Slack engineering has launched an strategy known as agentic testing that explores how AI agents can be incorporated into end-to-end testing to enhance resilience in dynamic software program methods. to enhance resilience in giant distributed methods. The work targets a typical problem in steady supply environments, the place end-to-end exams steadily fail as a consequence of consumer interface or service adjustments slightly than precise useful regressions, resulting in elevated upkeep overhead for engineering groups.
Conventional end-to-end exams depend on fastened steps, secure selectors, and predictable flows throughout UI or APIs. In fast-changing methods, these assumptions typically break, rising upkeep effort. Slack engineers describe agentic testing as shifting a part of this duty from static scripts to AI-driven brokers that execute based mostly on higher-level intent.
On this mannequin, a check is expressed as an goal slightly than a strict sequence of actions. An AI agent interprets the intent and makes an attempt to finish the workflow by interacting with the appliance by means of UI or API surfaces. The agent evaluates the appliance state at every step and selects actions dynamically. When minor adjustments are encountered, resembling modified UI construction or relocated components, the agent makes an attempt alternate paths to proceed execution as a substitute of failing instantly. The execution is then validated in opposition to anticipated assertions outlined by engineers.
The workflow usually begins with a check intent being handed to the agent layer. The agent performs planning, executes actions in opposition to the system beneath check, observes outcomes, and iterates till the target is accomplished or a stopping situation is reached. The result is then recorded together with execution traces that seize the sequence of selections and interactions taken through the run. Slack engineers weblog that as a consequence of value issues, agent pushed testing is presently higher fitted to focused debugging and exploratory testing slightly than frequent execution in steady integration pipelines.
Conventional testing movement:
click on → click on → sort → assert
Agentic testing movement:
purpose → agent adapts → confirm outcome
Slack engineers word that deterministic exams proceed to function the first mechanism for validating crucial logic and contract correctness. Agentic testing is positioned inside the end-to-end layer the place workflows are extra delicate to UI and structural adjustments. The agent-based strategy is used to scale back failures attributable to superficial adjustments that don’t mirror useful regressions.
The system additionally consists of constraints to control agent conduct throughout execution. These embody limits on allowed actions, boundaries for exploration, and situations beneath which execution ought to cease. Observability is a key requirement, and execution logs are structured to supply visibility into every step taken by the agent, enabling groups to replay and examine failures.
Testing pyramid with 4 layers: Unit Exams, Integration Exams, E2E Testing, and Agentic Testing (Supply: Slack Blog Post)
Slack engineering positions agentic testing as a complementary functionality slightly than a alternative for present testing approaches. Deterministic end-to-end exams proceed to help quick, repeatable regression validation in CI, whereas agent-driven execution is utilized the place UI adjustments introduce brittleness. Scripted or generated exams present secure verification for predefined journeys, whereas agent-based execution operates from a goal-oriented mannequin, observing utility state and dynamically figuring out the way to attain the specified consequence. This makes it helpful for exploring advanced UI conduct, debugging flaky workflows, and reproducing manufacturing points, alongside conventional deterministic testing inside the end-to-end layer.