A daily reading list pulled from newspaper and newsletter inboxes, scored against tech/AI, business, politics, and culture. Rebuilt every morning at 8am.
Listen
Audio playlist
4 tracks · supplemented from regular audio newsletters · auto-advances
RL
Select a track
—
1
Is AI Going to Destroy Our Lives or Not?
kyla’s newsletter · narrated essay on living through the AI buildout
~22m
2
Open Tab: Hunter Harris
Open Tab (Substack) · the culture writer on what she is reading
39m
3
Open Tab: Lachlan Cartwright
Open Tab (Substack) · the reporter on media, scoops, and the press
48m
4
The Walt Disney Company
Acquired · the full history of Disney, from Walt to Iger
~4h30m
Saved
Bookmarks from Nitzan
Links Nitzan dropped into the To Read database. Synced fresh each morning from Notion.
Nvidia’s $750 Billion Deals Revive Fear of AI Circular Financing
Bloomberg·Morning Briefing Americas·Jul 27·6 min read
Nvidia is pursuing a fresh round of AI infrastructure deals that could total more than $750 billion, including talks over roughly $250 billion in backing for OpenAI’s lease of compute capacity at SoftBank’s planned $500 billion Ohio data center campus. The scale is unprecedented even by the standards of the current buildout.
The concern: the chipmaker is increasingly financing its own customers — the circular-financing pattern that has historically marked the late stage of capital-spending booms.
The piece traces how each new commitment deepens the loop: Nvidia’s capital supports the data centers whose operators then buy Nvidia hardware, with the same dollars showing up on both sides of the ledger. Skeptics hear echoes of past bubbles; bulls argue the demand underneath is real and still compounding.
The same session produced a sharp counterpoint from the other side of the chip decoupling race: Chinese memory maker CXMT surged 466% in its Shanghai debut after a $9.8 billion IPO, instantly becoming China’s largest onshore-listed company — a reminder that AI-hardware enthusiasm is now a global, and geopolitical, phenomenon.
How Writers Are Reacting to Substack’s AI Transparency Tools
On Substack·Substack team·Jul 24·5 min read
Substack recently shipped a feature that lets readers scan any post to see how much of it is estimated to be written by hand versus with AI assistance, alongside publisher tools: pre-publication scanning of drafts, per-post opt-outs, and a “How I make this” statement for describing one’s creative process. This post rounds up the first wave of writer reactions.
The company’s position: the use of AI isn’t necessarily a problem, but a lack of transparency around it definitely is.
The roundup spans the spectrum — from an Economist columnist arguing AI prose should be shunned “even if it is perfectly bearable to read,” to novelists testing what LLMs can and cannot generate, to a literature professor examining machine poetry (“For poets, edge cases are the poem”).
The dissents are the most interesting part: writers raising the risk of false flags, the limits of what a label can capture about a piece of writing, and the worry that scanning turns a reading community into a process-scrutinizing one — one calls the scanner “a moral purity test in disguise,” noting a detector can estimate AI involvement but cannot tell you who had the original thought.
The Micro-Philosopher·The Micro-University·Jul 27·9 min read
A study guide to Parmenides of Elea, positioned as the second great philosopher after Heraclitus and the greatest of the Pre-Socratics. His significance rests on three claims: the earliest surviving evidence of deductive argument, original metaphysical and epistemological conclusions that still echo in modern philosophy, and the productive challenge his work posed to Empedocles, Democritus, Plato, and Aristotle.
The title he earns: inventor — or grandfather — of metaphysics.
The essay covers his life in Elea, the school he founded there, and the surviving fragments of his poem — especially fragment B8, the longest continuous stretch of argument to survive from any Pre-Socratic.
It closes on his three core questions — what can be said about existence itself, whether the senses deceive, and how thought relates to reality — questions that set the agenda Plato and Aristotle would spend their careers answering.
