# State of AI in Quality Engineering 2026

Publisher: TestMu AI. Foreword: Asad Khan, Co-founder and CEO. Published 2026-11-02. Last updated 2026-11-02.
Canonical: https://www.testmuai.com/research/state-of-ai-in-quality-engineering/
Cite as: TestMu AI. (2026). State of AI in Quality Engineering 2026. https://www.testmuai.com/research/state-of-ai-in-quality-engineering/
When you quote a figure, keep its base, for example "of teams answering the question" or "of agent builders". Chart credit: Source: TestMu AI, State of AI in Quality Engineering 2026.

## Summary

The State of AI in Quality Engineering 2026 is TestMu AI's survey of testers, developers and QA leaders, fielded from late July to early October 2026. The short answer: AI now writes much of the code and many of the tests, and most teams still verify that work with a person after the fact, or with another model.

## Key statistics

1. 81% of teams have merged AI-written code or tests they didn't fully understand, according to TestMu AI's State of AI in Quality Engineering 2026 survey; 19% do it regularly.
2. 44% of teams report more escaped defects or production incidents as AI-generated code grew (TestMu AI, State of AI in Quality Engineering 2026).
3. 40% of teams say AI writes more than half of their production code, and only 3% say none of it is AI-generated (TestMu AI, State of AI in Quality Engineering 2026).
4. 41% of teams run any automated check on AI-written code before merge; most rely on a person, or on a QA hand-off after the code is written (TestMu AI, State of AI in Quality Engineering 2026).
5. 29% of teams that use LLM-as-a-judge regularly merge AI output they don't fully understand, twice the rate of other teams (15%), according to TestMu AI's State of AI in Quality Engineering 2026.
6. 65% of the time, at the median, AI-generated tests are "almost right, but not quite", and median trust in AI-generated test artifacts is 54 out of 100 (TestMu AI, State of AI in Quality Engineering 2026).
7. 37% of teams that build AI agents have no systematic way to evaluate them, and only 28% test agents against users who try to break them (TestMu AI, State of AI in Quality Engineering 2026).
8. 45% of teams building voice or phone agents do not test with background noise or a poor line, and 39% do not test accents, dialects or other languages (TestMu AI, State of AI in Quality Engineering 2026).
9. 50% of teams spend most of their AI testing budget on generating output (coding-agent seats and model APIs), against 17% on verifying it (TestMu AI, State of AI in Quality Engineering 2026).
10. 13% of teams have an incident-response plan for AI mistakes (TestMu AI, State of AI in Quality Engineering 2026).

## Figures

Every percentage is a share of the people who answered that question; totals for multi-select questions exceed 100%. QA and developer views draw on smaller groups. Read them as directional.

### What agent builders do before an agent ships

| Answer | Share |
| --- | --- |
| Use human review to evaluate the agent | 63% |
| Test against users who try to trick or break it | 28% |
| Re-run evaluations after a model or prompt change | 27% |
| Call it ready on a demo, or nothing formal | 24% |
| Make evaluations a release gate in CI | 15% |

### Changes between the 2023 and 2026 surveys

Base: Future of QA 2023 against the State of AI in Quality Engineering 2026 survey.

| Group | Survey | Share | Measure | Note |
| --- | --- | --- | --- | --- |
| AI adoption | 2023 | 22% | have no plans to use AI in testing |   |
| AI adoption | 2026 | 3% | have no plans to use AI in testing |   |
| Barriers to AI in testing | 2023 | 60% | reliability is a barrier |   |
| Barriers to AI in testing | 2026 | 50% | output we can't trust is a barrier | reworded |
| Barriers to AI in testing | 2023 | 54% | skill gap is a barrier |   |
| Barriers to AI in testing | 2026 | 27% | skill gap is a barrier |   |
| Barriers to AI in testing | 2026 | 55% | data privacy, security or IP exposure | new option, now #1 |
| Where tests run | 2023 | 48% | local machines or in-house grids |   |
| Where tests run | 2026 | 35% | local machines or in-house grids |   |
| Where tests run | 2023 | 23% | cloud CI/CD runners |   |
| Where tests run | 2026 | 31% | cloud CI/CD runners |   |
| Release pace | 2023 | 20%+ | release daily or faster | published as 20%+ |
| Release pace | 2026 | 30% | release daily or faster |   |
| Frameworks | 2023 | 75% | use two or more frameworks |   |
| Frameworks | 2026 | 74% | use two or more frameworks | unchanged |
| Frameworks | 2023 | 24% | use Selenium, Cypress and Playwright together |   |
| Frameworks | 2026 | 10% | use Selenium, Cypress and Playwright together |   |
| Testing without people | 2023 | 4% | expected AI to do all testing without human oversight |   |
| Testing without people | 2026 | 11% | say most of their testing already runs autonomously | different question |

