Skip to content

2025

  • 2 min read

About Vibe Coding

Impact of vibe coding on product desing

Software development is changing rapidly overnight. Indeed:

  1. OpenAI announced Codex on May 16, 2025, for their $200/month Pro users https://openai.com/index/introducing-codex/.
  2. Microsoft GitHub Copilot released its new coding agent on May 19, 2025. https://bsky.app/profile/github.com/post/3lpjxvgje7s2k
  3. Google announced a tool called Jules (jules.google.com) on May 20, 2025, making it available for free and
  4. Mistral releases devstral, an open-source model for coding agents on May 21, 2025. https://mistral.ai/news/devstral

These new coding agents—along with Cursor, Lovable, Windsurf, V0, Bold.new, and others—are all tools that support some form of “vibe coding” (a term coined by Karpathy indicating AI-assisted coding).

This gives rise to a lot of FUD (fear, uncertainty, and doubt) from the corporate gatekeepers. The short-term opportunity is this: in a design thinking approach, a “research prototype” that checks the basic hypotheses (who is this product for, what problem does the product solve) can be developed much faster using vibe coding.

Even with the expected “valley of disappointment” that may follow (because users tend to overreact to the initial prototype, which will likely need to be rewritten from scratch), in the end, the chance of building a product that resonates with users is much higher and it will be ready sooner—if the same good old software process is followed, from prototype to Minimum Viable Product (MVP) to Version 1 accepted by users.

  • 2 min read

The future Ai ecosystem will be open

a79b47e679f3b1e93e1d2a2aadbb3461875225293794331ce2b9f471931c3f44.jpg
The future AI ecosystem will be open

I was just reading the article The walled garden cracks: Nadella bets Microsoft’s Copilots—and Azure’s next act—on A2A/MCP interoperability and this is how I see what's happening in the AI landscape:

  • Antropic: best user experience and defined MCP (Model Context Protocol)
  • Google: best model on all leaderboards with Gemini and defined A2A (Agent-to-Agent)
  • Microsoft: let's build an open AI ecosystem with MCP and A2A, releases supports for A2A and MCP in VS Code
  • Deepseek: after they pulled of DeepSeek V3, Deepseek released open model DeepSeek-Prover-V2 that tackles advanced theorem proving achieving an 88.9% pass rate on the MiniF2F-test benchmark (Olympiad/AIME level theorems) and solving 49 out of 658 problems on the new PutnamBench. This means that Deepseek is cracking the reasoning part of LLM's.

And OpenAI:

Seeing all those signals, we conclude: the AI future will be open!