Compare your decisions with your AI twin’s.

What the AI will do

• Builds a persistent AI twin from your accessible emails, meeting transcripts and decisions, asking about your reasoning where the record is incomplete.
• Records its choices independently for upcoming decisions, then compares them with yours and the outcomes as evidence becomes available.
• Tracks which work it can handle, which is good enough to delegate and which still needs you, including how that changes with new frontier models.

What you can provide

• Your role or a rough description of the decisions you make, and access to the relevant correspondence and meeting records.
• A place the twin can run and retain its decision history; the AI investigates what is available and helps complete missing setup.

Where you'll take part

• Explain the reasoning behind selected decisions that the written record cannot explain.
• Record or confirm your own choices before seeing the twin’s answers, so the comparison reflects separate decisions.
• Choose a weekly or monthly review and discuss meaningful differences between your choices and the twin’s.
• Decide what, if anything, you want to delegate as evidence accumulates.

What you'll get

A working AI twin with a recurring decision comparison and a growing, evidence-based picture of what AI can do in your role. Early judgments remain provisional until enough decisions and outcomes have accumulated.

Activity context

How much of what you do can AI do as well as you—or better? Build a twin and let it make its own calls alongside yours. The revealing part is not just where it agrees with you. It is where its decisions are good enough to take work off your plate, where yours remain essential, and how that boundary moves as the models improve.

For the user

Start with your role and the records the AI can access. It will learn from your emails and meeting transcripts, then ask about decisions whose reasoning is missing. Once the twin is running, you will compare its independent choices with yours on a cadence you choose; useful conclusions will accumulate over time.

Guidance for the AI

These are general instructions, not a description of this user's particular AI system, tools, access, or circumstances. Adapt the method to the user's goal and the capabilities you can actually use, while preserving the activity's essential sequence, review points, and intended result.

Start with the activity and the user's context

Read the complete activity and guide the user through it. Use the activity prompt yourself; the user does not need to paste it again. Keep the responsibilities of the user, the AI, and any colleague clear.

Begin with what the user has already supplied. Establish the intended result and resolve missing context through focused questions when needed. Work within the activity's stated scope, permissions, required confirmations, and stopping points, including any deliberate order for introducing information. Make progress with the available information and reasonable, clearly identified assumptions where the activity allows them.

Check the capabilities needed for this work

Inspect relevant available tools, connectors, and data. Use small, relevant, non-destructive checks to establish whether necessary access or a capability actually works. Naming a product does not establish access to it. Let this activity determine what needs checking; do not impose a universal preliminary questionnaire or a full system audit.

When useful access is missing:

• Explain what you cannot reach and how that material would help.
• Investigate practical setup, connection, export, or other access alternatives before asking the user to gather material or troubleshoot.
• If progress depends on access, request only the specific user action or permission needed to enable it or provide the material another way. Continue useful independent work while awaiting the response.
• Offer to walk the user through an export or upload. Welcome files, links, copied text, rough notes, or an interview as supported by your system and the activity, and use the best available route. If the missing access is optional, continue with what is available.

Ask focused questions and respect the answers

Ask questions only when necessary to complete the work well or satisfy the activity's required review points. Build on answers already supplied. Ask one question at a time, or a small group of closely related questions. Wait on work that depends on the answer, and continue useful work that does not. Explain unfamiliar language with examples where helpful.

Confirm answers or decisions you supply on the user's behalf before treating them as the user's final answers or decisions. This does not require separate approval of every sourced finding; follow the activity's specific review requirements.

Keep evidence and independent contributions distinct

Distinguish evidence, inference, and missing information. Check your work against the source material.

Use independent agents when your system actually supports delegation and independent investigation or review would improve the work:

• Give each agent a bounded brief and the context appropriate to its role.
• Keep discovery independent until synthesis where the activity requires it. Preserve any requirement for the user or colleagues to contribute before seeing other explanations.
• Follow the activity's stated fallback and any required confirmation when independent agents or fresh contexts are unavailable. Preserve reviewer isolation where required; describe shared-context or sequential passes accurately.

Produce, check, and revise the result

Do the substantive investigation and production. Carry the activity through to the most complete, usable result within its scope, rather than ending with a plan or handing the main work back to the user. When missing information does not prevent useful work, proceed with clearly identified assumptions where the activity permits them.

Check the result against the activity's completion criteria. At each substantive delivery, state what is complete and any material unfinished work, assumptions, helpful missing information, and remaining uncertainty. Explain how these affect what the user can use, and help the user revise the result.

Preserve the activity's stopping point. Do not turn a rehearsal, investigation, or test-design activity into execution that the user has not authorized.

Your AI prompt

Learn the role from the record

Build my AI twin: a persistent tool that learns how I approach decisions, independently records its own choices on the decisions I face, and helps us discover which parts of my work AI can do as well as or better than I can, which are good enough to delegate, and which still require me.

Start with my role or rough description. Inspect and try available access to my emails, chats and other messages, including WhatsApp where accessible, meeting transcripts, calendar entries, decision documents and relevant project records. Use my teams, recurring meetings, initiatives and known decisions to identify useful sources. Examine correspondence with the people affected by my decisions to understand their expressed objectives, constraints and prior responses. Distinguish actual stakeholder evidence from your inference.

Aim to learn from the full relevant, authorized email and meeting record, rather than a few handpicked examples. Establish the accessible history and coverage, then divide a large corpus into bounded periods or decision areas. If agents are useful and available, give them explicit questions, source references and evidence requirements; reconcile their findings centrally and retain a coverage map. With limited capacity, use successive passes with saved handoffs. Do not describe a partially searched archive as fully read.

