7CTO Deep DIve

7CTO Member Led Deep Dive

How Your AI Engineering Team Should Address Quality​

Addressing AI Engineering Quality - June 12, 2026

This presentation by Jay, Founder and CEO of JDAQA, addresses the widening quality gap in engineering teams resulting from the rapid adoption of AI development tools. As boards push for higher engineering velocity, many organizations are silently accumulating massive “quality debt”. Jay outlines how teams can visualize this debt, transition to a quality-first development methodology, and leverage QA leaders as system architects rather than reactive firefighters.

The Core Problem: Invisible Quality Debt
With the introduction of AI tooling like Cursor and Claude, engineering output and velocity have surged—often by up to 40%. However, this speed comes with a hidden cost:
 
  • Spiking Defects: Total defects per sprint and bug severity scores are climbing rapidly.
  • Invisible Accumulation: Because companies primarily measure velocity or tickets closed, quality debt accumulates invisibly in every single commit.
  • Increasingly Severe Bugs: Teams are not just shipping more bugs; they are shipping severely worse bugs, such as silent crashes, generic timeout errors, and malformed JSON.
 
Actionable Strategy: Quantifying Quality Debt for the Board
 
Jay provides a direct playbook for CTOs to reframe conversations with CEOs and the board, shifting from “asking permission to slow down” to proactively flagging a crisis before it hits in 6 to 12 months:
 
  • Extract the Data: Pull the last six months of defect data from Jira or Linear into a CSV file, capturing sprint severity and whether the bug was found in QA or production.
  • Analyze with AI: Drop the file into Claude to calculate two metrics: a specific quality debt dollar amount and the overall trajectory trend.
  • Present to the Board: Use these metrics to demonstrate that while AI has increased engineering output by 40%, it will cost a projected amount over the next quarter if left unaddressed.
 
The Blueprint: Quality-First Development
 
To close the quality gap, teams must stop attempting to bolt quality onto the end of the development cycle. Jay introduces Quality-Led Development (QLD), which requires establishing constraints before developers prompt AI.
1. Developer-Led Quality (The Feature Level)
Instead of accepting raw AI code blindly, developers should attach two brief files to their AI prompts to create immediate constraints:
 
  • A Spec File: A brief, five-minute document mapping out inputs, outputs, requirements, and hard constraints.
  • A Test File: A baseline of general tests or historical edge cases that have broken in the past.
  • The Result: Forcing the AI model to build against these constraints ensures explicit error handling and covers critical failure scenarios before code is even generated.
 
2. QA-Led Quality Infrastructure (The Repo/Org Level)
Because relying entirely on developer discipline is unrealistic, a QA lead or architect must build an overarching quality infrastructure across three layers:
 
  • Claude.md File: Automatically loads into every AI session to define the stack, what “done” looks like, and strict testing conventions.
  • Claude Code Hooks: Pre-commit hooks that intercept actions between Claude and the codebase, automatically blocking commits if linters or type checkers fail.
  • Testing.md File: A single file written per repository by the QA lead that defines what a “good test” looks like, preventing junior developers from dictating standards on their own.
 
Key Q&A Insights: The Evolving Role of QA
During the collaborative discussion, several engineering leaders shared their experiences regarding how the AI Software Development Life Cycle (SDLC) is redefining team topographies:
 
  • The Human Strategic Layer: Human validation remains essential to prevent “garbage in, garbage out” loops, as AI cannot reliably dictate test strategy.
  • Moving Upstream: Participants agreed that QA must move further upstream into the product design, definition, and backlog grooming phases to catch bad design before engineering begins.
  • Managing AI “Slop”: Tech leaders noted that QA teams are currently overburdened with “slop” and recurring bugs reintroduced by AI tools multiple sprints later.
  • QA as AI Factory Managers: Rather than executing manual tasks, QA professionals are transitioning into managers of AI agents. They oversee automated agents that review user stories, check specs, and generate consistent test outputs.
Let’s Talk

Schedule your Discovery Call

Schedule your Discovery Call