Open-Source 'Lattice' Framework Targets AI Coding Chaos with Built-in Engineering Disciplines

Rahul Garg today released Lattice, an open-source framework designed to fix the fundamental flaws in AI-assisted programming. The tool directly addresses how current AI coding assistants skip design reviews and forget constraints mid-conversation.

Lattice introduces three tiers of composable skills—atoms, molecules, and refiners—that enforce battle-tested engineering practices like Clean Architecture and Domain-Driven Design. It also creates a persistent context folder (.lattice/) that accumulates project standards, decisions, and review insights over time.

“After a few feature cycles, atoms aren't applying generic rules—they're applying your rules, informed by your history,” Garg said. The system grows smarter with use, making it a living repository of engineering discipline.

The framework can be installed as a Claude Code plugin or used standalone with any AI tool.

Background: The AI Coding Assistant Problem

Current AI coding assistants jump straight to code generation, silently make design decisions, and forget constraints partway through a session. Output often goes unreviewed against real engineering standards, leading to tech debt and inconsistent architectures.

Garg previously published a series of posts on reducing friction in AI-assisted programming. Lattice operationalizes those patterns into a concrete, reusable framework.

SPDD Article Draws Massive Traffic, Gets Q&A Expansion

Meanwhile, an earlier Structured-Prompt-Driven Development (SPDD) article by Garg’s colleagues Wei Zhang and Jessie Jie Xia has generated enormous interest. The piece has been updated with a new Q&A section answering a dozen of the most common questions.

“The response has been overwhelming, and we wanted to clarify key concepts,” said Zhang. The Q&A covers practical implementation details and common pitfalls.

Jessica Kerr Builds Tool to Harness Double Feedback Loops

Software developer Jessica Kerr, known as Jessitron, has created a tool that works with conversation logs to reveal a double feedback loop in AI-assisted development. She observes that developers operate within two loops simultaneously.

“One is the development loop, with Claude doing what I ask and then me checking whether that is indeed what I want,” Kerr wrote. “Then there's a meta-level feedback loop, the ‘is this working?’ check when I feel resistance.”

Kerr emphasizes that frustration, tedium, and annoyance are signals to change not just the software being built, but the tools used to build it. “With AI making software change superfast, changing our program to make debugging easier pays off immediately.”

What This Means

These developments signal a shift toward discipline-first AI coding. Instead of treating AI as a black box that writes code, developers are building frameworks (Lattice) and methodologies (SPDD) that inject engineering rigor into the AI interaction itself.

Kerr’s double-loop insight highlights a deeper trend: AI is enabling developers to rediscover the joy of molding their own development environment—a capability largely lost with modern, polished IDEs. Garg’s Lattice framework productizes that idea by making engineering standards part of the AI’s “memory” rather than an afterthought.

For teams adopting AI-assisted programming, these tools offer a path to consistent, reviewable, and increasingly intelligent code generation—if they invest the initial effort to configure the context layer and skills.