody gets fired for buying IBM. Big companies can slop-code something together over a weekend, but making the thousands of tedious trade-off decisions that a real product requires? Nobody wants to do that if they don't have to.
The playbook: pick a vertical, go deep, accumulate domain knowledge, add workflows, build trust, and compound relentlessly. Build a "mega app." Leora started as "ChatGPT for lawyers," became "AI for lawyers," and is now becoming "everything for lawyers" — one of the fastest-growing startups right now. The takeaway: build Leora for whoever you know. Doctors, dentists, realtors, whoever. They're buying it for the domain knowledge and the trust.
Path 3: Swing for a VC-scale unicorn. If enterprise sales isn't your thing and you want to build at the frontier, there are massive unsolved problems waiting.
The Frontier: Where the Big Opportunities Are
GitHub Is Obsolete
Pull requests and code reviews made sense when humans wrote all the code. Now the code itself isn't even the main artifact — there's the prompt, the reasoning, the trade-off decisions that happened before a single line was generated. The entire GitHub review process doesn't fit the world we're in. Maybe the future of code review looks more like Linear, or maybe it's a real-time collaborative environment that combines something like Claude Code with an editor and a review process all at once.
Open Source Is in Trouble
Open source maintainers are drowning. They're getting dozens or hundreds of pull requests daily from people who haven't read the documentation, don't understand the codebase, and used some random AI agent to push the code — sometimes fully automatically with no human review. Teams are spending 10–30% of their capacity just triaging AI-generated PRs. Security vulnerabilities are multiplying. The old assumption that open source is inherently more secure is inverting: if your code is public, there's a growing chance it's already been exploited.
Business models are crumbling too. Open-source-core plus paid hosting? Easy to one-shot now. Open-source-core plus paid modules (like Tailwind's model)? Claude Code can replicate those modules on the fly. If you can solve the open source sustainability crisis — whether through new monetization models, better review tooling, or security infrastructure — that's a massive market.
The Mexican Standoff Inside Every Company
Every company now has an internal standoff between roles. Designers think they can replace engineers with AI-generated code. Engineers think they can replace designers. Product managers think they can replace everyone. This dynamic is playing out across industries: film, architecture, consulting, everywhere. Rethinking collaboration, decision-making, and role boundaries in this new reality is a huge opportunity.
The Age of Adoption
Every company has roughly a third of their people who have 10x'd themselves with AI tools, a third who could but haven't, and a third who never will. The problem is nobody knows which third is which. Helping organizations figure this out — and doing it with the right metrics, because lines of code are meaningless now — is a significant opportunity.
A key insight for individual software developers: you're no longer getting paid for the code. You're getting paid for trade-off decisions. In a way, you always were (especially as a senior developer), but now it's unambiguously true. Knowledge work has shifted from "I get paid for what I know" to "I get paid for what I decide, based on what I know, and then I auto-build."
Build vs. Buy vs. Build-It-Myself-Over-the-Weekend
The build-versus-buy debate is about to become the most painful discussion in every company. You'll have Bruce who says "I can just build this over the weekend." You'll have someone else arguing to just pay the monthly SaaS license. And then there's the boss eyeing an acquisition. Facebook Camera is the cautionary tale here — it was built internally as a competitor to Instagram, but shipped three months too late. With AI tools, companies can now move much faster on the "just build it" option. If you're in M&A discussions, expect the acquirer's R&D team to be testing whether they can replicate your functionality directly.
Other Frontier Bets
Prompt injection will be bigger than SQL injection. Models will get better at avoiding it, but it will be a massive security issue for years. All those vibe-coded apps the marketing team spun up? Securing them, standardizing them, making them maintainable — that's an entire industry waiting to emerge.
Surface area is the new moat. It doesn't matter which model runs in the background. What matters is owning the interface, the user relationship, the surface area. Claude.bot, for instance, doesn't care which model powers it — that's the power position.
Build for where models will be in 18 months. Don't build features for today's model capabilities. Assume the things that aren't quite good enough yet will be. Design for that future.
The Real Advice
Don't forget to have fun. You can't spreadsheet your way to success. Obsession is completely underrated — find something you're genuinely passionate about and go deep. Bias to action is underrated too: Open Club is a few weeks old and already reshaping the AI conversation. Just ship.
And don't listen to random YouTubers. Do your own thing.