Many Small Loops: Feedback That Hardens Agent Output
Agentic QA holds up through many small feedback loops in the process instead of one big test gate at the end. Three ideas for where a team can start.
AI & Software Quality
Artificial intelligence changes software quality in two directions: teams need to evaluate AI-based systems with probabilistic outcomes, while generative AI can also support analysis, test design, and automation. This topic separates these perspectives and connects technical opportunities with realistic quality risks.
The articles and podcast episodes cover areas such as testing large language models, data and model quality, suitable evaluation approaches, and the responsible use of AI in testing. They explore where AI provides productive support, where human judgement remains essential, and how experiments can become traceable quality practices.
Foundations
Articles
AI in Testing: The Focus Shifts from Writing to Testing
AI in testing writes tests, explores apps and generates test data. It also becomes a test object itself, from chatbots to RAG search.
Faster Horses: Why AI Rewards Teams That Talk to Each Other
Using AI for our old tasks just gives us faster horses. As the next abstraction layer, AI rewards precise intent and teams that talk to each other.
Between Pride and Throwaway Software: Ownership Takes Work
Simply accepting AI output creates no bond with the result. Ownership of code, architecture and tests grows from working through them yourself.
The Great Compression: When AI Leaves No Time to Think
Kamchatka 2008 with books instead of a ChatGPT itinerary: what the double compression of time and work does to us, and why deliberate brakes help.
AI: Are We Solving the Problems That Matter?
AI generates movies and apps while teams drown in legacy code. Why we should use AI for the problems that actually hurt, not for demos.
AI Hype Cycle: How to Stay Relaxed but Mindful
Caught in the AI hype cycle? Total refusal rarely works. Small steps and Gartner's model help you stay calm and alert amid the constant AI noise.
AI and Learning: What We Lose When We Only Take the Output
AI delivers results fast, and the learning process passes us by. Testers grow through questions, debugging and system knowledge, not generated tests.
Testers in the AI Era: Interpreters Between Code and People
The future of testing lies not in better tools but in humanity. Why testers are the interpreters between AI, code, and the people behind it.
AI Programming Languages: Must Code Stay Human-Readable?
Should AI programming languages stay readable for humans? If AI writes the code, compilers and syntax may change. What that leaves for developers.
Stoicism: Calmer Decision Making in the Age of AI
Stoicism helps with decision making in the AI era: machines do more of the work, we decide more often. Control, virtue and acceptance bring clarity.
Focus on People: Investing in Humans Alongside AI Tools
Focus on people: when AI writes much of the code, decisions on quality, ethics and vision land with humans. Companies should invest in them as in tools.
Test Data: First a Radical Cure, Then AI
Test data problems won't vanish with AI. First ask which data you really need, delete the rest and build a data architecture that works across systems.
Testing AI and Testing with AI: Rethinking Quality
AI in software testing helps with test cases, test data and error analysis. Testing AI systems means rethinking quality criteria such as data quality.
AI for Testers: Why Good Testers Won't Be Replaced
AI for testers helps with analysis, test cases, bug hunting and test data. Weak testers may lose out, good ones build AI into the way they work.
SDETs and Coaching Roles: Testing Needs Broader Profiles
SDETs, coaching roles and quality engineers in place of test managers: in a truly agile setup, quality becomes a shared responsibility of the team.
T-Shaped Testers: Software Testing of the Future
T-shaped testers who span requirements, architecture, development and DevOps: that is software testing of the future. ML systems need new approaches.
AI Evaluation Metrics: Why Pass or Fail Falls Short
AI evaluation metrics like precision, recall, and the F1 score judge a model across many test cases. Evaluating AI also covers its training data.
Podcast Episodes
How to Build Rules and Skills So That Agents Can Remember
AI coding agents need fewer manual reviews once you work on the system itself: rules, hooks and retro skills make their corrections stick for good.
Agentic Systems in Testing: Why Humans Stay in the Loop
Agentic systems are not just agents working alone. Between 8 and 12 agents can cover testing end to end, but humans still judge what AI cannot.
AI-Generated Code: Quality Assurance Must Keep Pace
More AI-generated code means more defects, and only about 30% of AI review comments help. Static analysis, test selection and reviews must scale too.
AI in Software Testing Raises the Value of Expert Testers
AI in software testing leaves a faulty fraction: hallucinated requirements, invented test cases. Business knowledge keeps expert testers in demand.
LLM Testing: A Second Model Acts as the Judge
LLM testing can be automated if you accept non-determinism: rubrics and acceptance ranges replace yes/no asserts, and a calibrated LLM judges outputs.
API Testing: Agent-Optimized Design Cuts Token Costs
API testing depends on the protocol: REST, GraphQL, webhooks or MCP. It runs more stably than browser testing, and agent-friendly APIs save tokens.