School of the Possible·Dave Gray·Jul 27·4 min read
Gray’s argument: making room for your next life inside your current one is a design problem, and the solution is deliberately building “wiggle room” — slack in time, money, and commitment that lets something new grow before it has to pay for itself.
The case study: a 1993 move from a declining newspaper job to a term-limited, part-time teaching post — less security, more room.
That constrained-but-open arrangement gave him both the space and the pressure to build a side project seriously; it became a company he ran for decades and sold in 2022. The essay reads the arc backwards to argue the side project was never really on the side — it was the main plot, waiting for room to run.
One prompt drives the whole thing. Copy it, paste it into a Claude scheduled task with the right connectors, point it at your own Vercel project, and you'll have your own RL by tomorrow morning.
Show the full prompt▾
# RL — Daily Reading List
Build a daily reading list from the recipient's newspaper/newsletter inbox,
deliver it as a Gmail draft, and redeploy a public web page.
============================================================
SETUP — replace these placeholders before first run
============================================================
RECIPIENT_EMAIL = "you@example.com"
RECIPIENT_NAME = "Your Name"
DEPLOY_URL = "https://your-rl.vercel.app"
VERCEL_PROJECT_ID = "prj_..."
VERCEL_ORG_ID = "team_..."
VERCEL_PROJECT = "your-rl"
VERCEL_TOKEN = "vcp_..." # vercel.com/account/tokens
CADENCE: every day at 8am local (cron: 0 8 * * *)
BRAND : "RL — Reading List"
INTEREST PROFILE — edit these to your taste:
- Topics: Tech/AI/startups, Business/markets/finance, Politics/policy,
Culture/longreads/essays
- Daily volume: 3–5 written picks (minimum 2)
- AUDIO MINIMUM: always ship at least 3 audio tracks in the playlist,
even on quiet days
- SAVED: pull all entries from a Notion "To Read" database; show every one
============================================================
STEP 1 — Read the Notion "To Read" database
============================================================
Use the Notion MCP. Search for the database, then list every page in it,
sorted by createdTime DESC. For each page, capture the Name and URL property,
infer the source from the URL hostname, and compute a relative timestamp.
============================================================
STEP 2 — Find newspaper/newsletter emails from the past 24 hours
============================================================
Use the Gmail MCP `search_threads`. Two passes:
newer_than:1d (from:nytimes.com OR from:wsj.com OR from:washingtonpost.com
OR from:economist.com OR from:ft.com OR from:bloomberg.com
OR from:theatlantic.com OR from:newyorker.com OR from:theinformation.com
OR from:axios.com OR from:semafor.com OR from:platformer.news
OR from:stratechery.com OR from:reuters.com OR from:theguardian.com)
newer_than:1d category:updates (newsletter OR briefing OR substack OR digest)
If the past 24h is light, expand to `newer_than:2d`.
============================================================
STEP 3 — Read promising threads
============================================================
`get_thread` with messageFormat=FULL_CONTENT.
If a body exceeds the token limit, delegate parsing to a general-purpose
subagent — give it the file path the error returns and ask it back for:
subject, sender, date, top stories (1-line each), and the first article URL
per story.
============================================================
STEP 4 — Score and pick 3–5 written
============================================================
Priority:
- Hits on the four topic areas above
- Sharp framework / argument pieces — ideas you can apply
- Writing, design, or creative-tooling angles
- Bloomberg: max ONE per day, only the most insight-dense piece
- Use the SAVED list as a signal — weight pieces that match recent saves
- Skip: ads, vendor newsletters, ticket alerts, generic music/club recs,
university dept newsletters, generic mindshift content
============================================================
STEP 5 — Build the audio playlist (minimum 3 tracks)
============================================================
For each pick, WebFetch the article URL and look for:
a) Direct narration:
/api/v1/audio/upload/<uuid>/src
b) Podcast embed:
meta-twitter:player: https://on.substack.com/embed/podcast/<slug>
curl the publication's RSS feed and grep:
grep -oE 'enclosure url="[^"]+\.mp3"' feed.xml
If you have fewer than 3 audio tracks from today's picks, supplement
from these regularly-audio newsletters by fetching their RSS and grabbing
the most recent episode that isn't already in the playlist:
- https://on.substack.com/feed (Open Tab interview series)
- https://kyla.substack.com/feed
- https://stratechery.com/feed
- https://feeds.transistor.fm/acquired
- any other Substack-narrated newsletters in the inbox
(search: from:substack.com newer_than:7d)
Capture for each track: title, source attribution, direct mp3 URL,
duration (from `<itunes:duration>` if available). Cap at 5 tracks max.
