AI Can Remove Work and Also Remove the Practice That Builds Professionals

AI agents can research, analyse, draft, recommend and execute work at a speed that changes the structure of a profession. The visible gain is productivity. The less visible risk is that people may lose the repeated practice that lets them notice a bad recommendation, recover from automation failure and exercise independent judgment.

In IRREPLACEABLE with AI #246, Pascal Bornet frames the problem through aviation. Pilots use automation extensively, but they are still required to practise emergencies, manual operation and takeover in simulators. The point is not to resist automation. It is to ensure that the human can still take control when the automation is wrong, unavailable or operating outside its intended conditions.

The professional risk is not only a wrong answer

A professional can be harmed in two ways. First, an AI system can produce an incorrect, incomplete or unauthorised result. Second, the professional can gradually lose the practical ability to recognise the error. The second failure is more dangerous because it removes the independent challenge that should exist around an automated system.

This is especially important in professions where routine work is also apprenticeship. A junior auditor learns what an exception looks like by testing evidence. A lawyer learns the shape of a matter by constructing chronologies and checking authorities. An engineer learns how constraints interact by working through calculations, inspections and design changes. If the repetitive work disappears, the profession must deliberately replace the learning it carried.

Automation must not become a black box apprenticeship gap

Bornet points to the risk of removing entry level tasks before organisations redesign how people gain capability. That is the central implication for Anunnaki.Live. The objective cannot be to automate every visible task and call the remaining human role “oversight.” Oversight without retained competence is only a label.

A Professional Anunnaki Twin should therefore not merely complete work. It should make professional practice observable. It should reveal the evidence considered, the governing source, the choices available, the conditions that caused an escalation and the reason an action was permitted or refused. That enables the human professional to learn from the work rather than becoming a passive approver of it.

What Anunnaki.Live must preserve

1. Manual mode: a professional must be able to take a matter back, inspect its state and continue without hidden system knowledge.
2. Challenge mode: the Professional Anunnaki Twin should expose the evidence, rule, authority and reasoning path so the professional can test the recommendation.
3. Practice mode: meaningful work should be available as guided exercise, including exceptions, incomplete evidence and conflicting instructions.
4. Takeover drills: teams should periodically rehearse work when a connection fails, authority changes, evidence is missing or an automated action must be stopped.
5. Replay: the record must allow an authorised reviewer to reperform a material action and determine whether the result was justified at the time.
6. Competence evidence: completion of a course is not enough. The system should be able to show that the professional can perform, challenge and recover within the relevant scope.

Professional Anunnaki Twin as capability infrastructure

The right design is not human versus machine. It is a governed relationship in which the Professional Anunnaki Twin carries authorised preparation, monitoring and machine executable work, while the professional retains and develops judgment, accountability, challenge and reserved acts.

That changes the definition of productivity. The measure is not simply how many tasks disappear. It is whether the professional and enterprise gain more capacity while preserving the ability to understand the work, intervene safely and grow future experts.

Replay creates the training surface

Replay is often described as an assurance mechanism. It is also a training mechanism. A replayable work receipt can show a trainee or reviewer the starting state, admitted evidence, applicable authority, action path, resulting state and unresolved conditions. The learner can compare the Professional Anunnaki Twin's path with an independent attempt, identify where their reasoning differed and practise the judgment that automation should never silently replace.

For Anunnaki.Live, this is a material product requirement. If a Professional Anunnaki Twin acts within a bounded role, the human must be able to inspect and challenge the boundary. If it prepares or executes work, the profession must retain a living path to competence. If it produces an outcome, the outcome must be explainable and replayable.

The leadership test

Every organisation adopting agents should ask five questions. Can the team still do the work if the agent stops? Can a qualified professional identify a weak recommendation? Are junior professionals still getting deliberate practice? Does the workflow distinguish automated preparation from a human reserved decision? Can the enterprise replay a material result without relying on memory?

Those questions turn AI adoption from a productivity programme into professional capacity design. The aim is not to slow AI down. It is to ensure that the new capacity makes professionals more capable, not less.

Anunnaki.Live should not make the professional a rubber stamp. It should make the professional's work clearer, more continuous, more challengeable and more capable of being taken over when necessary.

Source

Pascal Bornet, “IRREPLACEABLE with AI #246: The Skills AI Quietly Takes Away,” 9 October 2026.