LLMs in Software Testing: Start with Small, Bounded Tasks
LLMs in software testing pay off where language is the raw material: keyword docs, rule-based reviews, defect messages. Reasoning models review better.
Vibecoding and the Lack of System Understanding in Practice
Vibe coding risks grow where system understanding is missing. The future belongs to Pi-shaped developers who combine breadth with real depth.
Automatically Reviewing Documentation with AI Personas
Documentation testing with AI personas: a simulated junior developer burned 1.8 million tokens where a senior needed 300,000, exposing gaps in the docs.
Learning and Unlearning with AI: Skills That Outlast Tools
Learning and unlearning with AI: students who leaned on ChatGPT retained less. Which skills AI cannot replace and which work habits to let go of.
What AI Really Does to Trust and Team Dynamics
Heavy AI use erodes trust in teams: colleagues seem infallible, and friction and social learning fade. Strong teams onboard AI like a new team member.
Tester Skills: Communication Decides Whether Bugs Matter
Software tester skills start with communication: a bug nobody understands stays unfixed. Critical thinking counts when AI confirms wrong assumptions.
AI in Regulated Industries: Strategy First in a Medical Lab
AI in regulated industries starts with strategy: a medical lab mapped stakeholders, narrowed the scope and put compliance on one page with a bow tie.
Generative AI Ethics: The Case Against Hyperscale Models
Generative AI cannot be used ethically as long as training data is used without consent and billionaires control the models.
Chatbot Testing: Why Bugs Hide in Prompts and Retrieval
Chatbot testing has to look past the final answer into chunks, embeddings, prompts and fallback logic. How traceability makes these hidden bugs fixable.
AI-Assisted Testing: Testers Keep Control of the Process
AI-assisted testing turns a month of test case and test data work into about a week, and plain-language automation widens who can build tests.
Agentic Engineering: AI Accelerates Existing Principles
Agentic engineering is less about prompts than about the agent environment: your prompt is about 20 of 20,000 tokens. Cleanup agents keep quality up.
Security Testing for AI: Checking Outputs Instead of Just Inputs
AI security testing covers the whole life cycle, from training data to live queries. Why output filters often beat input filters, and what OWASP offers.
Trust in AI Agents Lives in the System Around the Model
AI agents earn trust through checks, guardrails and clear requirements. Testing them tracks trends and mean time between failures, not single runs.
QFD: The Cause-and-Effect Analysis LLMs Can't Deliver
Quality Function Deployment (QFD) links customer benefits to functions and tests, the cause-and-effect analysis LLMs can't do. The result: fewer tests.
AI-Generated Test Cases: Compliant Without Tool Validation
AI-generated test cases for medical devices: a RAG system retrieves in-house documents, a human reviews the results, so no tool validation is needed.
Next-Gen Software Engineering: A Test Model for Every Model
Next-gen software engineering pairs every model with a test model. AI code looks cheap, but someone still has to version, document and test it.
Critical Thinking in AI-Assisted Testing: Check the Source
Critical thinking in AI-assisted testing means asking where an output comes from. If AI writes code and tests, a tester still has to judge the result.
AI Test Assistant: Test Cases Compliant with ISO 29119
An AI test assistant generates ISO 29119 test cases for business teams without testing know-how. Munich's Sherlock exports them to TestLink or Xray.
Is AI in Software Testing Making the Tester's Job Obsolete?
AI in software testing is not making testers redundant. The job shifts toward test strategy, critical questioning and early work in refinements.
Testing in 2025: Realism Wins, Accessibility Grows
Testing in 2025 got faster with AI for test cases, scripts and test data, and accessibility became a legal duty. Next up: orchestrating AI agents.
Software Quality: AI Only Accelerates Well-Defined Processes
Software quality depends on a solid test process. AI doesn't fix weak processes, it exposes them faster. Testers also need gut feeling.
Testing Legacy Apps with AI: Visual Selectors Instead of IDs
Testing legacy apps without element IDs works with visual AI selectors. Deutsche Bahn cut mobile checkout regression from two weeks to three hours.
AI Bias: People Adopt Biases from Algorithms
AI bias spreads because people adopt it from algorithms even faster than from humans and rarely notice. Comparative data keeps AI decisions in check.
AI in Requirements Engineering Requires Precise Prompts
AI requirements engineering only delivers testable requirements with precise prompts: scope, context, requirement level and templates, refined in steps.
Natural Intelligence Tested: Humans Evaluate AI
Natural intelligence finds bugs no AI would look for, through curiosity, intuition and context. AI unit tests often hit just one equivalence partition.
System Knowledge Catches What AI Misses in Grown Code
For software testing, AI pays off in test data and document analysis. In grown systems, testers' system and domain knowledge still catch side effects.
