AI is not removing testing. It is changing who does it, when it happens, and what QA leadership is for.
This is Part 1 of The Director of QA Dilemma, a five-part series about one larger transition: the Director of QA must move from defending testing activity to engineering the confidence that lets the company ship AI-generated software.
The Director of QA has an uncomfortable job right now.
The CEO sees AI and asks why testing still takes so long. Engineering sees coding agents generating tests and assumes validation will soon happen automatically. The QA team hears both conversations and wonders which jobs will remain.
All three reactions are understandable. All three are dangerously incomplete.
The Director can defend the existing organization, its automation portfolio, its release process, and its headcount. That may protect the team for a while, but it can also turn QA into the group preserving yesterday’s work while the rest of engineering changes around it.
Or the Director can accept the simplest AI story: coding agents will write the code, write the tests, run the tests, and decide when the work is finished. That sounds efficient, but it asks the system that created the defects to provide nearly all the evidence that those defects do not exist.
Neither answer is good enough.
The Director of QA must stop owning testing and start owning confidence.
The Old Bargain Is Breaking
Traditional QA organizations were built around a practical division of labor. Developers created software. Testers challenged it. Automation engineers encoded repeatable checks. QA leaders staffed the work, managed environments, tracked defects, reported pass rates, and negotiated release risk.
That model was never as clean as the organization chart suggested, but everyone had a recognizable role.
AI breaks the bargain because implementation and validation now happen in the same generative environment. A coding agent can inspect requirements, modify code, launch the application, write tests, exercise APIs, click through the interface, review logs, repair failures, and repeat the loop before a traditional QA handoff would begin.
Testing does not disappear. It expands dramatically. Its ownership becomes less obvious.
If the Director defines the team’s value as executing tests, maintaining piles of scripts, filing bugs, or protecting a final release gate, the function becomes easier to bypass. AI can perform much of that work faster and cheaper.
The value is no longer the volume of testing activity. It is deciding whether the evidence means anything.
I Keep Hearing the Same Story
I have my ear to the ground on this transition. QA directors reach out to me privately to compare notes, ask what other companies are doing, and talk through what AI means for their teams and careers.
This is not a formal survey, and the people who call me are naturally more likely to be wrestling with the problem. Still, the pattern is remarkably consistent: they feel trapped between the team and processes they have today and the AI-first organization management increasingly expects.
Most try to play both cards.
They reassure the team that the current work remains valuable and little has to change. At the same time, they tell management that QA is adopting AI and will soon move faster.
The team sees the experiments as a threat anyway. Management sees that delivery speed, cost, and turnaround time have not changed enough. Instead of satisfying both sides, the Director disappoints both.
I have heard about QA teams going around their Director to upper management, engineering leadership, and even HR because they dislike the proposed changes or fear what those changes mean for their jobs.
I have also seen engineering and product teams quietly work around QA. They begin building validation loops with Claude, Codex, and other coding agents. They do not always remove the QA leader immediately. First, they create an alternative. Then they prove enough of it works. Only then do they return to decide what, if anything, they still need from the existing organization.
The Director is rotated out.
Two Bad Choices
The first bad choice is denial.
A Director can insist that AI-generated tests are unreliable, coding agents make mistakes, and human testing will always be necessary. Every statement may be true, but the conclusion does not follow. Imperfect AI can still automate a large percentage of the work, operate continuously, and generate more variations than a human team can review.
Defending every current task because AI cannot perform it perfectly is not a strategy. It is an argument for being replaced slightly later.
The second bad choice is surrender.
A Director can reduce the team, buy an AI testing product, ask developers to own quality, and assume the coding-agent harness will eventually handle the rest. This confuses test generation with confidence.
An agent can write a thousand tests and still miss the wrong risk. It can create assertions that repeat its own implementation assumptions. It can optimize visible correctness while missing privacy, accessibility, recovery, cost, latency, policy, or user trust. It can even make a dashboard greener by weakening the checks.
The question is not whether AI can test. It can.
The question is who decides what deserves to be tested, what evidence is independent enough to trust, which failures matter, and when the remaining uncertainty is acceptable to the business.
That is the Director’s opening.
Part 2, QA Is the Hard Thing Now, explains why this disruption should make an exceptional QA leader more important, not less, and what a wartime QA director should do next.
Learn more in the Testing AI knowledge guide.
Get Out of the Trap in 30 Days
At IcebergQA, we help QA and engineering leaders move before their organizations work around them. Our approach draws on experience testing at scale at Microsoft and Google, building AI testing startups, and using AI to test both conventional and AI-powered products.
In a focused 30-day AI quality transformation sprint, we can map how your company is already using AI, identify where the current QA organization is being bypassed, align leadership around a new confidence mandate, and launch a practical AI-first quality workflow with your team. The goal is not a strategy deck. It is visible evidence that QA can help the company move faster.
—Jason Arbon, IcebergQA