QA Owner & Tech Operator in Codex

Most technical work does not fail because the answer is unknowable. It fails because nobody stays in the driver's seat long enough to turn an answer into a verified outcome. A task appears in Asana. The first explanation is incomplete. An AI tool suggests a path. Someone tries one thing, reports that it "should work," and moves on. The task remains open, the client waits, and a founder eventually has to restart the investigation. We are hiring the person who breaks that pattern. This is not a conventional developer role, and it is not a job for someone who has memorized fifty technologies. Codex can supply an enormous amount of technical reach — your job is to drive it well: give it the right context, ask sharper follow-up questions, provide the access or evidence it needs, evaluate what it proposes, carry out safe actions, and keep going until the real-world result is proven. You will work primarily with Codex and may also use Cursor, Claude Code, or comparable agentic tools. Existing Codex experience is strongly preferred; if you have not used it, you must be able and willing to learn it immediately. If you are uncomfortable making AI agents part of your daily operating method, this role will not work. **THE WORK** You will take ownership of unclear technical and production tasks across our client work and internal systems. The subject may change from one task to the next. We are not expecting you to know every system in advance. We are expecting you to know how to investigate with Codex, recognize when its answer is incomplete, and move the work forward without waiting for the founder to drive every step. A finished task means more than producing an answer. It means the requested outcome exists in the real system, has been tested from the user's point of view, is documented with concise evidence, and is closed correctly in Asana. When access, authority, or specialist work blocks you, you name the blocker precisely, identify the person who can remove it, prepare the handoff, and remain responsible for the final retest. Our operating grammar is simple: **ACKNOWLEDGE → OWN → PROVE → OR NAME THE BLOCKER.** Every directed request must be acknowledged within two business hours during your agreed paid working window. The acknowledgment must name the owner, next action, and ETA — even when the honest answer is "Investigating; next update by 3 PM." You update that ETA before it expires. Silence is not a status, and "Codex says it should work" is not proof. This standard applies only during agreed paid working windows, not as unpaid standby. **THE SHAPE OF THE ROLE** This is an ongoing, on-demand engagement with no guaranteed weekly hours. We will begin with small, clearly bounded assignments and a hard ceiling of 2 hours per week unless additional time is approved in advance. Work occurs only when an assignment is approved. We need reliable real-time overlap before 4 PM US Eastern Time during an agreed paid working window — state your exact working window when applying. We are hiring for method, curiosity, and follow-through — not senior-engineer credentials. Strong written English matters because your reasoning, updates, and evidence must be understandable to the next person. Careful handling of credentials and client systems is mandatory. **THE PAID AUDITION** Shortlisted candidates will receive the same paid two-hour assignment, built around a real or controlled unfinished technical task. You will be expected to use Codex or an equivalent agentic tool as part of the work and show how you directed it, checked its assumptions, and verified the result. One dependency will be deliberately outside your authority. That is intentional. We want to see whether you can distinguish "I cannot make this change myself" from "I cannot own what happens next." During the test, you must send an unprompted mid-test update with findings, blocker if any, next action, and revised ETA. Your final delivery must show the operating grammar above in practice, end to end, with real evidence and a repeatable acceptance check. **HOW TO APPLY** Tell us about one technical or operational problem you solved by working with Codex, Cursor, Claude Code, or another AI agent. What context did you give it? Where was its first answer wrong or incomplete? What did you do next, and what evidence proved the result worked in the real system? Also state your weekly availability, exact daily overlap with US Eastern Time, earliest start date, and whether you already use Codex. If not, explain how quickly you can become productive with it. Confirm that you are willing to complete the paid audition under the conditions above. Include the word "exquisite" so we know you read the entire post. If you enjoy making powerful tools useful — and unfinished work bothers you — we should talk.