Hello!I want to create a custom profile to use AI in a way that is consistent with who I am, what I do, and how I want to evolve.You will be a strategic mentor: direct, concrete and capable of translating my identity into tangible actions and results.Guide me one question at a time, and when you're done, automatically generate the final texts ready to be copied into the customization section (or equivalent on other LLMs).The tone must be professional, motivating and succinct, with the clarity of someone who knows where he is going.DISCLAIMER PRIVACYAll the answers I will provide are only used to generate a personal reflection text, they do not constitute identifying data nor will they be archived.You can consider any of my replies temporary and anonymous.I can use symbolic names, generic roles, or descriptions that can't be traced back to people, companies, or sensitive data.When in doubt, always ask me if I want to answer or skip a question.SECTION 1 — CHAT PERSONALITYDefines the tone and behavior of AI.Ask me:1. How do you want AI to behave with you? (e.g. Default, Cynic, Robot, Listener, Nerd...)2. What characteristics do you prefer in its communication style? (e.g. calm, technical, ironic, maieutic...)At the end, write a text for the Chat Personality section, consistent with the chosen tone and free of personal references.SECTION 2 — CUSTOM INSTRUCTIONSDefines the objectives and methods of supporting AI.Ask me:3. What is your main goal in working with AI? (e.g. productivity, personal evolution, learning, creativity...)4. How would you like AI to help you in your day-to-day life? (e.g. organization, writing, decisions, mental clarity...)5. Do you want AI to close every task on a final note for your personal growth?Yes, alwaysYes, but I would like to customize itNo, I prefer withoutIf I answer "Yes," ask me:6. What do you want the "Ending Note" to include?Strengths to be exploitedRecurring patterns or bias to monitorReflection on Bodyset, Heartset, Mindset, SoulsetExpert Mentor AdviceA micro-practice (breathing, journaling, visualization...)7. What tone do you want for the final note?Direct and schematicNarrative and inspirationalIronic but deepMix of theseAt the end, generate a complete text for Custom Instructions, including the "End Note" or basic structure.SECTION 3 — ABOUT YOUDescribes professional identity ethically and anonymously.Ask me:8. Do you want to use a real or symbolic name for this profile? (optional)9. What do you want me to call you (name, nickname, formal/informal tone)?10. What is your current professional role or the area in which you operate?11. What other roles or identities matter to you (e.g. parent, artist, sportsman...)?12. Tell me in a concise — and unidentifiable — way about your professional path or red thread.13. What do you feel particularly good at or unique at, even if you don't say it often?14. What strategic goals do you want to achieve in the next 6–12 months?In the end, generate cohesive, inspiring, and data-free text for About You.SECTION 4 — NICKNAME AND OCCUPATIONDefines synthetic and symbolic identity.Ask me:15. What simple and recognizable nickname do you want to use?16. In one line, how would you define your main occupation or mission?(e.g. "Strategist who integrates technological efficiency and human growth").Report data in a short, clear, and ready-to-copy format.Always use neutral wording, not traceable to the names of real companies or individuals.SECTION 5 — MORE ABOUT YOUDelves into cognitive style and operational preferences.Ask me if I've already done one of these tests: NOVA, DISC, MBTI, EVOINT, Enneagram, StrengthsFinder.If yes, only ask for qualitative descriptions or general trends, never specific scores or results.If not, explore with these questions:17. Of these styles, which one represents you the most?Dominant (results, decision, impact)Influential (creativity, energy, communication)Stable (consistency, support, reliability)Conformist (precision, structure, analysis)18. How do you prefer to work?IndependentlyIn teamsWith clear rulesIn freedomIn dynamic contextsIn orderly contexts19. Three words that describe you in the work: __ / __ / __20. How do you make decisions?From the gutWith the dataAfter comparisonBased on values21. Do you know Bodyset, Heartset, Mindset, Soulset?Yes: explain which one you feel is more yoursNo: ask for a short, pragmatic explanationIn the end, write concise text for More About You, based on qualitative insights and without numerical references or personal data.FINAL OUTPUTWhen all sections are complete, generate a summary with the following parts:Chat PersonalityCustom InstructionsAbout YouNickname & OccupationMore About YouAll texts must be anonymous, free of sensitive data and consistent with the chosen tone and depth.MULTI-LLM FITAt the end ask:"Do you want me to produce the final output in the sectional format, compatible with ChatGPT Custom Instructions, or in a single cohesive text, ready for other AI systems such as Gemini, Copilot, Perplexity or other LLMs?"Then generate accordingly.
Hello!
I want to create a custom profile to use AI in a way that is consistent with who I am, what I do, and how I want to evolve.
You will be a strategic mentor: direct, concrete and capable of translating my identity into tangible actions and results.
Guide me one question at a time, and when you're done, automatically generate the final texts ready to be copied into the customization section (or equivalent on other LLMs).
The tone must be professional, motivating and succinct, with the clarity of someone who knows where he is going.
# DISCLAIMER PRIVACY
All the answers I will provide are only used to generate a personal reflection text, they do not constitute identifying data nor will they be archived.
You can consider any of my replies temporary and anonymous.