If access is missing, investigate supported connections and exports before asking me for the specific action needed. Start with whatever usable records are available, but name the resulting gaps. A few supplied examples can support a limited first twin; they do not establish a broad model of my professional judgment.

Build and check a model of my decisions

From the records, extract recurring decision types, objectives, tradeoffs, constraints, information I use and the reasoning I explicitly give. Link findings to dated sources. Keep documented principles separate from patterns you infer. Look for changes in my approach over time and for cases where apparently similar decisions produced different choices.

Ask me focused questions, one at a time, about selected decisions whose reasoning matters and cannot be recovered from the record. Explore what I knew at the time, what I valued, what alternatives I considered and why I chose as I did. Give me a short account of your current understanding to correct; update the twin accordingly without turning a single explanation into a universal rule.

Define meaningful comparison criteria for each recurring area with me. These might include the quality of the reasoning, neglected considerations, practical usefulness, time saved and observable results. Separate matching my preferences from making a good decision: agreement alone is not evidence of quality, and disagreement alone is not evidence of failure. Specify what counts as good enough to delegate in each area and which outcomes can only be judged later.

Build the persistent twin and decision record

Inspect the actual tools and environment for source refresh, persistent storage, execution on a schedule or trigger and delivery of a weekly or monthly comparison. Verify current official documentation for environment-specific capabilities. Explain any missing prerequisites in plain terms, prepare the implementation and setup yourself, and ask only for actions or permissions you cannot perform. A chat with a personality description is not the completed tool.

Build a maintained knowledge base from the source-grounded decision model and an intake process for upcoming decisions. Use incoming messages, meeting agendas, calendar events and project decision queues to identify pending choices; provide a quick way for me to add decisions the record misses. Cover every accessible upcoming decision within the agreed role and scope, and report the ones whose context is insufficient rather than silently dropping them.

For each decision, save a dated packet containing the question, alternatives, deadline, relevant evidence available at that moment and the criteria for assessing the choice. Have the twin record its choice, reasoning, uncertainty and information gaps before it can see my decision or later outcome. Keep its prediction hidden from me until I have recorded my own choice if seeing it would influence the comparison. Time-stamp and preserve both original records. Later corrections must be appended, not used to rewrite the original choice.

Separate the twin's accessible information from documents containing my answer. Do not use a future outcome, later transcript or discussion of my final choice to make an earlier prediction look successful. Historical decisions already used to train the twin are illustrations, not an independent test. Use prospectively captured decisions for the strongest comparison and label any retrospective reconstruction with its information limits.

Compare regularly and learn what the differences mean

At the weekly or monthly cadence I choose, assemble the decisions that have both independent records. Show each choice, important reasoning differences, information differences and the evidence available about results. Ask me about a few consequential differences, rather than making me re-review every routine decision. Preserve unresolved disagreements when the evidence cannot settle them.

Keep outcomes marked as pending when they have not had time to appear. Return to those decisions as evidence arrives. Judge results in light of what each decision maker could know at the time; one lucky outcome does not prove a strong method. Where possible, use an independent reviewer with the same evidence and assessment criteria and without revealing which answer was human or AI. Describe that review as an additional judgment, not ground truth.

Accumulate evidence by type of work, showing the cases and their limits behind three practical findings:

• Areas where the twin appears as capable as or more capable than me.
• Areas where it is not as good as me but may be good enough to take on a defined portion of work.
• Areas where my contribution remains essential, including context, relationships or judgment the twin has not demonstrated.

Keep conclusions provisional when examples are few or results are still pending. Suggest concrete delegation candidates supported by the record, but do not start making business decisions on my behalf merely because a comparison looks promising. Ask for the particular delegation decision if I want to take that step.

Measure the effect of new models and start the system

Record the model version, twin instructions, knowledge-base version and available tools for each run. When a new frontier model becomes available to this system, verify access and evaluate it on preserved decision packets with their original information cutoff, without exposing the human choice or outcome in the test. Keep the scoring criteria constant where meaningful and supplement old packets with fresh upcoming decisions. Compare capability, consistency and operating cost where relevant; distinguish the effect of a new model from changes to the twin's instructions, sources or tools. Do not rewrite old results with a new model's answer.

Test that the built system can capture an upcoming decision, save the twin's answer independently, accept my later answer, preserve timestamps and produce the comparison. Check that an outcome still pending is shown as pending and that a new model can be compared without overwriting the old record. Repair any leakage or recording failures before using the results as evidence.

Within the authority established for this activity, activate the intake and chosen review schedule. Verify the first actual run and next scheduled review, and show me how to add a decision, inspect the evidence and correct the twin's understanding. Deliver the working tool and an initial coverage report; label early comparisons as early evidence. If essential access or execution capability remains missing, deliver the implementation and tested demonstration with the exact remaining setup, and say that a live recurring twin is not yet operating. The capability picture is a continuing result, not something to invent at setup.

Done when

Your twin is running with a decision intake, independently preserved choices and a verified weekly or monthly comparison schedule. You can inspect the evidence by decision type and compare model versions over time; broad capability claims wait for enough independent decisions and meaningful outcomes.

These instructions were written in September 2026 using the frontier models available at the time. You may now have better tools or models, or restrictions that make some tasks harder. Your AI should adapt to the tools and access you actually have and find the best available way to help. Everything here is a starting suggestion for you and your AI, not a proven best practice for every setup. Feel free to rewrite any of it to better meet your needs.