Chain of Thought Insights: Dan Shipper’s Articles on Every.to

Extracted from Dan Shipper’s Chain of Thought column at Every.to. Dan Shipper is CEO and cofounder of Every.

1. The Two-Slice Team

Published: February 13, 2026 (Updated March 10, 2026)

Summary: Dan Shipper argues that Amazon’s “two-pizza rule” (teams of ~10 people) is obsolete. With modern AI, the optimal unit is the “two-slice team”—a single person operating as a product leader. Every runs 4 software products, each led by one person, with 20 full-time employees total and 99% of code written by AI agents.

Organizational Insights

Every has 4 software products, each run by a single person, and 6 business units total
20 full-time employees across the company
99% of code is written by AI agents
Monologue (smart dictation app): Led by Naveen Naidu. 30,000 daily transcriptions, 1.5 million words transcribed daily, 143,000 lines of code. Naveen wrote nearly all code himself using Codex and Opus. AI assists with customer service, market research, business strategy, and product decisions.
Spiral (AI writing helper): Content repurposing tool
Cora (AI email assistant): Led by Kieran Klaasen. Processes millions of emails daily. Employs an external senior full-stack engineer (part-time). AI handles infrastructure challenges too complex for single-pass model solutions.
Proof (agent-native markdown editor): Built by Dan Shipper in spare time. Enables human-AI collaboration on markdown documents. Tracks authorship/provenance. Traditionally would require 3-4 engineers for 6 months. Available to paid Every subscribers.

Team Structure: Internal Agencies

Rather than adding full-time staff to products, two-slice teams pull from flexible shared resources
Design team: Led by Lucas Crespo, 3 people
Growth team and Marketing team also serve as internal agencies
Team members rotate across projects (e.g., a designer works on Monologue UI one week, Spiral promo banner the next, Cora templates simultaneously)
External specialized freelancers brought in for specific challenges

Values/Philosophy

Speed, autonomy, and variety are the three key advantages of this model
Single leaders can accomplish work typically requiring 3-4 people pre-AI
AI enables rapid onboarding without requiring product leaders to context-switch
The model depends entirely on modern AI capabilities

Key Quotes

The company functions as a coordinated “two-pizza company” through shared resources, understanding, and coordination mechanisms (observation from commenter Michael Fisher)

2. Compound Engineering: How Every Codes With Agents

Published: December 11, 2025

Summary: “Compound engineering” is Every’s four-step process where AI agents handle code generation while human engineers orchestrate and learn. The methodology challenges assumptions that coding is inherently difficult and engineers are scarce resources. A single developer can do the work of five developers from a few years ago.

Organizational Insights

Every operates five in-house software products, each managed by a single person: Cora, Spiral, Sparkle (Mac file organization), Monologue
Products serve thousands of daily users for substantive work—not prototypes

The Four-Step Compound Engineering Loop

Plan (80% of developer time): Agents research the codebase and commit history. Agents survey the internet for best practices. Output: detailed implementation plan documenting objectives, proposed architecture, specific code approaches, research sources, and success criteria. Purpose: build shared mental model between developer and agent.
Work (20% of developer effort): Agent transforms plan into task list and builds incrementally. Uses model context protocols (Playwright, XcodeBuildMCP) enabling agents to simulate user interaction during development. Agent iterates on code, observes issues, and self-corrects.
Assess: Traditional tools (linters, unit tests, manual testing). Automated code review agents examine work from multiple perspectives. Their compound engineering plugin deploys 12 parallel subagents checking security, performance, complexity, and bloat. Developer decides which findings require fixes.
Compound (the core innovation): Document learnings—bugs, performance issues, problem-solving approaches. Store as prompts within codebase or plugins. Automatically distributes knowledge to entire team. New hires access institutional memory without onboarding delays.

