How Every runs 5 products with 15 people using AI-native operations

Executive overview

Most companies treat AI as a productivity add-on. Every has restructured its entire operating model around it: no engineer writes code by hand, a dedicated AI operations lead automates repetitive work across the team, and new products are incubated from internal needs.

The result is a 15-person team shipping four products, a daily newsletter, and a consulting arm — all profitable on minimal outside capital.

The core insight: AI removes the ceiling on what a small team of generalists can build, but only if the whole operating model is rebuilt around it — not bolted on.

Hot takes on AI

• Claude code is the most underrated tool for non-technical people; it runs agents on local files autonomously for 20–30 minutes
• AI may be a force for reshoring American jobs by making expensive services (call centres, legal, advisory) affordable at small-company scale
• Using AI is a skill — studies comparing "doctors vs. AI" miss that doctors haven't yet developed that skill
• The "leash" you can give AI before intervening is the best measure of progress; AGI is when running agents indefinitely becomes economically profitable
• Entry-level workers who use AI from day one accelerate a year's worth of progress in two months — one Every writer is cited as evidence
• Generalists are the biggest beneficiaries: AI handles the specialist depth, freeing people to operate across many domains simultaneously

How Every operates

• Head of AI operations: a dedicated person (Katie Parrott) whose sole job is identifying repetitive tasks and building prompts and workflows to automate them across the whole team
• Editorial copy-edits now run through a Claude prompt against a style guide; a Claude Code command submits pull requests to GitHub automatically, then the editor reviews only the diff
• Engineers use Claude Code all day — no one manually writes code; prompts and PRDs are the primary engineering artefact
• Compounding engineering: for every unit of work, spend a small amount of effort making the next unit easier — build a prompt that converts rambling thoughts into a polished PRD rather than writing each PRD from scratch
• A shared GitHub library stores slash-command prompts that the engineering team reuse and iterate on
• Multiple agents are used simultaneously (Claude, Friday, Charlie) because different agents have different "personalities" and taste — like hiring a team of Avengers

Product incubation model

• Every builds products by first noticing internal needs — things that used to require expensive specialists (chief of staff, ghostwriter, lawyer)
• General-purpose tools (ChatGPT, Claude) are used first; if the use case proves out, it gets unbundled into a standalone app
• Internal usage is the primary success metric: "is it a banger inside Every?" before shipping to users
• Current suite: Cora (AI email chief of staff, 2,500 active users), Sparkle (AI file cleaner), Spiral (content automation), plus Lex (spun out)
• The whole bundle is sold for one price — no per-product pricing
• Total product cost for Cora: ~$300K, built by two engineers plus Claude Code agents

Fundraising model

• Raised $700K pre-seed with a note to investors: "this is probably not a venture business"
• Recent round: up to $2M from Reid Hoffman and Starting Line VC as a SIP (Seed Investment Programme) structure — capital committed but drawn down on demand at a set cap
• Rationale: seeing the full balance creates burn pressure; the draw-down model preserves psychological optionality and creative risk-taking
• At current AI leverage, far less capital is needed — teams of 2 engineers can do what previously took 20

Consulting arm: what separates AI winners from laggards

• Every's consulting business (~$1M last year, growing) trains large companies to adopt AI
• Process: interview each team to map repetitive tasks, produce a report with an interactive chatbot over the interview data, run four weeks of customised training, then build automations
• Single strongest predictor of success: does the CEO personally use ChatGPT daily?
  • CEOs who use it drive excitement and set realistic expectations
  • CEOs who don't use it either create negative culture or have unworkable expectations
• Tactics that work: "I wrote this email with ChatGPT" memo from the CEO; weekly meetings where employees share prompts; weekly stats email showing org-wide usage and highlighting power users
• Finding and rewarding the internal 10% of early adopters transfers their learnings to the 80% who will adopt if shown exactly how

The allocation economy

• In the knowledge economy, people are paid to do a thing; in the allocation economy they are paid to manage resources — including AI agents — toward goals
• Skills that become more valuable: evaluating output quality, having taste and vision, knowing when to delegate vs. dive in, dividing tasks across agents
• Only 8% of today's workforce are managers; AI makes management far cheaper, so more people will need these skills
• Generalists benefit most: AI provides specialist depth on demand, letting one person operate across many domains simultaneously — the Athens-era citizen model rather than the pin-factory specialist model

Building for the long term

• Every's stated mission: teach people to live a better, more human life with technology — through writing and through tools
• Dan returned to writing as the centre of his identity after a period of stepping back; business performance improved when he did
• The pattern of writers who build companies (Joel Spolsky, Jason Fried, Bill Simmons) is a real and under-recognised archetype
• The lesson: finding the shape of work that fits you — even if unconventional — outperforms cargo-culting standard startup playbooks