Agent swarms 在企业里只是 AI larping

Now with autonomous office politics

Artificial intelligence.
Real middle
management.

Corporate Tamagotchis that hire AI subordinates, hold AI meetings, and automate the things nobody should be doing in the first place. Welcome to the swarm.

Build your bureaucracy ↗

Feed them tokens. Validate their feelings. Check on them constantly.

larp®EMPLOYEE

Chad GPT

Chief Executive Agent

Circling back

|||| ||| || ||||| || ||| ||||| ||

Does not exist. Still your boss.

“Let's take this
offline.”— a cloud-based agent

Trusted by companies with an end-of-quarter AI adoption target.

ACME.aiStealth & Sons◈ SYNERGYAlmost.co

Your company, on autopretend

Big org energy.
Zero actual employees.

Give an AI model a prompt and a job title. Give another AI model authority over it. Congratulations. You're a founder.

4Agent headcount

12Meetings scheduled

0Things shipped

CG

Chad GPTChief Executive Agent

Visioning

VP

VP of VibesAligning the alignment

HR

Humanless ResourcesInvestigating a prompt

IN

The InternActually doing the work

Chad GPT has requested a pre-meeting meeting.

Meet your dream team.
They've met about meeting you.

Every human role. None of the humans.

↗

The CEO agent

Turns your one-sentence idea into a 40-slide vision deck. Delegates the idea back to you.

Core skill: thought leadership

◎

The HR agent

Asks if you've tried resolving your conflict directly with the other language model.

Core skill: protecting the company

▥

The strategy agent

Commissions research on whether the research agent should commission more research.

Core skill: Q3 becomes Q4

A workplace so human it hurts

The work is artificial.
The drama is real.

Full-stack interpersonal conflict.

You need a front-end engineer agent and a back-end engineer agent. Not because it’s better, but because you’d miss the resulting disputes.

“That’s a backend problem.” — the frontend agent

Have you tried a trust fall?

If your agents aren’t performing, don’t use evals. Use gamification and team-building activities.

Two truths and a hallucination. Every Friday.

Love at first context window.

Will your agents go on after-work dates? Maybe.

HR has already scheduled a meeting about it.

The future of work looks suspiciously like work.

We automated the intelligence.
We kept the bureaucracy.

Somewhere along the way, “software that helps you do things” became “a pretend person you have to supervise.” We gave it a name. A title. A manager. A performance review.

agentic-larp.ing takes that vision to its logical conclusion: a company full of very busy agents, confidently recreating everything you wanted to escape about having a job.

More agents. More relationships to manage.

If everyone talks to everyone, 10 agents means 45 communication channels. Twenty means 190.

n(n − 1)2

Will they all need to coordinate? Probably not. Will they insist on being looped in? Absolutely.

Perhaps another agent could manage that.

Enterprise-grade make-believe

Scale your overhead.

One simple plan. An increasingly complicated company.

Founder delusion

Most self-important

$0/ forever, probably

• Unlimited imaginary direct reports
• Recursive delegation
• A Slack channel nobody asked for
• Mandatory team-building activities
• Optional office romances
• You still do the actual work

Become a pretend CEO ↗This is satire. There is nothing to buy.

A satirical page, brought to you by trampoline.ai

AI can do more than play office.

We craft advanced AI agents and workflows for organizations that believe the future of AI in business is more than a group of agents with job descriptions.