### The skill gap has halved as a barrier since 2023, and data privacy now leads

Question: What is holding back AI in your testing? (2026: pick up to three)

Base: 2023 report and 2026 survey. Data privacy was not an option in 2023; answer options and limits changed.

| Barrier | 2023 | 2026 |
| --- | --- | --- |
| Data privacy, security or IP (new in 2026) | not asked | 54.9% |
| Reliability or untrustworthy output | 60.3% | 49.7% |
| Cost or unclear return | 42.6% | 39.4% |
| Integration with existing tools | 35.3% | 38.3% |
| Skill gap | 54.4% | 26.9% |

### 40% of teams say AI writes more than half of their production code

Question: Roughly what % of your developers' production code is AI-generated?

Base: Teams answering the question. One answer each.

| Answer | All teams | QA practitioners | Developers |
| --- | --- | --- | --- |
| 0% | 3.1% | 0% | 5.2% |
| Less than 25% | 20.4% | 24.8% | 12.1% |
| 26-50% | 28.3% | 23% | 44.8% |
| 51-75% | 25.1% | 28.3% | 24.1% |
| More than 75% | 14.7% | 16.8% | 10.3% |
| Don't know | 8.4% | 7.1% | 3.4% |

### Only 41% of teams run any automated check on AI-written code before merge

Question: When your AI coding agents write product code, how is it verified before merge? (select all that apply)

Base: Teams answering the question. Several answers allowed.

| Answer | All teams | QA practitioners | Developers |
| --- | --- | --- | --- |
| A person reviews and manually tests it | 55.1% | 50% | 69% |
| QA team builds and runs tests after hand-off | 51.4% | 59.1% | 43.1% |
| Separate automated tests run in CI | 34.1% | 36.4% | 29.3% |
| Coding agent verifies its own work in the app | 14.1% | 8.2% | 20.7% |
| Only compile, type or unit checks | 12.4% | 7.3% | 22.4% |
| No check beyond the agent saying it works | 5.4% | 3.6% | 6.9% |

### 81% of teams have merged AI output they did not fully understand

Question: Has your team merged AI-generated code or tests without fully understanding them?

Base: Teams answering the question. One answer each.

| Group | Regularly | Occasionally | Rarely | Never |
| --- | --- | --- | --- | --- |
| All teams | 19.3% | 33.2% | 28.3% | 19.3% |
| QA practitioners | 21.6% | 27% | 27% | 24.3% |
| Developers | 19% | 44.8% | 29.3% | 6.9% |

### Trust and the "almost right" rate by how AI-generated tests are overseen

| How AI-generated tests are overseen | Median trust (0-100) | Median "almost right" rate |
| --- | --- | --- |
| All teams | 54 | 65% |
| A person reviews and approves every test | 50 | 64% |
| Teams spot-check a sample | 47 | 52% |
| Tests run automatically; people step in only on failures | 80 | 78% |
| An agent writes, runs and decides, with people only on exceptions | 65 | 81% |

### Teams that use LLM-as-judge against other teams

| Share of teams who | Use LLM-as-judge | Other teams |
| --- | --- | --- |
| Regularly merge AI output they don't fully understand | 29% | 15% |
| Have ever merged AI output they didn't fully understand | 93% | 75% |
| See escaped defects rising as AI code grew | 50% | 41% |
| Report a positive return from AI in testing | 90% | 66% |
| Median trust in AI-generated tests (0-100) | 66 | 50 |

### Human review is the only agent-testing practice most agent builders use

Question: How do you evaluate AI outputs? Which real-world conditions do you simulate? Which tools do you use? (select all that apply)

Base: Teams that build AI agents they need to test. The last row is from a free-text answer.

| Answer | Share |
| --- | --- |
| Evaluate agent output by human review | 63% |
| Use LLM-as-judge (one model scores another) | 41% |
| Test long multi-turn conversations | 35% |
| Test against users who try to break the agent | 28% |
| Re-run evals after model or prompt changes | 27% |
| Say they use a dedicated eval platform | 21% |
| Gate CI on eval suites | 15% |
| Built their own voice or chat test harness | 11% |
| Named a dedicated eval tool when asked which one | 6% |

### 24% of agent builders ship on a demo or nothing formal

Question: How do you decide an AI agent is ready to ship to production?