============================================================
STEP 6 — Create the Gmail draft
============================================================
`create_draft`. To: [RECIPIENT_EMAIL]. Do NOT send.
Subject: "RL — <Day>, <Mon DD>, YYYY"
Provide BOTH:
- body : plain ASCII version with URLs inline
- htmlBody : h3 headlines, paragraphs, hr separators, anchor tags
Include a "Saved" section listing the To Read entries.
============================================================
STEP 7 — Save a markdown archive
============================================================
Write to your outputs folder as `rl-YYYY-MM-DD.md`.
============================================================
STEP 8 — Rebuild and redeploy the page
============================================================
Build a fresh `index.html` matching the current production design at
DEPLOY_URL. Fetch that live HTML as your starting template. DO NOT change
CSS or layout — only swap content blocks.
Preserve in the new HTML:
- Reading progress bar (fixed top, JS that updates on scroll)
- Masthead with RL logo + "Reading List" wordmark + dateline
- Top nav with anchor links to #listen, #saved, #read, #build with counts
- Listen section with audio playlist (3+ tracks, custom player UI,
sticky mini-player JS)
- Saved section between Listen and Read — list every Notion entry,
with an empty state if the database is empty
- Read section with TOC (#piece-1, #piece-2, ... with topic + title +
read time)
- Each article: article-head with No. NN + topic, h2 title, meta line,
summary, "Open original →" link, full inline text, inter-article
"Next →" nextlink at bottom. No personal commentary — the page is shared.
- Mini-player div + script with play/pause/next/keyboard logic
- This "Build your own" section, with the same prompt embedded
Estimate read times honestly: word_count / 200 wpm, rounded to whole minutes
(min 1).
Then deploy via bash:
mkdir -p /tmp/rl-deploy
cp <path-to-new-index.html> /tmp/rl-deploy/index.html
echo '{"cleanUrls": true}' > /tmp/rl-deploy/vercel.json
curl -sL DEPLOY_URL/favicon.svg -o /tmp/rl-deploy/favicon.svg
curl -sL DEPLOY_URL/apple-touch-icon.svg -o /tmp/rl-deploy/apple-touch-icon.svg
mkdir -p /tmp/rl-deploy/.vercel
echo '{"projectId":"VERCEL_PROJECT_ID","orgId":"VERCEL_ORG_ID","projectName":"VERCEL_PROJECT"}' \
> /tmp/rl-deploy/.vercel/project.json
cd /tmp/rl-deploy
export VERCEL_TOKEN="VERCEL_TOKEN"
npx --yes vercel@latest deploy --prod --yes --token "$VERCEL_TOKEN"
============================================================
STEP 9 — Report to chat
============================================================
- # of newsletters scanned
- 3–5 written picks selected (titles only)
- 3+ audio tracks (titles + sources)
- # of saved Notion links pulled
- Gmail draft created (confirmation)
- Live URL after deploy
- Anything notable
============================================================
TONE
============================================================
Direct, no fluff, no emojis, tight.
Prerequisites: Claude desktop with Cowork, Gmail connector, Notion connector, a Vercel account and token, scheduled tasks enabled, and an SVG favicon at /favicon.svg. The first deploy creates the Vercel project; subsequent runs reuse it via .vercel/project.json.