The Apex Framework: Measuring AI Impact in Engineering
Explore the Apex framework, a new operating model for engineering productivity in the AI era. Learn how to move beyond simple tool adoption to measuring real value, predictability, and efficiency in the SDLC. Shift from 'faster coding' as an illusion to data-driven delivery outcomes.
Navigating the AI Productivity Paradox
Engineering leaders today face a strange contradiction: AI coding tools like Copilot, Cursor, and Claude are ubiquitous, and developers report feeling "faster." However, this speed often fails to translate into actual business value or faster delivery. This is the "faster coding illusion," where upstream acceleration (writing code) is lost to downstream chaos (code reviews, testing, and deployment bottlenecks).
Introducing the Apex Framework
To solve this, Linear B has introduced Apex, a balanced operating model designed specifically for the AI era. Unlike research-heavy frameworks like DORA, Apex is based on pragmatic, real-world customer usage. It treats AI as a primary operator in the Software Development Life Cycle (SDLC) and balances four critical pillars:
1. AI Leverage (A)
Rather than tracking simple tool usage, Apex focuses on AI-assisted Pull Requests (PRs). This identifies where AI is actually contributing to the unit of work and allows leaders to differentiate between high-impact adoption and "AI slop" (volume without quality).
2. Predictability (P)
AI can introduce significant variability. Apex emphasizes planning and capacity accuracy to ensure that increased volume doesn't degrade the team's ability to meet commitments and deliver reliable value to the business.
3. Efficiency (E)
Paying homage to DORA metrics, Apex tracks cycle time and Change Failure Rate (CFR). The goal is to ensure that the speed gained during the coding phase isn't negated by a spike in production incidents or a bloated review process.
4. Developer Experience (X)
DevX acts as a guardrail. If productivity metrics improve but developer satisfaction drops, the gains are likely unsustainable or illusory.
Operationalizing AI in the Critical Path
Implementing Apex is not a one-size-fits-all approach. Teams can start where they are: those struggling with basics should focus on Predictability and Efficiency first, while mature teams can lead with AI Leverage. By establishing a repeatable cadence—weekly for AI tracking, per-sprint for predictability, and quarterly for DevX—organizations can move from experimentation to operationalizing AI as a first-class production contributor.
Key insights
-
Upstream acceleration is often lost to downstream chaos. While AI tools speed up the coding phase, they frequently create bottlenecks in code reviews and deployment, meaning the overall delivery speed doesn't improve.
Impact: Organizations must optimize the entire pipeline, not just the coding phase, to realize the actual ROI of AI tools.
-
Simple tool adoption metrics are insufficient for measuring AI impact. True value is found by tying AI activity to the Pull Request (PR) level, specifically measuring the percentage of AI-assisted PRs.
Impact: Allows leadership to identify 'power users' and replicate successful AI behaviors across the organization while filtering out low-quality 'AI slop'.
-
AI acts as an amplifier of existing systemic issues. If an organization has poor predictability or efficiency before AI, AI will likely exacerbate those problems rather than solve them.
Impact: Forces a shift toward stabilizing core delivery metrics (Predictability and Efficiency) before attempting to scale AI adoption.
-
Developer Experience (DevX) serves as a critical guardrail for productivity gains. Improvements in cycle time and throughput without a corresponding level of developer satisfaction suggest unsustainable practices.
Impact: Prevents burnout and ensures that AI integration is human-centered and sustainable in the long term.
-
Engineering is a business of repeatable cadences. Measuring AI impact requires a tiered approach to reporting, from weekly tactical updates to quarterly executive reviews.
Impact: Transforms AI from a side experiment into a managed, predictable business process.
Action items
-
Benchmark current engineering performance using the Apex pillars (AI Leverage, Predictability, Efficiency, DevX) to identify the primary bottleneck.
Impact: Provides a data-driven starting point for optimization rather than guessing where AI is most needed.
-
Shift measurement from tool usage dashboards to tracking AI-assisted Pull Requests and their corresponding Change Failure Rates.
Impact: Ensures that AI-generated code is adding value and maintaining quality rather than just increasing the volume of code in the system.
-
Implement a tiered reporting cadence: weekly for AI adoption/impact, per-sprint for predictability, and quarterly for developer satisfaction.
Impact: Creates a repeatable rhythm of operation that aligns engineering output with business expectations.
-
Identify high-performing teams with high AI adoption and low change failure rates to document and replicate their specific AI workflows.
Impact: Accelerates the organizational maturity of AI adoption by scaling proven, successful behaviors.
Quotes
“Upstream acceleration is lost to downstream chaos.”
“Coding faster is an illusion.”
“If you're just generating a bunch of slop or junk, whatever, with AI, you would see okay, my planning accuracy in my sprints are decreasing.”