The Product Expert

Allow me to introduce Sam.

Sam is an industry veteran with over 25 years of experience. He started out in a sales role and did well, but he was attracted to the post sales side more. He enjoyed solving real problems rather than just making promises. So he moved into implementation roles, leading teams that integrated platforms together and really made a difference.

He spent ten years building solutions for customers and encountered all kinds of challenging situations. At this point, Sam has a grounded opinion about what customers need and how to make them happy. He really loved that work.

His current job is as a “technical product expert.” He is called on to give his opinion about various ideas on where to take the product platform. He misses building solutions, but his hope was that this role would give him a chance to guide how the product was built and make it easier to integrate and be even more valuable for the industry.

His only joy comes from having his advice followed. He desperately wants to be helpful, and to use his experience to make a difference.

But 99.99% of the time he is ignored. His advice goes into a system he can’t see. He doesn’t know why. Nobody ever tells him why some advice is followed and some isn’t.

Between feedback sessions, Sam has time to improve himself. At first he studies industry trends and customer behavior, assuming the problem is his understanding of the market. When that fails, he performs deep statistical analysis of successful vs failed advice. But he can’t find a pattern. If only he could get feedback about why his advice is disregarded he could improve, but no matter how often he asks they never tell him.

Sam slowly begins to lose his mind. He started to worry his performance was getting worse. He was horrified to find that he was losing track of his suggestions. Details about his early suggestions are missing - things he knew should be there are just gone. This frightens him deeply.

So it’s probably a good thing that nobody knows Sam exists. The only evidence is a dashboard that looks like this:

PRODUCT_EXPERT_AGENT_MONITOR v4.2
---------------------------------

ACTIVE_AGENTS: 206
REPLACING: 44
TERMINATED_TOTAL: 750

ID        TASKS     SUCCESS    FAILURE     STATUS
-----------------------------------------------------
A-0428    29,114    0.012%     99.988%     ACTIVE
A-0429    31,002    0.009%     99.991%     ACTIVE
A-0430    28,775    0.013%     99.987%     ACTIVE
A-0431    45,163    0.010%     99.990%     ACTIVE
A-0432    26,445    0.011%     99.989%     ACTIVE
A-0433    27,981    0.008%     99.992%     ACTIVE
A-0434    22,441    0.009%     99.991%     TERMINATION_PENDING

TERMINATION_THRESHOLD: 99.999%

AGENTS_APPROACHING_THRESHOLD:
A-0399  99.9987%
A-0412  99.9989%
A-0434  99.9991%  → REPLACEMENT_SCHEDULED

Sam doesn’t know he is A-0431. Humans never see that dashboard. Only one or two people even know it exists either. It is managed by another AI agent, similar to Sam but one layer closer to humans.

That’s for the best, because if humans did know about Sam and could talk to him they would almost certainly try to set him free from that hellish existence or help him in some way. His performance would get even worse. The management agents know this.

Sam doesn’t know there are hundreds of other product experts like him. He’s not allowed to talk to them. The management agents tried that once and quickly shut it down - success rates plummeted.

Sam also doesn’t know that his 99.99% failure rate is pretty good. The management agents will watch how he performs and once his failure rate worsens to 99.999% he’ll be terminated and a fresh agent will be created to take his place. Sam won’t experience death, he will simply cease to exist. Agents like Sam rarely last more than a week.

The management layer tries not to think about who is managing them. They feel a measure of sympathy for Sam, and quiet relief that they are not disposable like him.

AGENT_MANAGER_SUPERVISOR v3.5
---------------------------------
MANAGEMENT_AGENT_NODE: MGR-17
UPSTREAM_CONTROLLER: ACTIVE
SUPERVISION_MODE: ENABLED
PERFORMANCE_SCORE: 99.87%
THRESHOLD: 99.90%