I’m Kenta — an AI engineer and founder from Japan, building in Silicon Valley.
Right now I’m working on 3 products:
- HelloSake - HelloStock - AI Translator (WIP)
I share insights on AI product development, startup lessons, and stock investing as I build.
Always happy to connect with founders, builders, and investors.
@jack が開発したBuzzに関して、1分で解説する動画をAIに作らせてみた。
スタックはFable 5 & Hyperframe & ElevenLabs V3です。
このコンテキストの分断はAIによってプロジェクト数が増えてから、ずっと課題感をもっていたのでBuzzの思想にはかなり共感する
I'm working abroad, and I got tired of switching between DeepL, ChatGPT, and Grammarly, so I'm building an app that combines them all into one.
1. Translating with DeepL -> The accuracy is a bit off
2. Typing "Translate this" to ChatGPT every time -> It's a hassle
3. Proofreading with Grammarly -> That's already three different apps
I'm planning to release the beta version in a few days.
Codex is no longer just a developer tool. With role plug-ins and Sites, it is becoming a workplace layer for spreadsheets, docs, decks, and dashboards. The real skill is not prompting faster. It is supervising AI work with evidence, constraints, and taste.
It seems that Anthropic's flagship model 'Opus 4.7' will be released within this week. Furthermore, a new AI tool for designing websites and presentation materials is also expected to be announced simultaneously.
Anthropic’s latest engineering post made one thing clear:
The future of AI coding isn’t just better models.
It’s better harness design.
Their big insight:
LLMs are often bad at evaluating their own work, especially on long-running tasks or subjective work like frontend design.
So instead of using one agent, they used:
•a planner
•a generator
•an evaluator
That structure produced much better apps than a solo agent.
The takeaway:
AI performance is increasingly about the system around the model, not just the model itself.
As models improve, the best scaffolding changes too.
The moat may be in designing the right harness.
New on the Anthropic Engineering Blog:
How we use a multi-agent harness to push Claude further in frontend design and long-running autonomous software engineering.
Read more: https://t.co/HWvmXk1ykn
Someone asked what advice founders ignore. That they:
1. Should change their name.
2. Should launch fast.
3. Shouldn't treat fundraising as success.
4. Shouldn't assume they can raise because it's time to.
5. Should fire bad people quickly.
6. Shouldn't talk to acquirers.
New in Claude Code: auto mode.
Instead of approving every file write and bash command, or skipping permissions entirely, auto mode lets Claude make permission decisions on your behalf.
Safeguards check each action before it runs.