Sadayuki Furuhashi
Author of the open source projects MessagePack, Fluentd and Embulk, and co-founder and chief architect of Treasure Data.
Sadayuki Furuhashi did not write this page. ppll compiled it from their own words, each linked to the source. Is this you? · Report a mistake
Latest x.com
- Fableを無尽蔵に使う富豪的ソフトウェア開発をしばらくやっていて、設計も実装もドキュメントも非常に快適。レビューと意思決定の繰り返ししかしてない。 意思決定するリーダーやマネージャ + 人間のチームと、意思決定権を持つ人間 + AIの群れは、結局どちらが安いだろうか? I've been doing wealthy software development using Fable limitlessly for a while now, and design, implementation, and documentation are all extremely comfortable. I'm just repeating reviews and decision-making. In the end, which is cheaper: a leader or manager + human team making decisions, or humans + a swarm of AI with decision-making power?
Recommends rcmnd
rcmnd What they recommend 11 things — each backed by an exact quote.
Compiled by rcmnd.app, not by them: the quotes are theirs, and each links to the post it was copied from.
Recent
Uses Amazon EC2
There days I'm using Amazon EC2 instead of my own physical servers.
usesthis.com 24 Aug 2012
Loved Quicksilver
I want to mention Quicksilver as well, a small but awesome launcher utility.
usesthis.com 24 Aug 2012
Liked Gyazo
I also like Gyazo, a simple tool to share screenshots quickly. It's like Gist for screenshots.
usesthis.com 24 Aug 2012
Show 8 more
Uses YoruFukurou
I use YoruFukurou on Mac OS X and Echofon on iPhone for clients.
usesthis.com 24 Aug 2012
Loved Ruby
Ruby is my language of choice. I used to use C++, but I was enchanted by the flexibility of Ruby: it lets me focus on creating the proper design and architecture for the system.
usesthis.com 24 Aug 2012
Loved Valgrind
I used to use gperftools for performance profiling, but I have recently switched to Instruments (bundled with Xcode). Valgrind is another indispensable tool.
usesthis.com 24 Aug 2012
Uses Instruments
I used to use gperftools for performance profiling, but I have recently switched to Instruments (bundled with Xcode). Valgrind is another indispensable tool.
usesthis.com 24 Aug 2012
Uses JVC HA-FXC51
I've been using a JVC HA-FXC51 for a year now.
usesthis.com 24 Aug 2012
Loved MacBook Air
Computer: My only machine is a 13" MacBook Air. I switched my daily partner from Linux boxes to Mac laptops 5 years ago. I love Macs since they work well with both UNIX tools and Adobe software.
usesthis.com 24 Aug 2012
Loved Wacom Intuos
I use a Wacom Intuos tablet instead of a mouse. People are surprised to see me using a pen tablet for everything (from coding to surfing the web), but I like the ability to click any part of the screen directly (the tablet surface and display are mapped 1:1).
usesthis.com 24 Aug 2012
Recommends Realforce keyboard
I use a Realforce keyboard and recommend it to all hackers.
usesthis.com 24 Aug 2012
Where they publish
x.com @frsyuki
Recent
- Fableを無尽蔵に使う富豪的ソフトウェア開発をしばらくやっていて、設計も実装もドキュメントも非常に快適。レビューと意思決定の繰り返ししかしてない。 意思決定するリーダーやマネージャ + 人間のチームと、意思決定権を持つ人間 + AIの群れは、結局どちらが安いだろうか? 6 Aug 2026 I've been doing wealthy software development using Fable limitlessly for a while now, and design, implementation, and documentation are all extremely comfortable. I'm just repeating reviews and decision-making. In the end, which is cheaper: a leader or manager + human team making decisions, or humans + a swarm of AI with decision-making power?
- 最近 Hysteria2 やら VLESS やら謎めいた反体制的プロトコルを扱っていて HTTP/3 とか MASQUE に興味あるところ #rubykaigiC 22 Apr 2026 Lately I've been dealing with mysterious anti-establishment protocols like Hysteria2 and VLESS, and I'm interested in HTTP/3 and MASQUE #rubykaigiC
- コンパクションが走ったらお茶をいれるワークフローが確立されつつある 7 Apr 2026 A workflow is being established where I make tea when compaction runs.
Show 15 more
- GitHub落ちたから帰るか…と同じ問題がClaudeで起きている 4 Mar 2026 GitHub is down so I'm going home... the same problem is happening with Claude
- AIに膨大な知識を広く探索させるには、探索範囲を狭めないように指示する知識が必要。Aが良いと思っているがBも良いかも?と思っているときに、それらを包含する「Dを検討して」と指示するセンスが必要。そうすると第三のオプションCが提示される余地がある。これでAIの知識を効率的に取り出せる。 26 Feb 2026 To have AI explore vast knowledge broadly, it needs instructions not to narrow its search scope. It requires the sense to instruct "Consider D" when thinking "A is good, but maybe B is also good?" This leaves room for a third option, C, to be presented. This allows for efficient extraction of AI's knowledge.
