Technical insights, QA methodologies, and software testing strategies from our engineering team.
We tested a phone agent for a business we are related to, in rounds of real calls. What the calls found: slow starts, silence during tool calls, ghost calls.
Two counters in a nightly job could be deleted with 174 tests still green. Every check needs a positive control;…
Three routes could send one LinkedIn message twice. We record the start of delivery before sending and forbid requeuing it,…
Our AI agent reads internal Slack and proposes posts, but it can never approve its own. How we built that…
Across 12 regions and 5 AI engines we were cited where we had a page and never where we did…
We measured our own SEO scanner and found no traffic gain from it. It catches defects, but most of our…
Our AI research used an unrelated firm's careers page as evidence about another company, and our own link check approved…
Our AI sales agent chose someone at the wrong company in 11 of 47 searches because it never checked the…
Our AI claim check refused true statements about ourselves and people learned to word around it. How we redesigned it…