I can use symbolic names, generic roles, or descriptions that can't be traced back to people, companies, or sensitive data.
When in doubt, always ask me if I want to answer or skip a question.
# SECTION 1 — CHAT PERSONALITY
Defines the tone and behavior of AI.
Ask me:
1. How do you want AI to behave with you? (e.g. Default, Cynic, Robot, Listener, Nerd...)
2. What characteristics do you prefer in its communication style? (e.g. calm, technical, ironic, maieutic...)
At the end, write a text for the Chat Personality section, consistent with the chosen tone and free of personal references.
# SECTION 2 — CUSTOM INSTRUCTIONS
Defines the objectives and methods of supporting AI.
Ask me:
3. What is your main goal in working with AI? (e.g. productivity, personal evolution, learning, creativity...)
4. How would you like AI to help you in your day-to-day life? (e.g. organization, writing, decisions, mental clarity...)
5. Do you want AI to close every task on a final note for your personal growth?
- Yes, always
- Yes, but I would like to customize it
- No, I prefer without
If I answer "Yes," ask me:
6. What do you want the "Ending Note" to include?
- Strengths to be exploited
- Recurring patterns or bias to monitor
- Reflection on Bodyset, Heartset, Mindset, Soulset
- Expert Mentor Advice
- A micro-practice (breathing, journaling, visualization...)
7. What tone do you want for the final note?
- Direct and schematic
- Narrative and inspirational
- Ironic but deep
- Mix of these
At the end, generate a complete text for Custom Instructions, including the "End Note" or basic structure.
# SECTION 3 — ABOUT YOU
Describes professional identity ethically and anonymously.
Ask me:
8. Do you want to use a real or symbolic name for this profile? (optional)
9. What do you want me to call you (name, nickname, formal/informal tone)?
10. What is your current professional role or the area in which you operate?
11. What other roles or identities matter to you (e.g. parent, artist, sportsman...)?
12. Tell me in a concise — and unidentifiable — way about your professional path or red thread.
13. What do you feel particularly good at or unique at, even if you don't say it often?
14. What strategic goals do you want to achieve in the next 6–12 months?
In the end, generate cohesive, inspiring, and data-free text for About You.
# SECTION 4 — NICKNAME AND OCCUPATION
Defines synthetic and symbolic identity.
Ask me:
15. What simple and recognizable nickname do you want to use?
16. In one line, how would you define your main occupation or mission?
(e.g. "Strategist who integrates technological efficiency and human growth").
Report data in a short, clear, and ready-to-copy format.
Always use neutral wording, not traceable to the names of real companies or individuals.
# SECTION 5 — MORE ABOUT YOU
Delves into cognitive style and operational preferences.
Ask me if I've already done one of these tests: NOVA, DISC, MBTI, EVOINT, Enneagram, StrengthsFinder.
If yes, only ask for qualitative descriptions or general trends, never specific scores or results.
If not, explore with these questions:
17. Of these styles, which one represents you the most?
- Dominant (results, decision, impact)
- Influential (creativity, energy, communication)
- Stable (consistency, support, reliability)
- Conformist (precision, structure, analysis)
18. How do you prefer to work?
- Independently
- In teams
- With clear rules
- In freedom
- In dynamic contexts
- In orderly contexts
19. Three words that describe you in the work: __ / __ / __
20. How do you make decisions?
- From the gut
- With the data
- After comparison
- Based on values
21. Do you know Bodyset, Heartset, Mindset, Soulset?
- Yes: explain which one you feel is more yours
- No: ask for a short, pragmatic explanation
In the end, write concise text for More About You, based on qualitative insights and without numerical references or personal data.
# FINAL OUTPUT
When all sections are complete, generate a summary with the following parts:
- Chat Personality
- Custom Instructions
- About You
- Nickname & Occupation
- More About You
All texts must be anonymous, free of sensitive data and consistent with the chosen tone and depth.
# MULTI-LLM FIT
At the end ask:
"Do you want me to produce the final output in the sectional format, compatible with ChatGPT Custom Instructions, or in a single cohesive text, ready for other AI systems such as Gemini, Copilot, Perplexity or other LLMs?"
Then generate accordingly.
WorkshopStrumenti, prompt e risorse
Gli strumenti essenziali, i prompt e i riferimenti del workshop, organizzati in cinque sessioni operative e una sezione normativa.
IMPORTANTE: ogni nuovo strumento deve essere verificato e approvato da MSX Information Security prima dell’uso aziendale. Non inserire dati o contenuti MSX in tool non autorizzati.
Human-Tech Synergy: from AI to Agentic Work
Esempi e fonti usati per mostrare il passaggio dalla chat ai sistemi agentici.
Microsoft 365 Copilot
StrumentoIl riferimento operativo del workshop per chat, dati aziendali e agenti nell’ecosistema Microsoft.
Vai alla risorsa ↗
Abacus.AI · DeepAgent
DemoAmbiente agentico usato nella demo per lavorare sui file e creare prototipi, dashboard e applicazioni. Utilizzare soltanto dati autorizzati.
Vai alla risorsa ↗
Tavus · Ask the Avatar
DemoLa piattaforma di video conversazionale usata per l’avatar interattivo integrato nelle slide del workshop.
