Composition Shouldn't be this Hard
A new company, Cambra, announces its mission to tackle the chronic brittleness and fragmentation plaguing modern internet software development. The author, an industry veteran, argues that current systems force a choice between powerful and general-purpose tools, leading to tedious, error-prone, and inflexible architectures. Cambra proposes a 'general-purpose, domain-aligned, sealed model' to revolutionize software composition, asserting that even advanced AI agents benefit profoundly from such coherent foundational systems.
The Lowdown
The author, drawing on a decade of experience building data infrastructure at companies like Twitter, Google, and Snowflake, identifies a persistent problem in software development: the gap between conceptually elegant programming and the reality of brittle, hard-to-change systems. They contend that while engineers invest heavily in testing and deployment, the core issue lies in fragmentation and the inability to work within a single, coherent model across an entire system. This leads to the announcement of Cambra, a new venture aiming to develop a programming system that rethinks the traditional internet software stack.
- Models as Superpowers: Programs function using abstract models to simplify reality. Great models enhance reasoning, verification, optimization, and refactoring, but foundational low-level models (bits and instructions) necessitate higher-level abstractions like programming languages, operating systems, and databases.
- Sealed and Domain-Aligned: The ideal scenario involves working within a 'sealed,' domain-aligned model that rarely requires dropping to lower levels, maximizing the leverage of specialized tooling.
- The Fragmentation Problem: Modern software is assembled from components with incompatible internal models, forcing developers to reason about interactions using low-level, domain-misaligned models (e.g., networks and operating systems). This results in brittle systems characterized by tedium, inflexibility, errors, and performance issues.
- Costs of Fragmentation: These include contract mismatches between services, difficulty performing cross-component optimizations, high ceremony and risk associated with changes (like database migrations), and impedance mismatches (e.g., incompatible type systems between databases and programming languages).
- Coherent Systems and Their Limits: 'Coherent systems' operate entirely within a single, domain-aligned model, offering significant benefits in tooling, productivity, correctness, and performance. Examples include type systems, relational databases, and web frameworks. However, these are often domain-specific, forcing systems to fragment as application requirements expand across diverse domains.
- Cambra's Vision: Cambra rejects the conventional wisdom that fragmentation is an inevitable cost of using the 'right tool for the job.' They aim to build a novel 'general-purpose, domain-aligned, sealed model' for internet software, promising dramatic acceleration in development, system-wide verification, automatic optimization, and streamlined operations.
- AI and Models: The author addresses the popular narrative that AI will negate the need for better abstractions. They argue that AI's effectiveness is enhanced, not replaced, by precise, domain-aligned models. AI agents perform best in well-structured environments with clear rules, confirming the ongoing importance of good models.
Cambra believes that recent advances in programming language theory and database systems have opened a viable path to achieve this ambitious goal, promising to make developing internet software feel like working on a single, coherent system rather than a fragmented mess.