Posts

Testing Orders and Refunds with Autonomous Agents

By Bogdan Stefan Plesa Orders and refunds turn a browser journey into a stateful business investigation. The page may announce success while the underlying financial behavior is wrong. What changes Define external invariants: one order identity, the expected total, traceable status changes, and a refund that moves the correct amount exactly once. Then retry deliberately with the same idempotency context and after simulated interruptions. The FaST standard Compare independent layers such as UI state, order API, payment stub, and ledger. If they disagree, preserve the disagreement rather than asking the model to choose the friendliest interpretation. Adaptive exploration is useful only when the financial oracle remains deterministic. FaST: Fully AI Software Testing — Concept and Paradigm, created by Bogdan Stefan Plesa. Explore the open reference project , read about the author , and inspect the source notes and limitations . Prepared with AI assistance and reviewed for consistency with ...
Post a Comment
Read more

Testing User Creation, Disabling, and Stale Sessions

By Bogdan Stefan Plesa User lifecycle testing shows why adaptive investigation matters. Creation, authentication, disabling, and session revocation interact across time. A happy-path script can check every screen while missing the relationship between them. What changes Create a synthetic user and verify the durable identity, assigned role, and uniqueness rule. Establish protected access before disabling the account, preserve the session, then probe that same session immediately after revocation. The FaST standard Check every layer: UI status, identity record, fresh authentication, and established-session access. Differences are findings. The FaST reference deliberately includes a stale-session fault because a convincing workflow can easily miss it when actions occur in the wrong order. FaST: Fully AI Software Testing — Concept and Paradigm, created by Bogdan Stefan Plesa. Explore the open reference project , read about the author , and inspect the source notes and limitations . Prepar...
Post a Comment
Read more

How an AI Agent Should Test Login Flows

By Bogdan Stefan Plesa A login test that enters valid credentials and sees a dashboard is useful but narrow. An autonomous agent should investigate authentication state: valid and invalid credentials, role-specific access, logout, lockout, expiry, and revocation after account disabling. What changes Observe more than the page. Capture the final URL, response status, session transition, visible identity, and a protected-resource probe. A dashboard heading may be stale content; a redirect may hide an error page. The FaST standard Use synthetic users with known roles and states. Preserve an old session before testing a fresh login, or the new attempt may destroy the evidence needed to reveal a stale-session defect. The output should be an evidence-backed map of authorization transitions. FaST: Fully AI Software Testing — Concept and Paradigm, created by Bogdan Stefan Plesa. Explore the open reference project , read about the author , and inspect the source notes and limitations . Prepared...
Post a Comment
Read more

Evidence-First Oracles for AI Software Testing

By Bogdan Stefan Plesa Every test needs a way to decide what an observation means. In autonomous testing, the oracle cannot simply ask the model whether the result looks correct. Coherent interpretation is not an independent check. What changes Prefer consequences over friendly messages: refund counts, authorization decisions, inventory movements, protected-resource responses, and durable audit events. UI text can support a finding, but should rarely be the only evidence. The FaST standard FaST keeps pass, fail, inconclusive, and policy violation separate. Each result links the invariant to the observation, timestamp, and tool provenance. Known-fault fixtures then test whether the oracle catches real defects without failing healthy behavior. FaST: Fully AI Software Testing — Concept and Paradigm, created by Bogdan Stefan Plesa. Explore the open reference project , read about the author , and inspect the source notes and limitations . Prepared with AI assistance and reviewed for consist...
Post a Comment
Read more

Why AI Test Agents Need Tool Boundaries

By Bogdan Stefan Plesa Testing agents need power to create data, call APIs, operate browsers, and inspect results. That power is also the main safety problem. A prompt that says “be careful” is not an access-control system. What changes The planner should request actions through narrow, typed interfaces. An independent execution layer validates the action, arguments, target environment, data scope, and remaining budget. A model cannot grant itself a new tool or expand its own permissions. The FaST standard Blocked actions must remain visible in the evidence ledger. Web pages, logs, and tickets are untrusted observations, not authority. Autonomy without enforceable tool boundaries is uncontrolled execution with a persuasive narrator. FaST: Fully AI Software Testing — Concept and Paradigm, created by Bogdan Stefan Plesa. Explore the open reference project , read about the author , and inspect the source notes and limitations . Prepared with AI assistance and reviewed for consistency with...
Post a Comment
Read more

The Five Layers of an Autonomous Testing Agent

By Bogdan Stefan Plesa An autonomous testing agent is not one prompt wrapped around a browser. A defensible design separates five responsibilities: objective, observation, planning, tools and policy, and evaluation. That separation prevents one fluent model response from silently becoming the product verdict. What changes The objective states the question and limits. Observation preserves browser state, API responses, identifiers, timestamps, and provenance. Planning chooses the next experiment. The tool layer validates every requested action against permissions and budgets. Evaluation checks independent business invariants. The FaST standard Tracing and reporting cut across all five layers, but a trace is not an oracle. The architectural goal is to make errors inspectable: when the planner is wrong, the evaluator should still expose the missing or contradictory evidence. FaST: Fully AI Software Testing — Concept and Paradigm, created by Bogdan Stefan Plesa. Explore the open reference ...
Post a Comment
Read more

From Test Cases to Testing Objectives

By Bogdan Stefan Plesa A test case tells a system which path to follow. A testing objective asks a bounded question. In Fully AI Software Testing, that difference lets an agent adapt its next experiment to what it actually observes instead of replaying steps after the evidence has changed. What changes Consider the objective “establish whether disabling a user revokes every path to protected resources.” A fixed script may test only a fresh login. An agent can preserve an existing session, disable the account, probe that old session, and separately verify that new authentication fails. The FaST standard Useful objectives name the target, allowed tools, environment, business invariants, evidence required to stop, and escalation conditions. The agent gets room to investigate; it does not get authority to redefine success. Adaptive testing should remain purposeful, reconstructable, and accountable. FaST: Fully AI Software Testing — Concept and Paradigm, created by Bogdan Stefan Plesa. Expl...
Post a Comment
Read more