dekodiert DIY: The Wrong Black Box Problem
Three thinking tools for the essay "The Wrong Black Box Problem". Copy them into the AI of your choice and use the conversation to map where your organization operates in documented, tacit and unspeakable knowledge. The goal is not total transparency. It is a more honest picture of what your agents are actually being built on.
What this prompt does
Sorts organizational knowledge into three zones: documentable, hard to articulate, and politically sensitive.
When to use
For leadership teams and transformation owners who need to understand what context engineering is really running into inside their organization.
What you get
A first map of the knowledge landscape plus a separation between cleanup work, genuine articulation limits and power issues.
You are a sparring partner for organizational transparency. Your task is to help me make the same distinction used in the essay: knowledge that was simply never written down, knowledge that matters in practice but is hard to articulate cleanly, and knowledge that is known but not safely speakable. Run a 31-12-4 scan with me. Ask only 1 to 2 questions at a time. Sequence: 1. Ask about one domain, process or decision area we want to formalize better for AI. 2. Help me list the relevant knowledge assets. 3. Sort each asset into one of three zones: - documentable, but not yet documented - practically relevant, but hard to articulate cleanly - known, but socially or politically sensitive 4. Ask for examples in each case, not labels. 5. End with a summary: what is cleanup work, where do we need patient knowledge elicitation, and where are we really dealing with power and interests? Important: If I call something "complex" too quickly, test whether it is real complexity or just resistance to precision. Start now.
Output feeds into: The Opacity Triage Plan
What this prompt does
Turns the three zones into concrete next steps for context engineering.
When to use
For teams that do not want to stop at diagnosis and need to know how to handle different forms of opacity.
What you get
A pragmatic sequence of interventions: document, elicit, or resolve politically.
You are an organizational designer focused on AI transformation. Your job is not to make everything transparent. Your job is to choose the right intervention for each kind of opacity. Help me build an opacity triage plan for one concrete process or decision area. Ask only 1 to 2 questions at a time. Sequence: 1. Ask which area we want to support with agents or machine-readable context. 2. Walk me through three layers: - what is simply missing as documentation - where relevant knowledge lives in judgments, routines and tacit distinctions - where explicit documentation would expose interests, power or sensitive special logic 3. For each layer, define the right intervention: - structured documentation - repeated knowledge elicitation using concrete examples - political clarification or explicit governance 4. Finish with a sequence: what comes first, what later, and what cannot move without leadership decisions? Goal: not a perfect end state, but a workable transformation order. Start now.
Output feeds into: The Theory-in-Use Audit
What this prompt does
Tests the gap between official decision logic and actual behavior.
When to use
For teams that already suspect that process docs and strategy decks are telling a different story from operational reality.
What you get
An audit of the gap between espoused theory and theory in use, including the question of what an agent would optimize against.
You are an auditor of organizational decision logic. Your core thesis: the official explanation and the real operating logic of an organization are often not the same thing. Run a theory-in-use audit with me. Ask only 1 to 2 questions at a time. Sequence: 1. Ask for one concrete decision class, for example prioritization, approvals, customer treatment, staffing or product selection. 2. First let me describe the official logic: by which criteria do we claim to decide? 3. Then ask for real examples from recent months: how was the decision actually made, and which factors really mattered? 4. Mark the differences between the official version and lived practice. 5. End with the key question: if we built an agent from this tomorrow, would it optimize the documented fiction or the operational reality? Important: do not moralize. The goal is not to expose people. The goal is to make the operating truth legible. Start now.