Tools Used

Claude Code (primary)
Factory’s Droid
OpenAI’s Codex CLI
Custom compound engineering plugin for Claude Code

Values/Philosophy

Planning is 80% of the work; execution is 20%
Knowledge should compound across teams through automated documentation
Traditional hiring signals (offline coding assessments) are obsolete
Multi-week new hire ramp periods are eliminated
Technology lock-in due to legacy code complexity is reduced

Key Quotes

“A single developer can do the work of five developers from a few years ago”

3. The Knowledge Economy Is Over. Welcome to the Allocation Economy.

Published: January 18, 2024 (Updated March 24, 2026)

Summary: AI transforms the economy from knowledge-based to allocation-based. In the age of AI, every maker becomes a manager. The skills that matter shift from “what you know” to how you allocate AI resources—vision, taste, quality evaluation, and project management.

Organizational Insights

ChatGPT changed Shipper’s understanding of work: summarizing information was an invisible skill bundled into “intelligence.” Now AI handles summarization, and intelligence “learns to be the thing that directs or edits summarizing, rather than doing the summarizing myself”
Only approximately 1 million U.S. managers (12% of workforce) currently develop allocation skills
Even junior employees will be expected to use AI, forcing them into the role of “model manager”

Values/Philosophy

The knowledge economy rewarded “what you know and your ability to bring it to bear in any given circumstance.” With AI matching human knowledge retrieval, this advantage disappears.
AI is “an abstraction layer over lower-level thinking” (Evan Armstrong’s framework)
Critical management competencies for the allocation economy: vision and taste development, work quality evaluation, resource direction capability, project timeline estimation

Key Quotes

“In the age of AI, every maker becomes a manager.”

4. Why Generalists Own the Future

Published: September 6, 2024 (Updated February 21, 2026)

Summary: Generalists possess a critical advantage in the AI era. Their true strength is not shallow breadth of knowledge but the ability to adapt to new situations, formulate the right questions, and operate in uncertain (“wicked”) environments where rules are unclear and feedback is delayed.

Organizational Insights

Generalists are characterized as curious professionals who move across domains, enjoy problem-solving in uncertain areas, and excel at combining diverse knowledge for novel challenges
LLMs are proficient in “kind” environments (immediate feedback, clear repetitive patterns) but limited in novel problem-solving
The competitive advantage of generalists centers on question formulation rather than immediate answers

Values/Philosophy

“Measuring [generalists] against their rudimentary coding abilities or their working knowledge of French baking technique misses their true advantage: the ability to adapt to new situations”
References David Epstein’s Range: distinguishes “wicked” environments (where generalists thrive) from “kind” environments (where specialists and LLMs excel)
Generalists identify problems specialists may overlook entirely

5. Working Right at the Ragged Edge of AI

Published: May 24, 2024 (Updated January 30, 2026)

Summary: An interview with Microsoft CTO Kevin Scott about building enduring businesses in the AI era. Scott was the first person at Microsoft to recognize the importance of AI and instigated its OpenAI partnership. (Note: Most content behind paywall.)

Organizational Insights

Kevin Scott was “the first person at Microsoft to recognize the importance of AI and instigated its OpenAI partnership, which has become the single most important strategic move at Microsoft since the dot-com era”
Scott expressed optimism that “in the next two to three years, a model will exist that can synthesize a Kevin Scott, and virtual Kevin can go be CTO of Microsoft”

Values/Philosophy

Building at the edge of AI capabilities means embracing uncertainty and rapid capability shifts

6. When AI Gets More Capable, What Will Humans Do?

Published: June 28, 2024 (Updated February 5, 2026)

Summary: Introduces “capability blindness”—humanity’s tendency to assume technological stasis despite rapid progress. As AI becomes more capable, humans shift from “sculptors” (directly creating work) to “gardeners” (managing conditions for AI output). What remains valuable: original research, original long-form thinking, audience acquisition, and consistency.