Software Testing Survey: Basic Test Techniques Are Fading
The 2024 software testing survey of more than 800 respondents: a third use AI for coding and QA, yet only about 40 percent apply basic test techniques.
AI Code Remediation: Two Thirds of the Fixes Work
AI code remediation fixes about two thirds of static analysis findings well. The rest is faulty, and models almost never admit they don't know.
AI Code Generation: More Legacy Code, Produced Faster
AI code generation pushed up code churn and duplicates on GitHub by early 2024. Testing, precise questions and fundamentals set developers apart.
AI Agents in Software Testing: Treat Them like New Hires
AI agents in software testing need what a new hire gets: a selection process, a probation period, and ongoing performance checks, not fixed outputs.
Legacy Modernization: Why Missing Experts Are the Bottleneck
Legacy modernization usually stalls on missing subject matter experts. How RAG makes legacy code queryable and why working in slices lowers the risk.
AI Audit: How the AI Assessment Matrix Structures Testing
An AI audit needs a yardstick. The TÜV AI Lab's AI Assessment Matrix maps test dimensions against the lifecycle, so assessors pick the relevant fields.
Requirements Analysis with DEFOSPAM: Finding Gaps Early
Requirements analysis with DEFOSPAM: seven question areas expose unclear definitions, ambiguity and missing parts. AI adds ideas, people judge them.
AI Test Description: Test Cases Structured Like Code
An AI test description makes AI systems testable: five capabilities and quality criteria like correctness and bias turn standards into concrete tests.
AI-Supported Test Design: Method Knowledge Still Matters
AI-supported test design delivers equivalence partitions and boundary values in minutes, about 80 percent correct. Catching errors takes method skills.
Software Testing in 2024: AI and Legislation as Drivers
Software testing trends 2024: AI became standard in test tools, while the Accessibility Act, Cyber Resilience Act and EU AI Act pulled testing earlier.
Testing Trends: Testing AI and Testing with AI
Trends in software testing: AI is used most in programming and testing, far less in requirements engineering. Scrum dominates, even in aviation.
Generative AI in Testing: AI-Written Unit Tests Cement Bugs
Generative AI in testing speeds up automation with boilerplate code, test case reviews and edge cases. Unit tests generated from code cement its bugs.
Accessibility Testing: AI Checks Linguistic WCAG Criteria
Accessibility testing runs on a jumble of Word, Excel and browser tabs. AI checks linguistic WCAG criteria, and clustering picks representative pages.
Testing AI Systems: Determinism Decides the Test Strategy
Testing AI systems often boils down to one question: does the system behave deterministically? The answer decides whether classic methods suffice.
Software Testing Trends: Quality Becomes a Team Skill
Software testing trends like AI, test data and accessibility: what testers should watch, and why a new tool rarely fixes the real problem.
Future of Software Development: The Basics Will Stay
Future of software development: AI mainly helps developers create legacy code faster. What that means for quality and jobs over the next ten years.
A4Q Practical Tester: AI Builds the Exam, a Human Certifies
The A4Q Practical Tester certifies hands-on skills such as test plans and defect reports. AI generates the exam questions, a human makes the final call.
ATDD for LLMs: Failed Dialogs as Tests and Training Data
With ATDD for LLMs, each new requirement becomes a dialog-based test. After fine-tuning, the suite shows whether the model learned without regressions.
Happy Birthday: Podcast Software Testing Turns One
One year, 67 episodes, over 22,000 downloads: What became of a spontaneous video idea and where the journey will take us next.
ChatGPT in Software Testing: Strong in Exploratory Testing
Using ChatGPT for test cases, test data and exploratory ideas: What really works, where caution is advised and why prompting matters.
AI Quality Assurance: When There Is No Clear Test Oracle
AI quality assurance lacks a clear test oracle, so quality is measured statistically. Metamorphic and pairwise testing help, reproducibility stays hard.
Fair AI: Assurance Case Makes Fairness Testable
Fair AI sounds good, but what does it mean in measurable terms? Fairness measures contradict each other, so an assurance case breaks the claim down.
Speech Recognition Testing: Every Fix Means Retraining
Speech recognition testing means testing a black box: every fix costs days of retraining. Golden test sets and generated variations find errors early.
Prompt Engineering for Test Cases: Review Remains Mandatory
Prompt engineering for test cases gets better results with few-shot examples and embedded specs. Still, two of eleven generated tests remained wrong.
Software Testers and AI: Routine Goes, Judgment Stays
Software testers and AI: routine checks move to AI, while judging requirements coverage stays a human task. A carpenter explains his path into testing.
AI Test Automation: Machine Learning Finds Root Causes
AI test automation turns 200 failed tests into seven real causes. Machine learning groups failures by root cause and frees testers to write new tests.