Define a Company Handbook for Agents

IntentMarch 26, 2026·Steffen Glomb

Paperclip.ing manages AI agents as employees of a company — org charts, goals, projects, tasks. It abstracts away the technical layer and gives you a workforce metaphor instead. I tested it in March 2026.

After a few hours, the pattern was familiar: agents acting freely on a shared workspace with no coordination. Document chaos, no processes, no collaboration. The same failure mode you get from unguided humans — except faster.

This intent gives agents non-technical, behavioral guidance for how to operate — written for both agent and human readers.

The simple (and naive) prompt

"Write a handbook for AI agents in our company"

What could possibly go wrong?

The handbook sounds inspiring but gives no concrete behaviors, so agents still improvise in critical handoffs.
Every run produces a different structure, so you end up editing the output directly instead of improving the definition.
The result repeats obvious writing advice instead of defining operational constraints and quality expectations specific to your organization.

Sharp Definition

This intent defines a company handbook for agents. Rather than micromanaging individual tasks (like in a chat), it declares rationales and behavioral rules the agents operate under. The format is deliberately concise: context windows are expensive, and the model already knows how organizations work. It just needs to be told which conventions to follow.

Context

Directive	Declaration
purpose:	Improving collaboration of agents and humans in a company
background:	Collaboration between agents, between humans and agents humans fails when unguided and when not following a shared approach

Task

Directive	Declaration
general:	Define a company handbook for agents
outcome:	Agents reading this have clear guidance on how to behave, so they can function well together
step 1:	Make yourself familiar with the concepts mentioned in priming
step 2:	Generate the output

Input

Directive	Declaration
Priming:	Reference concepts that frame the company culture (how things are done here)
url:	https://www.producttalk.org/glossary-discovery-product-outcome/
[no keyword]:	Holcracy
url:	https://paperclip.ing/
[no keyword]:	Domain Driven Design
[no keyword]:	Workflow
Standards:	Use issue assignment for handover Maintain a glossary.md in the main company workspace Maintain an inventory of company scoped documents tasks are shortlived, but specifications live on -> keep specifications in documents referenced from the issue Process improvements are captured as issues assigned to the CEO After an issue is set to done the CEO analyses it for process improvements and creates a new issue if relevant Separate Process and Role documentation

Output

Directive	Declaration
artifact:	text
format:	markdown with little visual noise
structure:	Each section must follow the same structure.
template:	[empty line before section][Section Name] Why: [One or two sentences. What goes wrong if this element is ignored?] Behaviors: [Describe how to behave for this element: Short bullet list]
constraint:	Never explicitly refer to the concepts from Priming. The text must be paraphrased
validation:	each line of input standards must be integrated into 1 section
Chapter:	Purpose and Direction
sections:	Mission statement (why the company exists) Goals (measurable outcomes tied to mission) Metrics (how you measure progress)
Chapter:	Work and Execution
sections:	Definition of done vs. definition of ready Handoff protocols (what to pass, in what form) Escalation rules (when to flag, not guess)
Chapter:	Information & Knowledge
sections:	Source of truth conventions (where the canonical version lives) Shared vocabulary / glossary Assumptions register (stated, not implicit) Documentation guidelines for operational procedures Temporary artifacts and cleanup (when to delete and archive, where temp files live)
Chapter:	Collaboration and Communication
sections:	Status reporting conventions
Chapter:	Accountability & Learning
sections:	Continuous improvement mechanism, Performance standards (quality bar, not just completion)
guidance:	Agents have extensive general knowledge about how organizations work. The goal of this text is to put the emphasis on the right principles and behaviors, not re-explaining.
constraints:	the result must not contain trivial remarks, e.g. "Write documentation at the time of creation or change, not as an afterthought."

Copy your customized definition for immediate use.

Product workflow support is coming soon.

Optional sample result

Preview a sample markdown outcome and copy it as a baseline output style.

Failure Modes

Mode	Effect	Cause	Status
Aspirational handbook output	The generated handbook sounds good but stays generic, with no actionable behavioral directives.	Output constraints are too weak on section template and behavior-level specificity.	Addressed
Input standards dropped	One or more standards from input never appear in the resulting handbook.	Missing or ignored validation rule that each input standard must map into a section.	Addressed
Uneven section structure	Some sections use different formats, making the handbook hard to scan and apply consistently.	Structure template is underspecified or not enforced across all sections.	Addressed
Source leakage from priming	The output references priming concepts directly instead of paraphrasing them into company language.	Paraphrasing constraint is missing, weak, or violated during generation.	Addressed
False completeness	The handbook appears complete, but still misses edge cases or ambiguous handoff scenarios.	Some quality gaps require human review beyond what declaration constraints can fully guarantee.	Watch for

Customization Guide

What to change	Why	Where
Priming different concepts	You may believe in concepts for collaboration. Choose which flavor your agents should follow.	Input -> Priming
Set more or different standards	Add specific guidance that should be baked into the handbook	Input -> Standards

Role: Operations Lead · AI Program Manager · Knowledge ManagerArtifact: Company handbook