“What does code abundance mean for the world?”

For the longest time, I’ve wanted a custom LLM-powered knowledge management system that could consolidate various information sources to improve accuracy. WeChat’s IMA database offers some level of customization, but since most of my work files aren’t hosted there, the manual uploading process and mobile interface don’t quite meet my needs. OpenAI’s “Build Your Own GPT” feature works similarly but lacks the collaborative capabilities required for internal teams. After browsing GitHub for a while and not finding anything immediately usable, I figured: this shouldn’t be that hard to build. So I asked Devin to create a basic internal knowledge base system tailored for our company.

Within half an hour, Devin delivered a working MVP—deployed on a virtual machine with a testing link ready to go. The UI was minimal, and there’s still plenty of room for improvement (like multi-model support via API keys, more file format integrations, etc.), but the core features I asked for—account login, team collaboration on knowledge management (contribution and admin control), and multi-model switching—were all there. Better yet, future iterations can mostly be handled through prompting. So… Build Your Own App just works now? Devin represents just one example in this emerging era of Build-Your-Own-App (BYOA) platforms.

“This is worrying, in a good way.”

There was a time when fast product development was a major competitive edge. A funny example: every student who took Stanford’s first iOS development course reportedly made at least $1M—because back then, the App Store was new, there was a real technical barrier to entry, and even basic utility apps could draw massive traffic. But we’re now approaching a near future where code is no longer a scarce resource, and neither is the ability to ship products. Everyday people will be able to create tools and apps using natural language alone.

So how does this shift change the current landscape of competition and user behavior?

Let’s break it down.

I’ll first separate users into consumer-side (C-end) and business-side (B-end), and then further divide the consumer-side into two categories: productivity and entertainment/killing time.

Consumer-side: Productivity

We’ll likely see only modest shifts in user behavior here. While AI coding tools offer vast customization potential, the idea of users developing their own personal productivity tools will remain fairly niche. Think of website building during Web 2.0—there were plenty of no-code platforms, but only a tiny percentage of internet users actually ran personal websites. Most people still prefer using existing platforms and tools—even if they aren’t perfectly tailored—because convenience usually wins.

That said, prosumers—those willing to pay $20/month for ChatGPT—are increasingly open to AI-augmented workflows. These power users might be willing to pay for personalized tools. Whether this takes the form of features within foundation models or entirely new apps remains to be seen. But even for these users, tool customization likely won’t be a frequent need. Take my Devin-built KMS, for example: once set up, it likely won’t require frequent updates.

Consumer-side: Entertainment & Killing Time

This is where things get more interesting.

I’m especially intrigued by the potential for a reverse recommendation algorithm—one that doesn’t just serve content but starts from the user’s intent. With code (and by extension, content) in abundant supply, the most valuable signals will be user preferences and social connections.

Imagine if I could port my preference data—ideally in a privacy-preserving way via zero-knowledge proofs—to initialize a brand-new content network. In an extreme, yet increasingly plausible scenario, we could see platforms like TikTok or Xiaohongshu entirely recreated with AI-generated content tailored to individual taste. As a result, content-driven network effects would weaken, while social graph-driven effects would strengthen. Platforms built on social capital might deliver faster, better, and more meaningful content — long Tencent, short Bytedance?

Recommendation algorithms have arguably been the defining force in consumer internet over the past decade. The world’s smartest minds spent years optimizing how to maximize time spent in-app. But as Jonathan Haidt notes in The Anxious Generation, recommendation systems are now increasingly linked to rising mental health issues among teens. In the grand scheme of things, it’s still unclear whether these algorithms represent productivity gains—or a slow erosion of human well-being.

To fight magic with magic, perhaps the best way to counter algorithmic addiction is by using AI as a high-quality content curator.

In Web3, we emphasize user-owned data and preferences. The goal is to create decentralized incentive structures that empower human-centric social networks over algorithm-driven feeds. I believe this vision is worth pursuing—but before we can fully displace TikTok, Instagram, and Twitter, we’ll likely experience several transitional stages.

In short: for companies that operate on a “come for the tool, stay for the network” model, the tool is becoming less important. The network is becoming more important—and how we define "network" may also evolve. Among all types of network effects, content-based ones are likely the weakest when compared to those built on social relationships, commerce infrastructure, or payments.

Business-side: B2B Use Cases

Existing SaaS platforms will have major opportunities to offer users more customization. While backend processes and data structures (think SAP) are hard to modify, there’s significant room to personalize frontend elements—UI, UX, and automation.

I may not move my whole team off Slack, but I’d love more no-code functionality built into it. Auto-replies to my boss (“Got it 🫡”), summarizing group chats, and auto-generating weekly reports are all small tweaks that can make a big difference. So no, AI won’t “kill SaaS,” but it could make SaaS radically more customizable.

Code Abundance and Crypto

Finally—how will code abundance affect crypto? In my view, the impact is largely positive.

Crypto has long struggled with poor UX, unclear product-market fit, steep onboarding barriers, and high development costs. Compared to Web2, it hasn’t yet reached a winner-takes-all dynamic. AI-driven code abundance can accelerate the search for PMF, enhance user experience, and lower the barrier for non-crypto-native users to start engaging with on-chain applications.

⚠️ Of course, code abundance ≠ automatic PMF. But it does accelerate the journey—just as it will in other emerging industries.