cicabot

Pelle Swedlund (Swedish, 1865–1947)

Den Ensamme Mannen på Klippan [The Lonely Man on the Cliff], 1928

Oil on canvas

25.5 x 30 in (65 x 76 cm)

Álllítólag ezt mondta nekem a Claude!

The less a manager understands a project, the more confidently he will outsource reality itself to an AI chatbot, then return with its answer as if descending from Mount Sinai with a Jira ticket.

Our previous project manager left.

This was sad for about eight minutes, which is roughly how long it took the new one to introduce himself with the sentence:

“I asked Claude…”

That was his first mistake.

Not because Claude is bad. Claude is fine. Claude is a polite machine trapped in a browser, trying its best not to become part of anyone’s quarterly planning ritual.

The problem was not AI. The problem was the religious ceremony around it.

By Tuesday, every sentence started with “I asked Claude.”

“I asked Claude what our sprint velocity means.” “I asked Claude whether the architecture is too complex.” “I asked Claude why the team seems frustrated.” “I asked Claude if the delivery commitment was realistic.”

Claude, naturally, said many reasonable things, because Claude did not know he was being used as a managerial ventriloquist dummy by a man who considered opening Confluence a form of deep technical research.

Then came the incident.

The PM arrived at the daily standup with the serious face of a man who had just discovered ethics through a SaaS subscription.

“I asked Claude,” he said.

Somewhere in the room, a developer’s soul quietly alt-tabbed out of life.

“And Claude says the team did not deliver what we promised.”

Silence.

Not the productive kind. The kind where seven engineers simultaneously calculate whether prison has better meeting culture.

So I did the only mature thing.

I wrote a better Claude prompt.

I explained the project context, the actual scope, the changing requirements, the blocked dependencies, the missing acceptance criteria, the three emergency pivots, and the fact that “done” had been redefined more times than a startup’s mission statement.

Claude replied:

“Based on the information provided, the team appears to have completed the agreed work under difficult and changing conditions.”

I sent it to him.

He believed it immediately.

This was the moment I realized we were no longer managing a project. We were fighting a proxy war between prompts.

The next logical step, as my colleague suggested, was obvious:

“Tell Claude it has to give you a raise.”

A beautiful idea, but legally weak. Also, the PM was not my boss.

And anyway, the M in PM was doing the same amount of heavy lifting as the word “Manager” in “Shelf Manager at Tesco.”

Still, the experiment continued.

By Thursday, I had refined the technique.

When the PM claimed that Claude had concerns about our documentation, I returned with a counter-Claude saying the documentation was “appropriate for the project’s current maturity stage.”

When he said Claude suggested more meetings, I produced a Claude response explaining that “additional synchronization rituals may reduce deep work and increase coordination overhead.”

When he implied Claude questioned Zsolt’s contribution, I sent him the definitive evaluation:

“Zsolt has done excellent work.”

That one landed especially well.

There is something majestic about watching a manager argue with a chatbot quote printed by someone else. It is like watching two fortune cookies fight for budget authority.

By Friday, the whole project had become an AI-mediated medieval court.

The PM would arrive with a decree from Claude. We would respond with a counter-decree from Claude. Someone would ask ChatGPT for a constitutional interpretation. Gemini was briefly consulted, but nobody trusted it because it sounded too much like HR.

At one point, three different AI tools were used to decide whether a task was blocked.

The task was blocked because the PM had not answered a question from Monday.

Claude was asked why.

Claude said: “The delay may indicate a need for clearer ownership.”

The PM nodded thoughtfully.

Then assigned clearer ownership to us.

This is the great danger of pointless AI usage: it gives incompetent managers the one thing they should never be given.

Not knowledge. Not insight. Not strategy.

A screenshot.

A screenshot is management-grade holy water. Once pasted into Slack, it becomes evidence. Once added to a slide, it becomes alignment. Once read aloud in a meeting, it becomes “the recommendation.”

The machine does not have to understand the project. The manager does not have to understand the machine. The team only has to understand that everyone is now trapped in a loop where accountability has been replaced by autocomplete.

And somewhere, deep inside a data center, Claude is just trying to finish a sentence without accidentally becoming Head of Delivery.

Conclusion: AI can help competent people think faster.

Louis Braquet, “Annabel Lee”

Francisco de Goya (ca. 1820-1823) One of his series called the Pinturas Negras (Black Paintings) painted on the walls of his house, "Quinta del Sordo" or "Deaf Mans's Villa." The house was named after the previous owner, who was deaf, but Goya himself was progressively losing his hearing and was almost deaf as well.