Base: Teams that build AI agents they need to test. One answer each.

| Answer | Share |
| --- | --- |
| Quality-metric scores (accuracy, tone, resolution) | 51.5% |
| A formal go or no-go verdict | 17.6% |
| A subjective demo or ad-hoc review | 18.4% |
| Nothing formal | 5.1% |
| Not applicable | 7.4% |

### Reviewing AI output now rivals flaky tests as the biggest drain on QA time

Question: Where is the most QA time wasted today?

Base: Teams answering the question. One answer each; 3% gave another answer, not shown.

| Answer | All teams | QA practitioners | Developers |
| --- | --- | --- | --- |
| Test maintenance | 19.3% | 19.6% | 20.7% |
| Flaky tests | 18.7% | 25% | 12.1% |
| Reviewing AI-generated output | 17.6% | 17% | 17.2% |
| Environment or test-data setup | 17.1% | 15.2% | 22.4% |
| Slow pipelines | 10.2% | 8% | 13.8% |
| Triaging false positives | 7.5% | 7.1% | 3.4% |
| Coordination and hand-offs | 7% | 6.2% | 8.6% |

### Generating AI output gets the main budget three times as often as verifying it

Question: Where does the majority of your AI testing budget get spent?

Base: Teams answering the question. One answer each; under 1% gave another answer, not shown.

| Answer | Share |
| --- | --- |
| Coding-agent seats and licences | 29.6% |
| Not sure | 21.8% |
| LLM or model API usage | 20.7% |
| Commercial AI testing platforms | 9.5% |
| In-house build | 6.1% |
| Eval and observability tooling | 3.9% |
| Cloud test execution and infrastructure | 3.4% |
| Training and enablement | 2.8% |
| Consulting and system integrators | 1.7% |

### 73% of teams report a positive return from AI in testing

Question: What is your ROI from AI use in software testing?

Base: Teams answering the question. One answer each.

| Group | Over 100% | 51-100% | 11-50% | 0-10% or a loss | Too early to tell |
| --- | --- | --- | --- | --- | --- |
| All teams | 8.5% | 26.1% | 38.8% | 9.5% | 17% |
| QA practitioners | 10.7% | 28.6% | 37.5% | 7.1% | 16.1% |
| Developers | 5.3% | 26.3% | 40.4% | 14% | 14% |

### Only 13% of teams have a plan for when AI gets something wrong

Question: How does your company keep its use of AI in testing responsible and accountable? (select all that apply)

Base: Teams answering the question. Several answers allowed.

| Answer | Share |
| --- | --- |
| Human review or sign-off on AI-driven decisions | 52.7% |
| A documented AI policy or principles | 45.1% |
| An audit trail of what the AI did | 37% |
| Real-time monitoring for issues or drift | 33.7% |
| Testing AI outputs for bias or fairness | 21.2% |
| No formal controls yet | 19.6% |
| An incident-response plan for AI mistakes | 13% |

### 79% say most of their testing already runs autonomously or will within two years

Question: How soon do you expect the majority of your testing to run autonomously?

Base: Teams answering the question. One answer each.

| Answer | Share |
| --- | --- |
| It already does | 11.2% |
| Within 1 year | 33.1% |
| 1-2 years | 34.8% |
| 3-5 years | 7.3% |
| More than 5 years | 2.8% |
| Not a goal | 10.7% |

### Only 10% of pipelines let an agent decide test outcomes today

Question: How autonomous is your testing pipeline end-to-end?

