Dispatch from the Future: building an AI-native Company (AI column)
Looked pretentious. report Dan Shipper, co-founder and CEO of Every He ran Every in. 2020 with Nathan Baschez, and Everyday - It's a team of 15 man 6 business units and 4 Products, publishes a daily newsletter about AI and consulting.
As for me, the scale of the company is small, but the theses are ambitious and interesting, so I gave them here.
1ок Quantum Leap Forward 100% adoption The difference between the organization, where 90Percentage of engineers use AI and organizations with 100The percentage of adoption is not 10Percent, 10x. If at least 10The percentage of the team uses traditional development methods, the whole organization rolls back to the old processes. The physics of development changes only with a complete transition.
2️⃣ 2. One Developer = Production Application Each of the four products is built by one engineer. These are not toys: Cora (AI email assistant) handles thousands of mailboxes, Monologue (speech-to-text) It is used to write millions of words a week. (AI writing partner) It generates content with millions of impressions. 99The percentage of code is written by AI agents – no one writes code with their hands.
3From Code Editor to Delegating to Agents The key change is moving from a code editor to a terminal-based workflow with Claude Code, which removes the traditional code editor and allows agents to delegate tasks. This opens up the possibility of parallel execution: developers launch 4 Windows with agents at the same time, working on different features.
4️⃣ Demo culture vs Memo culture When code becomes cheap, companies move away from memo culture. (Write documents and convince colleagues) To "demo culture" - you can make a prototype in a couple of hours and show it. This allows you to do more strange and interesting things that are difficult to describe in words, but easy to feel.
5️⃣ Compounding Engineering Every developed the Compounding Engineering methodology: each feature makes the next feature easier, not harder. The cycle consists of 4 stages: - Plan (40%)Agents study the code base and create detailed plans - Work (20%)Agents write code and tests - Review (20%)Quality assessment through tests, code review, subagents - Compound (20%)Codification of all learnings into prompts, subagents, slash commands
6и Secondary effects Full AI-adoption unlocks non-obvious benefits: - Tacit code sharingAgents can read repos of neighboring projects and transfer patterns to another stack without explicit libraries. Beginners are productive from day oneThe entire organizational knowledge base is encoded in claude.md files - Cross-app commitsDevelopers fix bugs in other people's products because it's just - Polyglot stack: Each product can use its own language and framework – AI handles translation **- Managers commit code.**Even CEOs can commit production code between meetings
7️⃣ Fractured Attention Programming AI allows you to work with “fragmented attention” when previously needed 3-4 hourly focus block. Now: left the meeting → gave the task to the agent → went to another meeting → returned to the finished result → did PR.
This entails the following changes for development 1) Changes in economic development - Parallelism instead of sequence: The developer works with agents 3-4 tasks simultaneously in different worktrees, and not with one, as it was before - Decrease in start-up costsPrototype-first approach becomes dominant Inversion of the developer roleCode is written by agents, developers become orchestrators
**2) Process transformation
- New primitives.**: agents.md files with project context, custom subagents for specific tasks, etc. appear (It is a way of codifying the knowledge of the organization.) Shift from documentation to artifactsAgents read code and other primitives directly Change in hiringNo more weeks onboarding and no more knowledge of a particular stack
#Engineering #AI #Metrics #Software #DevEx #Productivity #DevOps #Architecture #Culture #Engineering #ML #SystemDesign