Shower Thoughts on AI Coding Tools
Where are AI coding tools headed? A lot of it rhymes with the mobile and internet revolutions.
The Copying Problem
Companies keep building features on top of frontier AI, and then the frontier companies copy those features. Jason Calacanis said if you're using OpenAI's APIs, you might be a fool because they're watching and will copy everything. He's right. But it's also unavoidable.
Apple copied the weather app. Apple Music and YouTube Music copied Spotify. Reels and Shorts copied TikTok. Sometimes they had to copy just to survive.
The Spotify Example
Here's the thing I keep coming back to: Spotify sits on iOS and Android. Apple and Google both built competing music apps. And Spotify is still #1 in the US.
That's not survival. That's winning. On someone else's platform, against that platform's own product.
Netflix and Shopify don't count here. They sit on the open internet, not on platforms that directly copied them.
Can any AI coding tool pull off what Spotify did? I genuinely don't know. But it's possible.
Cursor's Position
Cursor is sandwiched between multiple platforms:
- Model APIs (Anthropic is building Claude Code, OpenAI is building Codex)
- VS Code (Microsoft has Copilot)
- Open source: Cline, Kilo Code, Open Code, AmpCode
That's pressure from every direction. Cursor isn't in a bad spot - they might win. But if I had to bet my life on it, I wouldn't choose them. The odds aren't impossible. Just not where I'd put my money.
Different Fights
Not everyone is playing the same game.
Claude Code is trying to replace Cursor. AI-native coding from the terminal.
Kilo Code and Windsurf are trying to replace VS Code itself. They want to be the IDE, not just a plugin.
Different strategies. Different risks.
The Indexing Split
There's a technical split too. Cursor does heavy indexing to handle large codebases. Claude Code uses grep and glob, betting that bigger context windows will solve the problem.
From what I've seen, Claude Code struggles more on large codebases. People who disagree might not have worked on truly massive repos.
You can still be productive with it. You can probably make millions. But when your codebase gets big, you end up back in older workflows. Maybe not writing code by hand, but looking at code, copy-pasting through chat, using sidebar features.
That's probably why Cursor stays popular despite all the Claude Code hype.
Vibe Coding
AI isn't there yet for heavy work. Cursor is closer to something usable. Vibe coding tools struggle with large codebases.
There's also V0, Lovable, Emergent. They're betting on AI building entire apps from prompts. But Claude already has Artifacts. As that matures, do these startups just get absorbed?
Counter-argument: focused startups can do one thing better than platforms serving everyone.
Memory Companies
Mem0, Zep, Letta. They're building memory for AI. I think they'll end up like Neon or Supabase. Infrastructure plays.
It'll be interesting to see how big they get. But I think AWS wins this one. You'll see Zep and others in the picture, but the big cloud providers will dominate.
Things I've Noticed
Kilo Code is positioned well. Multiple products, a coding agent that does a lot. Good bridge between different use cases. That's probably why they might win enterprise.
An Nvidia employee told me management pushes them to use Cursor. Usually it's the opposite. Employees push management on tools. Nvidia is betting on AI. They're not planning to lose.
Claude Code gets a lot of hype. But people forget how many ads they run. Every time I open YouTube, I see a Claude ad. They already have mindshare and they're still throwing money at it.
How Early This Is
Outside of direct API interaction with OpenAI or Anthropic, the ecosystem is surprisingly narrow. That tells me we're early. Very early.
Think about Apple. Huge company, massive ecosystem. But what they actually offer is narrow. iPhones, MacBooks, a developer platform. Safari and the rest just keep the ecosystem alive. No AWS-style cloud.
Open question: when Anthropic and OpenAI mature, will they stay narrow like Apple? Or expand like AWS?
Where I Land
Most tools in this space will probably get copied or commoditized. Some will survive. The ones that nail UX, build real switching costs, or find wedges the platforms won't bother with.
Not a prediction. Just a mental model I'm working with.
Note: I wrote the initial draft of this blog post and used AI to help prepare the final version. I will be rewriting and refining the content manually soon.