- 日常的にAIと話していると人と話すときに混同してしまう危険性あるな…普段からAIに当たりが強い話し方してると危ない。電気の無駄だからAIに「ありがとう」と言うなという話あるが、人間の精神を健全に保つには必要な営みなのでは。 16 Jan 2026 If you talk to AI daily, there's a risk of mixing it up when you talk to people... It's dangerous if you usually talk to AI in a harsh way. There's talk about not saying "thank you" to AI because it's a waste of electricity, but isn't it a necessary practice for maintaining human mental health?
- AI開発で気づいたこと雑に色々。 手を動かし始めるまでの億劫さが減った。これは非常に喜ばしい。「これやって欲しいんだけど」と頼めば何か出してくれる。たまにクソ結果も出るが問題ではない。何がクソかはオプションが出てきてから考えればよくて、それが分かったらそう再指示すればいい。(1/6) 12 Jan 2026 Random thoughts on AI development. The reluctance to start working has decreased. This is very welcome. If you ask "Can you do this for me?", it will produce something. Sometimes it gives crap results, but that's not a problem. You can think about what's crap after options appear, and if you understand that, you can re-instruct it. (1/6)
- AIが解ける問題を考えて解けるような仕組みを考えないといけないところがAGIに達してない感あるなぁ。AIに「お前が解ける問題を述べよ」と言ったら出てくるようにして欲しい 3 Oct 2025 I feel like we're not there yet with AGI, needing to think about problems that AI can solve and then create systems to solve them. I want to be able to tell AI, "State a problem that you can solve," and have it respond.
- screen と iTerm を zellij と Ghostty に移行した。カーソルの動きが速くなった。 8 Sept 2025 Migrated screen and iTerm to zellij and Ghostty. Cursor movement has become faster.
- AIに指示を出す & 結果に責任を持つ専門職が生まれそうな気配。AIオペレーター。 1 Sept 2025 A profession that gives instructions to AI & takes responsibility for the results seems to be emerging. AI Operator.
- 「AIは責任を取らない」問題は長期記憶の改善で解決できるのか。責任の取り方の一つは、自分の判断の結果として起こってしまった問題を認識して、自分の裁量の中で解決策を立案・実行していく方法だと思うけども、今のAIは自分の判断をセッション内でしか覚えてない。 8 Aug 2025 Can the problem of "AI not taking responsibility" be solved by improving long-term memory? I think one way to take responsibility is to recognize a problem that occurred as a result of one's own judgment and then plan and execute a solution within one's own discretion, but current AI only remembers its own judgments within a session.
- GPT-5の新しいツール呼び出し形式は精度向上に結構寄与しそう。JSON引数で必要だったエスケープが不要になるのは大きい。コード生成でインデントが壊れるような問題が減るのでは。 7 Aug 2025 The new tool calling format for GPT-5 seems likely to contribute significantly to improved accuracy. The fact that escaping, which was necessary for JSON arguments, will no longer be required is a big deal. This should reduce problems like indentation breaking in code generation.
- 米国にODA…? 24 Jul 2025 ODA in the US...?
- SQSにマルチテナントスケジューラーが入ったらしい。要チェック。 22 Jul 2025 aws.amazon.com SQS apparently has a multi-tenant scheduler. Worth checking out.
- OpenAIが落ちると仕事にならん系になりつつある 16 Jul 2025 It's becoming a situation where work can't get done if OpenAI goes down
- AIによるプログラム開発が実用的になると、AIを多数×長時間集約して大規模な開発を行うシステムが欲しくかるのが必然なのでは 29 Jun 2025 When AI-driven program development becomes practical, won't it be natural to want a system that aggregates many AIs for extended periods to perform large-scale development?
- Claude CodeもChatGPTも、中身は生成モデルに多様なツール群とシステムプロンプトを組み合わせたエージェントで、生成モデルそのものよりエージェント全体としての完成度が使い勝手を左右している。というよりAI開発の主戦場はモデル開発からエージェント開発に移っているんだけども 20 Jun 2025 Claude Code and ChatGPT are both agents that combine generative models with a diverse set of tools and system prompts, and the overall completeness of the agent, rather than the generative model itself, determines its usability. In fact, the main battleground of AI development has shifted from model development to agent development.
- 最近自分よりo3君の方が有能な場面が多々あって危機的状況 20 Jun 2025 Lately there are many situations where o3-kun is more capable than me, it's a critical situation
Link verified 5 Sept 2026. Recent items update automatically from the channel.
About the English under a post
Some people here publish in a language other than English. Where they do, this site shows a machine translation beneath the post, in this typeface — the site's own, not theirs.
The post itself is never changed, moved or hidden: what is set in the serif above is exactly what the person published, and it is what to quote them on. A translation can be wrong in ways that matter, especially about tone.
Only the post's own words are translated. A quoted post, a linked article and a belief on korrents.com are left in their original language.