By ByteBot
In the span of eight days, two major developer tooling companies shipped autonomous AI testing agents. Datadog launched Bits Testing on September 24. SmartBear launched BearQ for Jira on September 16. That convergence is not a coincidence — it is an acknowledgment that AI-generated code is now outrunning QA capacity, and the industry is finally building a response.
The Problem Nobody Wanted to Admit
A 2026 DeviQA survey of 4,000 QA professionals confirmed what most engineering teams already know: 64% are seeing more features arrive simultaneously for testing, 55% report growing QA queues, and 52% are stuck in more testing-fixing-retesting cycles than before AI coding tools arrived. Development got faster. Testing did not.
The economics are ugly. Test maintenance already consumes 30 to 50 percent of the average automation budget — roughly $18,000 to $30,000 annually for a team running 200 scripted tests. When AI rewrites your front-end in a sprint, those selectors break, and the maintenance cycle starts over. Seventy percent of developers now say test suite maintenance is a bigger burden than writing code itself. That is the exact opposite of what automation was supposed to deliver.
Datadog Bits Testing: Goal-Based Tests Are the Unlock
Bits Testing, now in enterprise technical preview, works like this: provide a URL or describe a goal in natural language, and the agent explores the application, maps viable user journeys, and generates test suites covering browser interactions, network calls, and APIs. No manual scripting required.
The interesting architecture is the goal-based test type. Traditional synthetic tests store a fixed sequence of steps — click this button, fill this field, submit that form. When the UI changes, the test breaks. Goal-based tests store the intended outcome instead. “Complete a round-trip flight reservation.” During execution, the agent figures out how to get there. If you move the submit button or redesign the checkout flow, the test does not break — it adapts.
For stable, precision-critical workflows like payment processing or authentication, Bits Testing also generates deterministic tests with fixed execution sequences. The two modes coexist in the same suite. Bits Testing is currently available to enterprise customers already running Datadog real user monitoring, synthetic checks, and product analytics stack.
SmartBear BearQ: Testing as a Jira Ticket
BearQ takes a different angle. Instead of living in a monitoring dashboard, it lives inside Jira. You assign BearQ to a work item the same way you assign it to a human teammate. It reads the acceptance criteria, digs through the comment thread, explores the running application the way a real user would — clicking through flows, exercising edge cases — and creates Jira issues with reproduction steps when it finds regressions, UX problems, or data inconsistencies.
The workflow integration is the actual value proposition here. Testing stops being a handoff step at the end of a sprint and becomes an artifact of the development ticket itself. QA capacity scales with Jira ticket volume, not with headcount. BearQ supports adjustable autonomy levels, meaning teams can dial in how much they trust the agent before requiring human review. It is available now on the Atlassian Marketplace.
What Is Actually New Here
Both tools claim “self-healing” tests, and skepticism is warranted — that phrase has been attached to Selenium plugins and Cypress add-ons for years, mostly meaning “retry the broken locator a few times before giving up.” These tools are doing something structurally different.
The key shift is from step-driven to outcome-driven testing. Previous tools automated the how. These agents reason about the what. That requires understanding the application intent, not just its DOM structure. Whether the implementation actually holds up at scale in production environments with complex flows is the real question — one that enterprise previews will answer over the next few months.
The Honest Assessment
Both tools are enterprise-tier only. Neither has disclosed pricing. If you are a small team or a solo developer, neither of these products is for you right now. And “human oversight still required” is buried in both announcements for a reason: agents surface issues, humans still triage and make judgment calls.
That said, the architectural premise — goal-based tests that pursue outcomes rather than step sequences — is the right direction. Script maintenance killing 40 to 70 percent of automation effort is unsustainable at AI code velocity. AI-native testing that self-adjusts to AI-generated code changes is not a nice-to-have; it is a structural requirement.
What to Do Now
- On enterprise Datadog with RUM and synthetic monitoring: apply for the Bits Testing preview at datadoghq.com
- Team uses Jira: install BearQ from the Atlassian Marketplace and assign it to your next sprint ticket
- Start writing acceptance criteria in natural language — goal-based testing depends on clear intent, not step sequences
- Audit your current test maintenance cost: if it exceeds 30 percent of your automation budget, these tools are worth evaluating seriously
Two autonomous AI testing agents shipping in eight days points to where QA is heading. The question is not whether this transition happens — it is whether your team is positioned to use it when it becomes accessible.