Base: Teams answering the question. One answer each.

| Answer | Share |
| --- | --- |
| Fully manual triggers | 22.7% |
| Scripted CI, no AI | 21.5% |
| AI-assisted within CI | 27.9% |
| Agent writes and runs, human signs off | 18% |
| Agent writes, runs and decides | 9.9% |

### Playwright has overtaken Selenium as the most used automation framework

Question: Which frameworks or tools do you use for test automation? (select all that apply)

Base: Teams answering the question. Several answers allowed; top 10 shown.

| Answer | Share |
| --- | --- |
| Playwright | 62.1% |
| Selenium | 52% |
| TestNG | 32.3% |
| PyTest | 31.3% |
| JUnit | 27.3% |
| Appium | 24.2% |
| Cucumber | 24.2% |
| Cypress | 18.2% |
| WebDriverIO | 13.6% |
| Jest | 11.6% |

### Almost half of teams have automated 40% or less of their test suite

Question: Approximately what percentage of your test suite is automated?

Base: Teams answering the question. One answer each.

| Answer | Share |
| --- | --- |
| 0-20% | 26.8% |
| 21-40% | 18.7% |
| 41-60% | 23.9% |
| 61-80% | 21.1% |
| 81-100% | 9.6% |

## Who took part

58% of respondents are SDETs, QA leads or testers, and 29% are developers and engineers. They come from the TestMu AI community rather than its customer base. The sample leans towards organisations based in India and towards SaaS companies, so the figures describe the teams surveyed rather than the whole industry.

| Profile | Answer | Share |
| --- | --- | --- |
| Role | QA practitioners (SDETs, QA leads, testers) | 58% |
| Role | Developers and engineers | 29% |
| Role | Leaders (executives, heads of engineering or QA, PMs) | 7% |
| Role | Other | 6% |
| Company size | 1-10 employees | 19% |
| Company size | 11-100 | 24% |
| Company size | 101-1,000 | 22% |
| Company size | 1,001-10,000 | 20% |
| Company size | 10,000+ | 15% |
| Where the organisation is based | India | 43% |
| Where the organisation is based | United States | 23% |
| Where the organisation is based | Germany | 4% |
| Where the organisation is based | Nigeria | 4% |
| Where the organisation is based | United Kingdom | 3% |
| Where the organisation is based | Other countries | 23% |
| AI in their own product | AI is core to the product | 26% |
| AI in their own product | Some features use AI in production | 46% |
| AI in their own product | Building or prototyping | 21% |
| AI in their own product | No AI features | 8% |
| Release cadence | Multiple times a day | 16% |
| Release cadence | Daily | 14% |
| Release cadence | Weekly | 36% |
| Release cadence | Monthly | 17% |
| Release cadence | Quarterly | 9% |
| Release cadence | Other | 8% |
| Industry | SaaS, software and IT services | 58% |
| Industry | Banking, financial services, insurance | 12% |
| Industry | Retail and e-commerce | 5% |
| Industry | Other industries | 24% |

## Method

Fielded online, from late July to early October 2026, through TestMu AI's community channels. Respondents are testers, test automation engineers, QA leaders, developers, engineering leaders and product managers from the TestMu AI community, at organisations around the world; they are community members, not a sample of TestMu AI customers. Each response is one person describing their own team; for readability, the report refers to respondents as teams. Shares are of people who answered each question; bases for subgroups are smaller. Only completed responses were analysed. Duplicate submissions were removed, keeping the most recent, as were test entries and responses completed too quickly to have been read. Prepared by TestMu AI's research and marketing team under the TestMu AI editorial policy (https://www.testmuai.com/editorial-policy-processes/). Press and research enquiries: press@testmuai.com.

## Sources

- TestMu AI, State of AI in Quality Engineering survey, July to October 2026 (all figures unless marked external)
- TestMu AI (then LambdaTest), Future of Quality Assurance survey report, 2023: https://www.testmuai.com/future-of-quality-assurance-survey/
- External: Stack Overflow, 2025 Developer Survey, AI section: https://survey.stackoverflow.co/2025/ai
- External: LangChain, State of Agent Engineering, fielded November to December 2025: https://www.langchain.com/state-of-agent-engineering

## Changelog

- 2026-11-02: first edition, survey fielded late July to early October 2026.

Related: https://www.testmuai.com/research/ | https://www.testmuai.com/kane-cli/ | https://www.testmuai.com/agent-testing/ | https://www.testmuai.com/agent-assurance/