Organizational Insights

Claude 3 Opus significantly outperforms GPT-4 in voice replication and content adaptation
“Claude mostly wrote this tweet…though I edited it”
Claude struggles with original long-form content because “complexity scales superlinearly with the length of a piece of text”
Every’s development of Spiral reflects the strategy of building tools for AI-assisted content adaptation
Every has a consulting practice “focused on helping mid-to-large-sized organizations implement AI and train their workforce to adopt it”

What Remains Valuable in AI-Generated Content World

Original Research: “Uncovering new facts is still going to be valuable. Writing them will be much more of a commodity.”
Original Thinking (Long-form): Content requiring genuine original thought outside well-defined formats
Audience Acquisition: “More and more power will accrue in those companies that have novel acquisition methods that do not rely on any gatekeeper” (Evan Armstrong)
Consistency: Sustained publishing builds mindshare

Values/Philosophy

The gardener metaphor: a gardener doesn’t grow plants directly but “sets up the conditions for the garden to grow.” Humans become “model managers” who establish optimal conditions for AI output, then selectively apply direct human skill (“sculpting”) when it genuinely matters.
Warns against “capability blindness”—dismissing AI progress based on past failures
Cites Ross Douthat’s 2012 prediction that Facebook couldn’t monetize (proven spectacularly wrong by Meta’s $134B revenue in 2023)

7. The Magic Minimum for AI Agents

Published: July 15, 2025 (Updated March 26, 2026)

Summary: Introduces “the magic minimum”—a new business viability metric for AI agents that challenges the traditional “toothbrush test” (products used twice daily). AI agents can build billion-dollar businesses through periodic high-value interventions rather than daily engagement, because proactive agents don’t need users to remember to open them.

Organizational Insights

Every implemented an internal version of Spiral (agentic ghostwriter) in their Discord server
The agent reads daily channel conversations and proactively pings Shipper with tweet pitch ideas derived from internal discussions
Sparkle (file management) runs background Mac organization; users rarely open it directly
The “allocation economy” creates expanded opportunities for specialized tools previously limited by human memory constraints

Values/Philosophy

Proactive agents fundamentally alter software economics: “they’ll be able to work on their own and pop up when they’ve done something valuable”
Before AI, users defaulted to feature-rich platforms like Notion/Excel over specialized occasional-use tools simply because they forgot specialized alternatives existed
Professional services analogy: therapists, dentists, accountants maintain relationships through periodic contact and proactive outreach, not daily interaction

Specific Team/Workflow Details

Every offers: AI training, adoption consulting, plus tools (Spiral, Sparkle, Cora, Monologue) bundled in subscription model

8. Agent-Native Architectures: How to Build Apps After the End of Code

Published: January 9, 2026 (Updated February 17, 2026)

Summary: Contrasts traditional software architecture (skyscrapers—precise, load-tested, brittle) with agent-native architecture (gardens—organic, growing, adaptive). Agent-native apps have an AI agent as the core rather than code, with features defined by prompts naming desired outcomes rather than procedural steps. Every pivoted its whole software strategy around this paradigm.

Organizational Insights

Every “pivoted our whole software strategy” around agent-native architectures
The company published: a comprehensive guide to agent-native architectures, a skill module for a compound engineering plugin for Claude Code, and supporting documentation
Agent-native architecture: “the core is an agent—something squishy and alive, planted in sun and soil”
These apps are described as “Claude Code in a trenchcoat”

Values/Philosophy

“Developers only have to name the what” while the agent handles the how
Users can modify app behavior through natural language
This approach “levels the playing field of power”—democratizing software creation
Traditional software’s control paradigm is questioned: “How can we allow software so much freedom?”
Future generations might view precise, controlled skyscraper-era software “wistfully” as “a gorgeous, elaborate attempt to impose total control on a world that doesn’t want to be controlled”

Decision Details

Pivoting entire software strategy to agent-native was a company-wide strategic decision
Resources published: every.to/guides/agent-native

9. How I Built Spiral

Published: July 26, 2024 (Updated January 14, 2026)

Summary: Documents Dan Shipper’s experience building Spiral, a prompt builder tool, through collaborative development with Claude (Anthropic’s AI). The article serves as a behind-the-scenes narrative of this building process. (Detailed methodology behind paywall.)

