Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
Claude Code Can Be Your Second Brain
TL;DR
Noah Brier turned Claude Code into an Obsidian-native second brain — he runs Claude Code at the root of his full Obsidian vault so it can search ~1,500 notes, pull relevant prior research into a project folder, and maintain daily progress logs for talks, ideas, and ongoing work.
His key prompt trick is forcing AI into “thinking mode,” not writing mode — Brier built a custom “thinking partner” sub-agent that asks questions, logs emerging ideas, and explicitly avoids drafting, because models keep trying to jump straight to artifacts instead of helping him reason.
The wildest part is mobile deep work from a basement server — using a mini PC, Tailscale VPN, private GitHub sync, and the iPhone terminal app Termius, he can research, write notes, and even ship tiny code fixes from his phone while out at breakfast or sitting by a pond.
Brier thinks voice AI is finally making phones useful for real intellectual work — he says Grok voice mode is currently the best, citing stronger tool use and research ability, and described a 2-hour drive to New Hampshire spent researching a piece plus an hour-long session on transformer self-attention.
His broader AI thesis is that models can dissolve organizational bureaucracy by acting like “English muffin” software — instead of forcing every team onto Asana, Jira, or Linear, AI can sit in the middle as a fuzzy interface that translates across systems and “gets into the nooks and crannies.”
He’s just as energized by AI for kids and education as for productivity — his 10-year-old used V0 on his phone to make a Secret Santa app in 75 revisions, and Brier argues schools should focus less on anti-cheating panic and more on media literacy, citing Tim Harford’s kids book The Truth Detective.
The Breakdown
From Evernote power-user to Obsidian + Claude Code obsessive
Dan Shipper opens by framing Noah Brier as a longtime tools-for-thought nerd — from the old Superorganizers days through Percolate, Variance, and now Alephic. Brier says the headline isn’t really “vibe coding,” it’s that Claude Code has become his main interface for thinking inside Obsidian, where every note is just markdown and folders Claude can actually work with.
Why phones suddenly became useful for deep work
Brier says one under-discussed AI shift is that phones are finally good for serious research and coding-adjacent work. He’s especially bullish on voice: he spent a 5-hour solo drive after dropping his daughter at camp using Grok voice mode for two hours of research, and later used it for an hour-long explainer on transformer self-attention that he called the best explanation he’d heard.
His talk-writing workflow: build a project, then make AI read everything
For a new talk, Brier creates a dedicated project inside Obsidian, seeds it with source material, and starts Claude Code at the root of the entire vault so it can scan everything. For his upcoming Brand.ai talk — tied to the OSS Simple Sabotage Field Manual, Wild Bill Donovan, and a working thesis called “Transformers are eating the world” — he first had Claude search his whole archive for related notes and pull useful material into the project.
The custom “thinking partner” agent that stops AI from drafting too soon
His favorite move is a sub-agent explicitly told: facilitate thinking, don’t write. He even showed all-caps instructions like “I DO NOT UNDER ANY CIRCUMSTANCES WANT YOU TO TRY TO WRITE IT,” because models constantly lunge at outlines and drafts when he really wants questions, synthesis, and a running log of what he’s uncovering.
Daily logs, saved chat transcripts, and resuming broken flow
Brier saves full transcripts from ChatGPT, Claude, and Grok into a chats folder, clips web material into research, and has AI generate end-of-day summaries of what changed in his thinking. That makes one of his most useful prompts dead simple: “Can you catch me up on the last three days of research?” — a way to recover context after getting interrupted by, you know, his actual job.
Bureaucracy, transformers, and the “English muffin theory” of AI
A lot of the conversation turns into live theory-building for his talk: bureaucracy as an innovation, AI as a way around enterprise software change-management pain, and maybe even “bureaucracy as positional encoding,” which he admits is still half-baked. His most memorable metaphor is the “Thomas’ English muffin theory of AI”: models can slip into the nooks and crannies between existing tools and teams instead of forcing everyone onto one rigid system.
The basement server that lets him use Claude Code from his phone
Then comes the setup everyone was waiting for: a mini PC in his basement, Tailscale for a personal VPN, private GitHub sync for his Obsidian vault, and the iPhone terminal app Termius. Through that stack, he can open Claude Code on his phone, ask what’s new in the last two days, resume a project, or even pull down a repo and push a small PR while sitting outside — which he says has completely changed how and where he works.
Kids, vibe coding, and why school should care more about literacy than cheating
Brier’s 10-year-old recently used V0 on his phone to build a Secret Santa chooser app in 75 revisions, and he lights up describing how she independently discovered abstractions like modeling “groups” instead of hardcoding “kids” and “adults.” That feeds into his bigger point: schools are too fixated on AI cheating and not focused enough on teaching students why writing matters, how to defend ideas in person, and how to spot bad information — skills he says matter just as much for social media as they do for hallucinations.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
