Are we just throwing spaghetti against the wall?

We're in the AI era of ship until they succumb.

If you talk to product teams at large enterprises or high-profile AI startups right now, you’ll hear the same thing: everyone is launching, everyone is moving fast, and no one knows what will stick.

Are we in AI’s spaghetti-on-the-wall era?

This moment reminds me of the early internet boom—a period of hyper-experimentation where new ideas, interfaces, and business models were being tested daily. Some ideas stuck and became companies that touch billions today, but many more vanished overnight.

Meanwhile, smaller, hyper-focused teams are quietly bootstrapping—or leanstrapping—their way to millions in ARR by mastering one need exceptionally well. Software development. Meeting transcripts. Creative workflows. AI support agents.

Just for fun: let’s consider a V/E ratio for private companies — venture dollars raised per dollar of earnings (arr).

Perplexity V/E ratio: $665M raised / $65M ARR: 10

Cursor V/E ratio: $160M raised / $100M ARR: 1.6

This suggests two dominant strategies at play”

1️⃣ Brute-force a horizontal play across multiple verticals and use cases, fueled by capital.

2️⃣ Find a niche, dominate it, and expand from there. A well-loved feature can become the next empire.

The pattern we’ve seen across every major technology shift: the best startups won not because they were first, but because they reshaped the way people interact with computers.

We’re watching this shift closely.

• What are the new interfaces that will feel inevitable?

• Which behaviors are AI changing permanently—and which are just hype?

• What are founders seeing that isn’t obvious yet?

Here’s what mattered in AI this week:

Anthropic’s new ‘hybrid reasoning’ AI model is its smartest yet: Anthropic is releasing Claude 3.7 Sonnet, its first “hybrid reasoning model” that produces near-instant responses or extended, step-by-step thinking. In addition to a new model, Anthropic is also releasing a “limited research preview” of its “agentic” coding tool called Claude Code. Claude has been powering the popular AI IDEs like Cursor and Windsurf, leading several to wonder what happened to the other high-profile coding agent startups like poolside who raised millions of dollars last year.

Inception emerges from stealth with a new type of AI model: Inception, a new Palo Alto-based company started by Stanford computer science professor Stefano Ermon, claims to have developed a novel AI model, Mercury, based on “diffusion” technology. Inception calls it a diffusion-based large language model, or a “DLM” for short.

Unlike traditional autoregressive models that generate text sequentially, Mercury processes text in parallel, promising significantly faster generation speeds, up to five times faster, and potentially higher quality outputs. This promising approach could lead to much faster — like 10x faster— and efficient inferences.

Hugging Face launches FastRTC to simplify real-time AI voice and video apps: Hugging Face introduced FastRTC, an open-source Python library that removes a major obstacle for developers when building real-time audio and video AI applications. FastRTC offers automated features handling the complex parts of real-time communication.

The library provides voice detection, turn-taking capabilities, testing interfaces and even temporary phone number generation for application access. The library’s primary advantage is its simplicity. Developers can reportedly create basic real-time audio applications in just a few lines of code.

Perplexity teases a web browser called Comet: Perplexity AI-powered search engine

DeepSeek is accelerating the launch of its next reasoner AI. Reports suggest they planned to release R2 in early May but now want it out ASAP. This comes after R1 went viral with its efficient yet powerful reasoning capabilities – matching industry leaders.