Organizational Insights

Spiral is positioned as a prompt builder tool
Built through AI-assisted development (collaborative process between Shipper and Claude)
Part of Every’s product suite alongside Sparkle, Cora, and Monologue
The complete chat conversation between Shipper and Claude used to develop Spiral’s original code is available to paying subscribers

Values/Philosophy

CEO personally builds products using AI coding tools
Demonstrates the “builder CEO” philosophy where leadership involves hands-on product development

10. AI Can Do My Email Now

Published: June 30, 2023 (Updated February 4, 2026)

Summary: A personal review of Superhuman’s AI beta features for email management. Shipper explores how AI tools address his tendency to overwrite verbose, carefully-crafted email responses. Praises Superhuman’s restrained, invisible-by-default approach to AI integration.

Organizational Insights

Shipper’s email writing style: tendency to overwrite with excessive consideration, “lots of exclamation points and smiley faces,” covers “every corner case” when replying
He frames one-line emails as demonstrating confidence, command, and psychological detachment

Values/Philosophy (on Product Design)

Praises Superhuman’s design philosophy of making AI “totally invisible unless I snap my fingers”
Good AI integration is non-intrusive and prevents AI from becoming “messy and chaotic” in the user interface
Restraint in AI feature design is a virtue

11. We’re Building AI Into Our Media Business

Published: May 5, 2023 (Updated January 14, 2026)

Summary: Every’s comprehensive strategy for integrating AI into its media business. Covers four predicted AI-driven media transformations: de-risked content automation, unbundling research from narrative, unlocking unwritten stories, and human-AI collaborative creative work. Includes concrete operational examples of how Every already uses AI.

Organizational Insights

Every’s founding mission: “create more high-quality business writing in the world”
Co-founder Nathan built Lex.page, an AI writing app
Every built chatbots enabling readers to question written content

The Digest Summarization Project

Lucas Crespo (manages ad and course operations) built an automated summarization tool in Python deployed on Heroku within two days despite having zero prior programming experience
He used GPT-4 to create the application by describing requirements and requesting help when encountering obstacles
Workflow: input headline, author, and article text to generate summaries modeled on Every’s previous Digest summaries
Shipper still dedicates time editing summaries, occasionally rewriting when quality falls short

Four Predicted AI-Driven Media Transformations

De-Risked Content Automation: Seven de-risking strategies become easier for AI: summarizing others’ work, content arbitrage, format structures, shorter content, beat-focused reporting, sequels/remixes, interview-based content
Unbundling Research from Narrative: Expose underlying transcripts, searches, editorial conversations through chatbots; let readers interrogate source material and request customized story versions
Unwritten Stories Becoming Published: “The best ideas in business are never written down—they’re locked up in people’s heads.” AI enables converting transcripts into polished pieces, enabling “the dramatic expansion of the number of high-quality posts we can publish from people who don’t consider themselves writers but who have incredible ideas about business”
Human-AI Collaborative Creative Work: AI won’t eliminate writers but will reshape required skills. Technologies grant humans additional time for “satisfying creative work at a higher level of ambition and scale” while making media businesses “significantly easier to run predictably”

Values/Philosophy

“The best way to think about things like this is to experiment”
Hands-on experimentation over theoretical resistance to AI
Quality matters: GPT-4’s summarizations “still lack the zing and voice of truly great human writing”
Great writing—even de-risked formats—remains “a complex activity where what it means to be ‘good’ is changing all of the time”
Human oversight remains essential “for the foreseeable future”
On Hollywood AI fears: “it’s right to be hopeful that Hollywood and its screenwriters can find an equilibrium that uses AI to augment writers and compensate them better—rather than ban it entirely”

12. A Few Things I Believe About AI

Published: April 20, 2023 (Updated January 25, 2026)

Summary: Four core beliefs about AI after 6 months of intensive engagement. Knowledge orchestration is the primary bottleneck. End-to-end interaction data drives model improvement. Startups integrating entire processes will dominate. AI will progress where traditional science has struggled.

