Write Things Down | Stratechery by Ben Thompson
Read the Article: https://stratechery.com/2026/write-things-down/ Links: Stratechery: https://stratechery.com Sign up for Stratechery Plus: https://stratechery.com/stratechery-plus Sharp Tech website: https://sharptech.fm
Read Video · Transkript & Insights
Diese Folge hat ein vollständiges Transkript + KI-Insights
Kostenloses Konto · keine Karte nötig · 150 Credits nach Registrierung, genug für diese Episode
- 📄 Vollständiges Transkript mit Zeitstempeln
- ✨ KI-Zusammenfassung, Keywords & Mindmap
- 💡 Kernaussagen & Zitate
- Sprecher 1
Episoden-Zeitlinie
Ben Thompson reflects on David Allen's Getting Things Done and the RAM analogy for mental clutter.
- David Allen's Getting Things Done argues the mind functions like RAM—a focusing tool with limited capacity, not a storage place, so incomplete items constantly distract us.
- Thompson admits he was a beta user of OmniFocus but sucked at using the system, eventually hiring an assistant to manage his inbox and tasks for him.
- The core lesson is that writing things down lets you empty your RAM and focus on actual work, a theme that recurs throughout the episode.
Jensen Huang declares AGI has arrived with GPT-6 Astra, but Thompson disputes the definition.
- NVIDIA CEO Jensen Huang declared on X that AGI has arrived with GPT-6 Astra trained on roughly 100K+ Grace Blackwell GPUs, his second such claim this year.
- Thompson's personal definition of AGI is AI that learns continuously, which current LLMs fail because their weights are frozen after training.
- He cites his own experience with Claude, whose January knowledge cutoff made it give wildly outdated RAM pricing advice, proving it couldn't update its understanding over time.
Thompson argues the real AGI moment may have been the harness—deterministic software that writes things down.
- Claude Code's 2025 launch introduced a workflow of writing copious notes and markdown files that could be read into context to keep models on task.
- This crude form of memory simulates continuous learning even with a frozen model, mirroring how writing propelled human evolution from oral communication to scalable, extendable knowledge.
- Biological evolution is the slowest form of learning, oral communication is like a lossy context window, but writing things down made learning scalable.
Schlüsselkonzepte
- write things down— The central thesis: externalizing thoughts into written form is the foundation of both human progress and AI capability.
- Getting Things Done— David Allen's productivity philosophy that frames the episode's argument about clearing mental RAM.
- AGI— Artificial General Intelligence, which Thompson defines as AI that learns continuously, unlike current LLMs.
Bemerkenswerte Zitate
AGI is AI that learns continuously.
🔥— Thompson offers a crisp personal definition that directly contradicts Jensen Huang's declaration, reframing the AGI debate around a testable criterion.
My solution to not being the sort of person who is organized enough to run Omnif Focus was to hire someone to do it for me.
🎯— A candid admission that even the productivity guru's biggest fan outsourced the system entirely, revealing that tools alone don't change behavior.
Konkrete Handlungen
✅Personal Productivity
Your mind is a focusing tool, not a storage place; open loops consume RAM and create anxiety.
This week, do a brain dump of every incomplete task into one trusted list, then review it once daily.
Systems only work if you actually use them; even the biggest GTD fan outsourced his OmniFocus.
Pick one tool you already use and commit to it for 7 days instead of switching to a new app.
🤖AI Literacy
Current LLMs are not AGI because they don't learn continuously; their weights are frozen.
Test your AI assistant with a question about a recent event and note where its knowledge cutoff fails.
LLMs write token by token and re-read the entire context each time, so they are not persistent entities.
When using an AI agent, maintain an external notes file it can read into context to simulate memory.
Transkript und Insights werden KI-generiert und können Fehler enthalten. Die Genauigkeit hängt von der Audioqualität und der Deutlichkeit der Sprecher ab — bei Unklarheiten ist das Originalaudio die maßgebliche Quelle.
Episoden und Videos zum Lesen
Podcast-Episoden

Grit: The power of passion and perseverance | Angela Lee Duckworth
TED Talks Daily
6. Sept. 202610:23EN
PA Replay: GOING ALL IN....What You Need To Know
The Pure Athlete Podcast
14. Juli 202652:14EN
Diogo Almeida - Deep Learning: Modular in Theory, Inflexible in Practice - TWiML Talk #8
The TWIML AI Podcast (formerly This Week in Machine Learning & Artificial Intelligence)
23. Okt. 201646:11EN
#2 Cold Case Ötzi
Wer wir sind und warum das nicht klappte ...
17. Apr. 202536:19DE
EP692 | 🍺
Gooaye 股癌
29. Aug. 202649:59ZH-Hant
Радио-Т 1026
Радио-Т
15. Aug. 2026RU
Videos

Jack Dorsey's Buzz: Clearly Explained (and how to use it)
Greg Isenberg
28. Juli 202638:44EN
Everyone Is Still Undersizing the AI Market | Eric Vishria
Invest Like The Best
11. Aug. 20261:17:53EN
Qwen3 is a fantastic open-source model
Matthew Berman
29. Apr. 202514:05EN
#36 Ludwig der Bayer - der dem Papst trotzt
99 mal Geschichte
11. Dez. 202559:57DE
从「上瘾模型」到「专注力训练」,如何在被算法理解的世界里重新找回主动?| 英文访谈 S9E33
声动活泼
16. Okt. 202550:48ZH-Hans