From Engineering Toil to AI-Native Ambition
Within Meta’s Reality Labs, the Horizon Experiences group has been running a focused experiment in AI-native engineering, treating AI not as a product feature but as part of the development fabric itself. The starting point was a familiar problem: engineers losing large portions of their day to toil—updating tests, fixing minor breakages, and reviewing mundane code changes. Instead of accepting this as inevitable, a small team asked how AI could systematically remove this drag so people could spend more time exploring ideas and solving hard problems. Their vision is to move engineers from being primarily builders to becoming explorers and innovators, with AI handling a growing share of repetitive implementation work. The initiative is intentionally framed as an engineering excellence effort, tying AI tools directly to measurable improvements in implementation quality, test health, code complexity, and documentation, rather than treating AI adoption as a side project or curiosity.
Assess and Grow: A Software Development Maturity Model for AI
To make that vision operational, Reality Labs created the Assess and Grow framework, effectively a software development maturity model for AI-native engineering. Instead of pushing one-off tool experiments, teams complete structured assessments that capture where AI is already embedded in their workflow and where manual effort still dominates. The model looks across areas such as test coverage, code quality practices, onboarding documentation, and everyday modernization tasks, then maps how AI could progressively support or automate each of these. Multiple teams have now run these assessments, revealing repeatable patterns rather than isolated wins. This maturity model gives leaders a way to benchmark AI-driven development workflow adoption, track progress over time, and prioritize investments in tools and training. By naming stages and practices explicitly, Assess and Grow turns an abstract idea—“use more AI”—into a concrete progression that engineering organizations can manage and measure.
Building a Community to Drive Engineering Team Transformation
The framework did not emerge in isolation; it grew alongside a grassroots movement called AI4P (AI for productivity). Over roughly seven months, what began as a small, ring-fenced experiment inside one product area expanded into a community of more than 400 participants. The team started deliberately small to create a safe environment where engineers could admit knowledge gaps, share missteps, and iterate quickly on practices without heavy scrutiny. Early work focused on internal tools and clear engineering excellence goals: improving test health, lifting code quality, and streamlining documentation. Informal brown-bag sessions surfaced emerging patterns of success, which were then codified into shared documentation and playbooks. This bottom-up approach meant that when the Assess and Grow framework arrived, it landed in an ecosystem already primed for change. The model became both a mirror for current practice and a shared roadmap for engineering team transformation across Reality Labs and beyond.
AI-Native Workflows Are Rewriting Engineering Norms
Meta’s experience fits into a broader shift toward AI-native engineering, where AI systems are embedded throughout the lifecycle rather than tacked on at the edges. Across the industry, AI is now assisting with code generation and refactoring, automated debugging, AI-driven QA, infrastructure recommendations, documentation generation, and even workflow orchestration. This is compressing development cycles dramatically, allowing smaller teams to ship software at speeds that recently seemed unrealistic. The Assess and Grow framework highlights what this means organizationally: success is less about experimenting with a single coding assistant and more about redesigning the entire engineering system to assume AI participation by default. As more companies adopt similar software development maturity models, the gap will widen between organizations that systematically cultivate AI-driven development workflows and those still debating whether AI belongs in their day-to-day engineering practice.