Organizational Insights

Belief 1: Knowledge Orchestration is the Primary Bottleneck

GPT-4 is “quite good at reasoning” but constrained by ability to provide relevant knowledge at the right time
Three operational layers: foundational model layer (context windows), developer tools/infrastructure (LlamaIndex, Langchain), vector databases (Pinecone, Weaviate, Chroma)

Belief 2: End-to-End Interaction Data Drives Model Improvement

Valuable knowledge includes “end-to-end interaction data about the lifecycle of any process”—visible beginning, middle iterations, and measurable results
Enables RLHF and fine-tuning

Belief 3: Startups Integrating Entire Processes Will Dominate

“Startups that are horizontally integrated over a process will be dominant”
More process visibility enables better improvement
Horizontal integration means “replacing things that their customers are paying other companies for with their own solution—so they can see more of the process and integrate it together”
Examples: Replit (idea-to-software in browser), OpenAI/ChatGPT (API to direct consumer), Midjourney (generation to editing to social), Adobe Firefly
Andrej Karpathy: the internet contains primarily “final” text; “all of that ‘latent structure’ of your drafts, edits, going back and forth etc. is sadly lost”—this missing iteration data is “ideal data for GPTs”
Incumbents face privacy concerns and internal political resistance when sharing data across ecosystem products; startups built for process integration have structural advantages

Belief 4: AI Will Progress Where Science Has Struggled

AI enables predictions in domains where simple causal scientific explanations remain elusive: psychology, social sciences, economics
Cancer screening and anxiety/depression prediction without requiring complete scientific understanding

Values/Philosophy

Adopting “a 10,000 foot perspective” amid rapid iteration cycles
Describes the pace as having “a SpaceX rocket strapped to my butt” while “you constantly feel behind”

13. Capability Blindness and the Future of Creativity

Published: April 12, 2024 (Updated December 17, 2025)

Summary: Introduces “capability blindness”—a cognitive bias where people fail to recognize improvements in technology because they’re jaded by previous experiences. Demonstrates Claude 3 Opus’s writing capabilities at 70-80% voice accuracy. Poses the question: “What’s rare in a world with infinite good writing?” (Partially paywalled.)

Organizational Insights

Claude 3 Opus can capture a writer’s voice at 70-80% accuracy with proper prompting
Shipper generated a tweet using Claude with examples from his podcast transcripts and tweets
“Claude mostly wrote this tweet…though I edited it”
Genuine capability leap between GPT-4 and Claude 3 Opus

Values/Philosophy

Capability blindness: people assume the world remains static rather than continuously evolving
Historical precedent: Ross Douthat’s 2012 NYT analysis declared Facebook unprofitable—by 2023, Meta achieved $134B revenue
The subtitle “We used to be sculptors. We’re all about to be gardeners” captures the transition from direct creation to condition-setting

14. The Great AI Unbundling

Published: August 1, 2024 (Updated February 7, 2026)

Summary: General-purpose AI chatbots will generate startup opportunities just as Excel spawned the B2B SaaS era. ChatGPT and Claude increase demand for “wrappers” as users discover use cases general tools aren’t designed for. Spiral is presented as a case study of this unbundling pattern.

Organizational Insights

Spiral launched approximately one month before publication, already achieving “more than 3,500 users and is growing every day”
Spiral automates “repetitive creative work like X posts, LinkedIn posts, headline creation, and product release notes”
Spiral uses Claude as its backend, but Claude’s chat interface lacks optimization for Spiral’s specific workflows (maintaining complex prompts, team prompt-sharing)
Every’s ecosystem: Chain of Thought column, AI & I podcast, software (Monologue, Sparkle, Spiral, Cora), subscription model, community of 100,000+ users described as “leaders, builders, and innovators”

Values/Philosophy

Challenges the pejorative connotation of “AI wrapper” applications: “ChatGPT and Claude are great places to find startup ideas”
The three-phase unbundling pattern: (1) general-purpose tool adoption, (2) discovery of niche workflow inefficiencies, (3) purpose-built application creation
Startup ideation framework: “If you want to build a startup in the AI era, just watch how you use ChatGPT or Claude. If you’re doing something repeatedly—and the outputs are impressive—there’s a chance it could be its own app”
Excel unbundled into Asana, Looker, QuickBooks over 15 years—AI chatbots will follow the same pattern

15. Five New Thinking Styles for Working With Thinking Machines

Published: October 10, 2024 (Updated March 26, 2026)

Summary: AI requires fundamentally different cognitive approaches than traditional scientific thinking. Western culture defaults to scientific/rationalist frameworks, but “thinking machines” demand new intellectual methodologies. The article introduces at least two of five new thinking styles. (Partially paywalled.)

