Builder's note
Why JobAIRisk was built around tasks, not replacement headlines
Your job title is only the baseline. The useful questions are which parts of your work AI can accelerate, which parts remain more durable, and what you can do next.
People rarely arrive here because they want another abstract AI score. They want a decision: should I change how I work, strengthen a skill, reposition my experience, or consider a different role? A replacement percentage cannot answer that honestly.
The title is useful, but it is not the person
A title gives us a repeatable place to start. It lets JobAIRisk compare 968 occupations using the same task-level method. It does not tell us how a specific employer divides work, how seniority changes responsibilities, or how someone spends a real week.
That distinction matters because 898 occupations in the current release contain tasks from at least two exposure classes. Structured reporting can sit beside negotiation. Routine code production can sit beside architecture and incident ownership. Documentation can sit beside embodied care.
Your job title is not the useful unit of action. Your weekly tasks are.
Why the score stays free
The occupation result is precomputed, instant, and requires no signup. It shows the score, rank, highest- and lowest-exposure tasks, calculation, limitations, and a useful next move before asking for payment. That is enough for many visitors, and it should be.
The paid plan exists for the gap the title cannot close. It uses the work a person chooses to describe, recalculates the task mix, states its confidence, and turns the result into supervised AI moves, a human moat, and 7/30/90-day actions. The model helps structure and explain the input; deterministic application code calculates the final score.
Why we do not publish replacement predictions
Task capability is not a firing decision. Hiring demand, cost, regulation, adoption, management choices, and new work created around a technology all sit outside this score. JobAIRisk therefore does not claim a probability that someone will lose a job or a date when a profession will disappear.
This makes the product less sensational and more useful. A high task-exposure result becomes a prompt to redesign the week, not a verdict. A low result is not a guarantee of safety. Both should lead back to the actual work.
What we will show instead of testimonials we do not have
Until real customers give informed permission to publish their experience, trust has to come from inspectable evidence: 16,578 released task values, visible arithmetic, a complete fictional sample, checksum-locked data, explicit corrections, and a seven-day refund policy.
The first major correction came one day after the initial import. Rather than hide it, we published why the earlier classifier failed and rescored every task. That release is the next story.