shogo

Hi, I'm Shogo👋 I'm 22, and I wear two hats: co-founder & CTO of TPS Collective, and a PM at @mercari_jp I live and breathe AI and startups — how they actually work, what's real vs. hype, and where this is all heading. What I'll post here:
→ the AI & startup news that actually matters
→ what I'm learning as I build
→ my honest take, not just headlines If that's your thing, stick around. Let's figure this out together.

this is a thoughtful proposal from demis:

@sama One of the most thoughtful takes on AGI I've read. The biggest challenge isn't whether we'll reach AGI—it's whether our ability to govern and evaluate these systems can keep pace with their capabilities. Innovation and safety shouldn't be competing goals.

@sama @TheAhmadOsman The most interesting part isn't that Sam asked for feedback. It's that frontier AI is now improving through continuous user feedback, much like modern software. Models are becoming products, not just research artifacts.

@TheAhmadOsman thanks! why? how can we improve?

I like 5.6 Sol more than Fable 5 btw.

@TTrimoreau Thinking about the next problem. Code is no longer the bottleneck. Deciding what to build, what to automate, and what to verify is.

I'm curious while AI is writing your code, what are you doing?

Today, I published my very first iOS app to the Apple App Store. It's only available internally for now, not to the public yet, but shipping that first app feels like a huge milestone. Looking forward to building many more.

Interesting move from Anthropic. They've extended Fable 5 access again, along with the 50% higher Claude Code rate limits. My guess is that their infrastructure is handling demand better than expected. Every extra week gives them more real-world usage data, helps validate GPU capacity, and lets more developers build Fable 5 into their workflow before usage-based pricing kicks in. A smart way to stress-test a frontier model at scale while increasing long-term adoption.

I can’t use ChatGPT because of an error — is it just me?

OpenAI just released the GPT-5.6 family. Instead of one model, GPT-5.6 now comes in three tiers: ☀️ Sol — Maximum reasoning and performance for the hardest problems. 🌍 Terra — The balanced model for everyday work, offering the best mix of capability, speed, and cost. 🌙 Luna — The fastest and most cost-efficient model for lightweight tasks. I really like this direction. Rather than asking one model to do everything, you can now choose the right model depending on the task—similar to how we choose between different cloud instance types or CPUs. AI models are becoming products with clear trade-offs, not just benchmark leaders.

Notion just launched Notion Agents, a standalone app that feels like the evolution of Notion AI. Instead of opening Notion just to chat with AI, you now have a dedicated AI workspace that's deeply connected to your notes, docs, databases, and connected tools. A few things I really like: Access to frontier AI models in one place (Claude Fable 5, Sonnet 5, GPT, Gemini, Grok, and more depending on your plan). Every conversation is automatically saved, organized, and searchable inside Notion. Unlike most AI chat apps, your conversations become part of your long-term knowledge base, so you can continuously build on previous discussions instead of starting from scratch every time. To me, this feels less like "another AI chatbot" and more like an AI-powered second brain. This is a direction I'm excited to see.

MercariでPMインターンをしている菊池翔吾より、インターン期間中に開発したClaude CodeのSkill "mercari-pm-agent" について紹介です！ PMがおこなう「問題の発見→データ収集→PRD作成→UIモック」の一連のワークフローを、1つのセッション内で処理するエージェントです。 engineering.mercari.com/blog/entry/202…

You will always be the GOAT. 🐐🇵🇹 Thank you, Cristiano Ronaldo.

Thanks for everything, Cristiano Ronaldo

Portugal lost 1–0 to Spain today, and Cristiano Ronaldo’s World Cup journey has most likely come to an end. It hurts to see it end this way, but nothing can erase the story he has written. A boy from Madeira who grew up with very little, left his family at 12, carried pain, pressure, criticism, and loss, and still turned himself into one of the greatest athletes in history. Cristiano Ronaldo was never just a football player to me. He was proof that hard work can change your life. Proof that where you start does not have to decide where you finish. Proof that obsession, sacrifice, and belief can take a person further than anyone imagined. He gave dreams, hope, and strength to millions of people around the world. I’m truly grateful that I got to live in the same era as him. To me, Cristiano Ronaldo will always be the GOAT. Thank you for everything, Cristiano.

Today, Cristiano Ronaldo played what is likely the final World Cup match of his legendary career. From a boy growing up in poverty on the streets of Madeira, leaving home at just 12 years old, fighting through loneliness, overcoming a heart condition, losing his father at a young age, and then becoming one of the greatest footballers the world has ever seen—his journey has never been about talent alone. It has always been about relentless work, discipline, and an unbreakable belief in himself. I feel incredibly grateful to have lived in the same era as Cristiano Ronaldo. He didn’t just win trophies or break records. He proved that your circumstances do not define your future. He gave millions of people around the world the courage to dream bigger, work harder, and never give up, no matter where they started. To me, he will always be the GOAT. Thank you, Cristiano. You inspired generations, changed football forever, and showed us that greatness is earned every single day. The world was lucky to witness your story.

God, please. Let Cristiano win today’s FIFA World Cup match. Please let him have this victory. 🙏🇵🇹

Genuine question for people using Claude Code. A lot of people say the Frontend Design skill produces websites that don't have the typical AI-generated look. But after trying it myself, it still feels like an AI trying not to look like AI. Has anyone consistently gotten truly premium, human-quality UI out of it? If so, what changed the results? Better prompts? Better reference images? Different workflow? Additional design systems or constraints? I'd love to hear what's actually working in practice.

Most people are testing Claude Fable 5 on simple prompts and missing what it's actually built for. After reading Anthropic's latest prompting guide, one thing stood out: Claude Fable 5 isn't just a better chatbot. It's designed to operate like an autonomous worker. Key takeaways: • Give it difficult, long-running tasks—not just quick questions. • Describe the goal and constraints instead of micromanaging every step. • Use high effort for complex work. • Prevent unnecessary refactoring by explicitly telling it to solve only what's required. • Let it verify its own work with tools before reporting progress. • Build a memory system so it learns from previous runs. • Use parallel agents for independent tasks. The biggest mindset shift: Stop prompting AI like a search engine. Start managing it like a highly capable teammate. The better your delegation, the better your results.

I usually end up using CapCut because Adobe is too hard to use.

本当にその通りだと思います。 新しいAIモデルやツールを追い続けること自体が目的になってしまって、結局何も作れず、何も売れない人は少なくないです。 AIはもう十分実用的なレベルに達しています。これから差がつくのは、最新情報を知っている人ではなく、それを使って実際に価値を生み出し、お金を生み出せるビジネスを作れる人だと思っています。

正直、今の自分は「個人開発したアプリでご飯を食べていく」ことにしか興味がないので、それ以外の「新しいAIモデルが登場した」とか「ループエンジニアリング」とか、今はあまり興味がないんですよね。もうAIは十分賢いので、実直にビジネスに向き合うことで成果を出していきたいと思っています。