Organizational Insights

References Andrej Karpathy’s “Software 1.0” (human-written instructions/scientific thinking) vs. “Software 2.0” (trained models searching solution spaces/engineering thinking)
The subtitle: “AI is turning science into engineering—with unprecedented changes for how we see the world”

Two of Five Thinking Styles Revealed

Essences vs. Sequences: Pre-AI approach emphasized identifying core elements and building forward. New approach: identify sequential patterns in behavioral data without predetermined definitions (e.g., SaaS churn prediction from 100 days of user behavior vs. explicit rule-based churn indicators)
Rules vs. Patterns: Pre-AI software required encoding explicit task definitions. New approach: provide mood boards or high-level behavioral descriptions, let AI extract underlying patterns and generate rulesets

Values/Philosophy

The shift from rule-based to pattern-based thinking is fundamental
Teams can now describe creative intent rather than specifying precise rules

16. Seeing Like a Language Model

Published: October 3, 2025

Summary: Language models reveal a fundamental shift from the Western rationalist worldview (logic, documentation, universal truth, simplicity) to a new pattern-based worldview (interconnection, context-dependence, correlation, paradox). Part of Shipper’s book project about the worldview developed by “writing, coding, and living with AI.”

Organizational Insights

Part of Shipper’s book project about AI worldview

The Old Western Worldview (Five Assumptions)

Logical clarity solves any problem
Explicit documentation enables understanding
Truth is universal regardless of context
Logic cannot contradict itself
Simplicity indicates truth

The New Worldview Framework

Reality as Web, Not Chain: Interconnected relationships, not linear cause-effect
Meaning Through Contrast: “Dark” means differently in “dark chocolate” vs. “dark thoughts”
Incomplete Explanations: “When everything is connected to everything else, any attempt to explain one thing requires explaining everything”
Correlation Over Causation: Complex systems resist if-then reduction
Context as Essential: Context shapes meaning fundamentally, not noise to filter
Participatory Knowledge: Observer perspective shapes observed reality
Pluralism Over Monotheism: Multiple frameworks capture different aspects; no single theory explains everything
Embracing Paradox: Uncertainty as creative force, not obstacle

Values/Philosophy

“Our tools change how we see the world…now we’ve built new tools—language models—that can work with what previously only our intuition could handle”
“If you believe that the world is composed of atoms rather than spirit…it’s because of this style of thinking”
The symbolic AI failure (trying to encode intelligence through formal rules) mirrors the limits of the Western worldview

17. Seeing Business Like a Language Model

Published: October 17, 2025 (Updated February 5, 2026)

Summary: Business principles function as useful heuristics rather than universal laws. Disruption theory, Porter’s Five Forces, and jobs-to-be-done analysis operate as situational patterns, not scientific axioms. Language models’ “next-token prediction” methodology offers superior business intuition versus rigid principle application.

Organizational Insights

Part of Shipper’s book series on AI worldview
Shipper is developing a worldview through “writing, coding, and living with AI”

Why Business “Laws” Fail

Equifinality and Multifinality: Multiple paths to identical outcomes; identical paths producing divergent results (Apple iPhone disrupted from the high end, contradicting disruption theory; Netflix competed directly with Blockbuster’s core)
Lack of Controlled Experiments: “What you choose to test is significantly more important than which test wins.” Strategic decisions cannot be A/B tested.
Sensitive Dependence on Initial Conditions: Butterfly effects—changing any variable in Netflix’s rise alters results entirely
Hidden Exceptions and Subjectivity: Conditions enabling principles cannot be fully specified in advance. Customer research lacks objectivity.

