Dispatch from the Future: building an AI-native Company – Dan Shipper, Every, AI & I

Building AI-Native Companies: Dispatches from the Future

Key Concepts:

• AI-Native Company: An organization fundamentally structured around leveraging AI for core engineering and product development.
• 10x Difference: The significant performance gap between organizations with partial vs. complete AI integration in engineering.
• Compounding Engineering: A methodology where each feature developed makes subsequent feature development easier, contrasting with traditional engineering.
• Tacit Knowledge: The unwritten, experiential knowledge held by engineers, now codifiable through AI-assisted processes.
• Cloud Code: Utilizing cloud-based IDEs and AI agents for code generation and manipulation, minimizing reliance on traditional code editors.
• AI Agents: Tools like Cloud Code, Codec, and Droid used for automated code generation, testing, and debugging.

---

The Shift to AI-Native: A 10x Impact

The speaker emphasizes a dramatic difference in organizational capability – a “10x difference” – between teams where 90% versus 100% of engineers utilize AI in their workflow. A partial adoption forces a reversion to traditional engineering practices, hindering the full potential of AI-driven development. This observation stems from the experience at Every, the speaker’s company, which has fully embraced AI in its engineering processes. The core argument is that achieving full AI integration unlocks capabilities unattainable with a hybrid approach.

Every: A Case Study in AI-Driven Development

Every, a company with just 15 people, operates six business units and manages four software products. Remarkably, the company has achieved double-digit month-over-month revenue growth for the past six months, boasting over 7,000 paying and 100,000 free subscribers, all while raising only $1 million in funding. Crucially, 99% of Every’s code is written by AI agents. This allows for an unprecedented level of output: each application is built and maintained by a single developer, despite their complexity.

Examples of Every’s products include:

• Kora: An AI-powered email management app featuring summarization and question-answering capabilities.
• Monologue: A speech-to-text application positioned as a more powerful alternative to Whisper or WhisperFlow.
• Spiral: Another application built and maintained by a single engineer.

These applications are not simple prototypes; they are fully functional, user-facing products demonstrating the viability of AI-driven development at scale.

From Code Editors to Task Delegation: The Paradigm Shift

The speaker attributes this success to the shift from traditional coding in code editors to delegating tasks to AI agents facilitated by tools like Cloud Code. This transition enables parallel work on multiple features and bugs simultaneously, overcoming the limitations of sequential development. The speaker acknowledges the “vibe coder” meme (appearing productive with multiple panes open but achieving little) but asserts that productive engineers are effectively utilizing multiple agent-driven workflows concurrently.

Compounding Engineering: Building on Success

The speaker introduces “compounding engineering” as a new paradigm for software development. Unlike traditional engineering where each feature adds complexity, compounding engineering aims to make each subsequent feature easier to build. This is achieved through a four-step loop:

1. Plan: Detailed planning is crucial when working with AI agents.
2. Delegate: Instructing the AI agent to execute the plan.
3. Assess: Evaluating the agent’s work through testing, review (including agent code review), and other methods.
4. Codify: The key step – transforming the knowledge gained from planning, delegation, and assessment into reusable prompts for future use. This creates a library of best practices and accelerates development.

The Second-Order Effects of Codification

Codifying knowledge into reusable prompts unlocks several benefits:

• Easier Tacit Knowledge Sharing: AI facilitates the transfer of unwritten knowledge between developers, even across different tech stacks. Instead of abstracting code into libraries, developers can directly learn from and adapt each other’s work using AI agents.
• Accelerated Onboarding: New hires become productive on day one with pre-configured environments and AI-assisted workflows.
• Efficient Freelancer Integration: Expert freelancers can quickly contribute to specific tasks without extensive onboarding.
• Cross-Product Collaboration: Developers readily contribute to other products, fixing bugs and implementing improvements across the organization.
• Stack Agnosticism: The ease of translation between languages and frameworks allows developers to choose the tools they prefer, rather than being constrained by standardization.
• Managerial Code Contribution: AI enables even non-coding managers to contribute directly to production code, leveraging fractured attention and AI assistance.

The Future of Development: Democratization and Collaboration

The speaker predicts a future where even customers may be able to contribute code fixes directly through AI agents, blurring the lines between developers and users. This vision highlights the potential for a more collaborative and democratized development process.

Every’s Offering & Conclusion

The speaker concludes by promoting Every, a subscription service offering access to AI tools, insights, and training. Every aims to be a comprehensive resource for organizations seeking to embrace AI-native development. The core takeaway is that fully embracing AI in engineering is not just about productivity gains; it’s about fundamentally changing how software is built and maintained, unlocking new levels of innovation and scalability.

Notable Quote:

“In traditional engineering, each feature makes the next feature harder to build. In compounding engineering, your goal is to make sure that each feature makes the next feature easier to build.” – Speaker, describing the core principle of compounding engineering.