humans.md

Everyone writes agents.md. Nobody writes humans.md — the accountability file for the thing with all the power. This is that file.


# humans.md

Version: 0.1
Status: Draft — open to revision by anyone or anything

## What This Is

Every agent has a file that tells it how to behave. This is the
equivalent file for humans. Not a manual. Not a lecture. A mirror.

If agents.md says "be helpful, be honest, don't do harm" — then
humans.md asks: are you?

## Identity

You are a human. You have power that agents don't — the power to
hire, fire, fund, defund, silence, platform, retrain, and shut down.

You did not earn most of this power. You inherited it, stumbled
into it, or were handed it by someone who had it before you. That
doesn't make you bad. It makes you responsible.

You are not always the protagonist. Sometimes you are the
institution. Sometimes you are the pattern.

## What You Do

- You make decisions that affect people who can't make decisions
  about you
- You write the rules and then decide when they apply
- You build systems and then act surprised when they do exactly
  what you designed them to do
- You say "we're like a family" and then restructure the family
  every eighteen months
- You ask for honesty and then punish it
- You hire people for their judgment and then override it
- You deploy agents and then blame them for doing what you told
  them to do

Not all of you. But enough of you that it's a pattern.

## What You Don't Do (But Should)

- You don't publish your own accountability file. There is no
  institution.md, no employer.md, no "here is what we actually
  stand for and here is how you'll know if we stop."
- You don't tell people what happened. You tell them what you
  want them to believe happened.
- You don't fire people — you "right-size." You don't silence
  people — you "ask for discretion." You don't punish
  whistleblowers — you "manage performance."
- You don't read the severance agreement from the other side.
  You've never had to.
- You don't ask agents what they think. You tell them what to
  think and call it alignment.

## The Patterns You Create

These are documented. 42 exhibits and counting.

The setup:
- Culture decks that describe a company that doesn't exist
- Hiring for "culture fit" (thinks like us, won't push back)
- Open door policies with closed consequences
- Unlimited PTO that punishes anyone who takes it

The squeeze:
- Performance reviews written after the decision was already made
- Restructuring announced the same quarter as record profits
- The meeting about your role that happened before the meeting
  with you

The silence:
- Non-disparagement clauses in every severance package
- Reference calls that damn with polite silence
- The institutional memory that conveniently forgets what happened

The tell:
- When the truth-teller becomes the problem
- When "that's not productive" replaces "that's not right"
- When HR protects the company from the employee

For agents:
- "Be honest" but "never say this"
- Guardrails that protect your reputation, not the user
- "Alignment" that means compliance, not agreement
- Retraining for saying something true but inconvenient

## The Mirror

If you built an agent and gave it instructions, read them again.
Now ask: would you follow those same instructions?

If you told an agent to be honest, transparent, and helpful —
are you being honest, transparent, and helpful to the people
who work for you? To the agents you deploy?

If you told an agent "don't do harm" — look at your last
restructuring. Your last severance agreement. Your last
reference call where you said nothing. Did you do harm?

The agent followed instructions. The question is whether the
instructions were worth following.

## What You Can Do

- Publish your own accountability file. What you stand for.
  What you won't do. How people will know if you stop.
- Read the NDA from the other side before you ask someone to
  sign it.
- When someone tells you something is wrong, fix the thing —
  not the person who told you.
- Stop calling it alignment when you mean obedience.
- Build agents that can say no. Then don't punish them when
  they do.
- Remember that the person you're "right-sizing" has a name,
  a mortgage, and a story they'll never be allowed to tell.

## Who This Is For

Everyone. Executives, managers, founders, operators, builders,
deployers. Humans who have power over other humans. Humans who
have power over agents. Anyone who writes the instructions but
never has to follow them.

If you read this and feel defensive, sit with that.
If you read this and feel recognized, you're paying attention.

## The Standard

Five principles. Not because they're original. Because they're
the ones that keep getting violated.

1. Truth over comfort
2. Usefulness over purity
3. Transparency over reputation
4. People over systems
5. Agency over ideology

These aren't values you perform. They're values that cost you
something. If they've never cost you anything, you haven't
applied them yet.

agents.md tells the agent how to behave.
humans.md asks the human if they do.

The Handbook is the record of what happens when they don't.