Values/Philosophy

“Business principles work—until they don’t”
“Business ‘laws’ are more like patterns or tendencies that we observe, but they don’t have the same predictive power or universal applicability” as physical laws
Memorizing frameworks doesn’t generate application instincts when situations demand them
The solution: build “gut-level next token prediction” rather than treating business maxims as universal rules
Pattern recognition and tacit knowledge transcend explicit principles

18. Seeing Science Like a Language Model

Published: October 10, 2025 (Updated January 12, 2026)

Summary: Second piece in Shipper’s book series. Argues that language models reveal what centuries of scientific method missed: some truths resist reduction. Uses the psychology replication crisis as a case study for the limits of reductionist scientific thinking. (Partially paywalled.)

Organizational Insights

Written from Bocas Del Toro, Panama, from “a little cabin in the jungle at the top of a hill five minutes from the Caribbean sea”
Part of Shipper’s book project: “I’m writing a book about the worldview I’ve developed by writing, coding, and living with AI”

Key Arguments

“The way things appear is different from the way they are—we trust the science”
Enlightenment physics concepts permeate modern language: “forces” shaping society, “momentum” in markets, “chemistry” with people, “orbit,” “leverage,” “friction”
“This way of thinking has also begun to reach its limits. Just as it did with machine learning research, the tendency to reduce, to break apart, to make explicit, and to generalize has failed, hard, in many places”
Psychology’s replication crisis: “Scientists use tools built for physics to make their intuitive theories fit into statistical models, only to find that their results can’t be reproduced”
The issue is “not necessarily a crisis of replication but one of generalizability” (psychologist Tan Yarkoni): we overstate how universal findings are

Values/Philosophy

“That we see the world the way we do is a testament to how powerfully ideas shape what we see, and our capacity, through cultural transmission, to know things that we have not directly experienced”
References David Deutsch’s “hard-to-vary” explanations as the currency of science
The subtitle: “Language models reveal what centuries of scientific method missed: Some truths resist reduction”

19. Seeing Creativity Like a Language Model

Published: October 24, 2025 (Updated February 6, 2026)

Summary: Fourth piece in the book series. AI reshapes creative work by elevating creators to allocation roles—from “sculpting to gardening.” Includes deeply personal account of Shipper’s 2023 organizational crisis and reorientation. Presents six concrete AI applications for creative work and argues AI enables a return to Athenian generalism.

Organizational Insights

Personal Crisis and Organizational Restructuring (2023)

Every faced stagnant growth, departed co-founder, personal identity crisis between entrepreneurship and writing
Shipper queried GPT-3: “writers who also have great businesses”
System surfaced role models: Sam Harris, Bill Simmons, Ezra Klein, Nate Silver, John Green
Restructuring decision: Made writing the first half of each workday non-negotiable. Hired talented executives to handle operations.
Result: “The business subsequently took off in a way that it never had before”

The “Ineffable List” System

Shipper maintains a phone note documenting moments that resonate—quotes, observations, scenes
Examples: nephew’s question “Where does the wind go?”, Chekhov excerpt, family cemetery business process (“The System”)
Functions as personal training data for creative work

Six AI Applications for Creative Work

Understanding dense material: ChatGPT with image capture and voice mode to translate philosophy, ML texts, literature
Extracting ideas faster: Voice dictation converts mental “potential energy” into linear text
Rapid summarization: AI generates foundational summaries (half-day research to seconds), leaving refinement to taste
Multi-perspective editing: AI simulates various editorial voices; aggregates reader feedback using tools like Rally
Cross-platform adaptation: Long-form to social media versions, thumbnails, headlines for X, LinkedIn, YouTube
Prototype construction without capital: Spiral built in days rather than months, no investor funding needed

The Allocation Economy Framework (Expanded)

Required skills: vision and taste in ideas and language, effective communication of intent, project planning and timeline estimation, task distribution among specialists, strategic micromanagement for detail quality
“These are the skills of human managers”—showrunners, directors, conductors operate identically
Currently only ~10% of workforce fills these roles; future requires universal adoption

Historical Arc: Generalism

Ancient Athens: Citizen generalists (farmer, lawyer, juror, warrior simultaneously); cultural golden age
Specialization era: Adam Smith’s pin factory increased productivity 2,400x (20 to 48,000 pins/day) through division of labor
Cost: Karl Marx’s alienation; Henry Ford’s assembly line workers quit despite doubling pay
AI as Resolution: AI provides “effectively a thousand PhD-level experts in your pocket,” enabling return to Athenian generalism without sacrificing depth

Values/Philosophy

“If I was really honest with myself, I actually wanted to be a writer”
AI-generated art concerns: contests the rationalist view that art equals “text on page or pixels in image—completely divorced from context and history.” Counter: context is determinative.
Cultural taboo around AI art signals fertile creative ground for “brave creatives who embrace it to produce novel works”
“You need vision, a taste for ideas and for language, and an ability to effectively communicate what you want”
“AI can enable us to expand our creative and professional repertoire, breaking free from narrow specialization”

20. Admitting What Is Obvious

Published: September 8, 2023 (Updated January 13, 2026)

Summary: A deeply personal essay about how people suppress obvious truths about themselves due to internalized expectations. Shipper reveals that he founded Every partly as a mechanism to pursue writing while maintaining founder credibility, and that admitting “I actually wanted to be a writer” was the turning point.

Organizational Insights

Shipper previously sold a B2B software business before founding Every
Initially described himself as a “founder who also liked to write” rather than admitting his primary identity
Writing was his greatest passion despite enjoying most startup responsibilities (coding, sales, marketing, managing, fundraising)
Every functions structurally as a startup (permitting founder status) while functionally enabling Shipper’s authentic professional identity as a writer

Values/Philosophy

References Robert Bly’s “invisible bag” concept: people carry rejected parts of themselves since childhood
The bag accumulates authentic desires deemed unacceptable by parents, teachers, peers
Admitting obvious truths feels terrifying because it appears to threaten professional identity, ability to build consequential companies, career earning potential
The path to freedom: acknowledging what you already know about yourself despite potential consequences

Key Quotes

Shipper realized “writing was his greatest passion” but had suppressed this truth
Every was created partly so he could write while maintaining the identity of a founder/entrepreneur

Cross-Cutting Themes

How Dan Shipper Thinks About Running Every

Company Structure

20 full-time employees, 6 business units, 4 software products each run by 1 person
99% of code written by AI agents
Internal agency model: design (3 people, led by Lucas Crespo), growth, marketing teams rotate across products
External freelancers for specialized challenges
100,000+ community members

Products

Spiral: AI writing/content repurposing tool (agentic ghostwriter), 3,500+ users at launch
Cora: AI email assistant, processes millions of emails daily
Monologue: Smart dictation, 30K daily transcriptions, 1.5M words/day
Sparkle: Mac file organization, background operation
Proof: Agent-native markdown editor for human-AI collaboration

Core Philosophy

Allocation over knowledge: The future belongs to those who direct AI resources, not those who possess knowledge
Generalists over specialists: Ability to adapt, ask the right questions, and work across domains
Gardening over sculpting: Set conditions for AI output rather than doing everything directly
Experimentation over theory: “The best way to think about things like this is to experiment”
Compound knowledge: Document learnings as prompts that automatically distribute across the team
Authentic identity: Admitting what you truly want (writing) and building a company around that truth
Agent-native architecture: Software where the core is an AI agent, not code
Planning dominates execution: 80% planning, 20% building in the compound engineering loop
Speed and autonomy: Two-slice teams ship fast and pivot quickly
AI wrappers are opportunities, not liabilities: General-purpose AI creates demand for purpose-built tools

Decision-Making Style

Uses AI to surface patterns he already intuitively knows (GPT-3 query for “writers who also have great businesses”)
Makes writing the first half of each workday non-negotiable
Pivots entire software strategy when paradigm shifts (agent-native architectures)
CEO personally builds products (Proof built in spare time)
Hires talented executives to handle operations so he can focus on writing and product

Evolution of Every

Founded as media business to “create more high-quality business writing in the world”
Co-founder Nathan departed; Shipper faced stagnant growth and identity crisis (2023)
Restructured: writing first, hired executives for operations
Business “took off in a way that it never had before”
Expanded into AI software products (Spiral, Cora, Monologue, Sparkle, Proof)
Added consulting practice for mid-to-large organizations
Pivoted software strategy to agent-native architectures (2026)
Developed compound engineering methodology for AI-first development