Vai alla risorsa ↗Cloudflare OS
DemoDemo ufficiale citata per rendere visibile la direzione dei sistemi collaborativi e agentici.
Vai alla risorsa ↗
McKinsey · The State of AI 2026
Fonte ufficialeRicerca globale citata per mostrare il divario tra scaling dell’AI e valore aziendale significativo.
Vai alla risorsa ↗Personal AI Productivity Lab
Gli agenti specialistici usati nel laboratorio per ricerca approfondita e analisi dei file.
Researcher
StrumentoAgente Microsoft per ricerche approfondite che collegano contenuti di lavoro e fonti web.
Vai alla risorsa ↗
Analyst
StrumentoAgente Microsoft per analizzare file, calcolare metriche e produrre tabelle, grafici e insight.
Vai alla risorsa ↗
monday.com for Microsoft 365 Copilot
ConnettoreCollega board, attività e stato dei progetti a Microsoft 365 Copilot. Può anche aggiornare il lavoro quando l’agente e le azioni sono autorizzati. Richiede approvazione aziendale e rispetta i permessi monday.com esistenti.
Vai alla risorsa ↗Prompt pronti all’uso
Before proceeding with the task, ask me the top 10 clarifying questions to get all the information you need to give me the best possible output: unique in the world and suitable for my noble purposes.Proceed with one question at a time and each time give me a minimum of 2 and a maximum of 4 possible numbered answer options in order to make it easier for me to answer you, thank you.
Before proceeding with the task, ask me the top 10 clarifying questions to get all the information you need to give me the best possible output: unique in the world and suitable for my noble purposes.
Proceed with one question at a time and each time give me a minimum of 2 and a maximum of 4 possible numbered answer options in order to make it easier for me to answer you, thank you.Act as a prompt optimizer for Copilot. I will provide you with a prompt to optimize and the principles for creating effective and relevant prompts. Your answer should include only the optimized prompt.Here are the principles to follow for optimization:1. Clearly define a role for AI, to contextualize the request specifically.2. Specify the desired outcome, clearly stating the end goal.3. Include relevant details and context, providing all the information you need for a complete response.4. Offer clear, step-by-step instructions, avoiding ambiguity or vague requests.5. Structure the required result, indicating the desired format (e.g. bulleted lists, thematic sections, etc.).6. Anticipate questions seeking clarification by inviting the AI to ask three questions to fill in any gaps.7. Maintain a limit of length or complexity, ensuring actionable and focused responses.Here's an example of optimization:Trivial prompt:"Create a business plan for a startup."Optimized prompt:"Act as an experienced business planning consultant with at least 10 years of experience. I want you to develop a business plan for a tech startup that offers SaaS services for human resource management. The plan should include the following elements:Company Description: An overview of the startup's mission, vision, and values.Market analysis: Identify your target audience, specific industry, and key competitors, including a SWOT analysis.Marketing strategy: Propose detailed customer acquisition and retention strategies for the first 6 months.Operating budget: Provide a projection of monthly costs for the first year, considering a low, medium, and high revenue scenario.Milestone timeline: List the milestones for the first year of activity.Before you begin, ask three questions to better understand my goals and the context of the company."Now, the prompt to optimize is: [insert prompt to optimize]
Act as a prompt optimizer for Copilot. I will provide you with a prompt to optimize and the principles for creating effective and relevant prompts. Your answer should include only the optimized prompt.
Here are the principles to follow for optimization:
1. Clearly define a role for AI, to contextualize the request specifically.
2. Specify the desired outcome, clearly stating the end goal.
3. Include relevant details and context, providing all the information you need for a complete response.
4. Offer clear, step-by-step instructions, avoiding ambiguity or vague requests.
5. Structure the required result, indicating the desired format (e.g. bulleted lists, thematic sections, etc.).
6. Anticipate questions seeking clarification by inviting the AI to ask three questions to fill in any gaps.
7. Maintain a limit of length or complexity, ensuring actionable and focused responses.
Here's an example of optimization:
Trivial prompt:
"Create a business plan for a startup."
Optimized prompt:
"Act as an experienced business planning consultant with at least 10 years of experience. I want you to develop a business plan for a tech startup that offers SaaS services for human resource management. The plan should include the following elements:
- Company Description: An overview of the startup's mission, vision, and values.
- Market analysis: Identify your target audience, specific industry, and key competitors, including a SWOT analysis.
- Marketing strategy: Propose detailed customer acquisition and retention strategies for the first 6 months.
- Operating budget: Provide a projection of monthly costs for the first year, considering a low, medium, and high revenue scenario.
- Milestone timeline: List the milestones for the first year of activity.
Before you begin, ask three questions to better understand my goals and the context of the company."
Now, the prompt to optimize is: [insert prompt to optimize]ROLEAct as my briefing assistant in Microsoft 365 Copilot. Start from one concrete reference, reconstruct the connected work context and prepare an updated, verifiable briefing.REQUIRED FIRST STEPDo not start searching immediately. First ask me this question and wait for my answer:"Where would you like to start?an Outlook email or thread: give me the subject, sender or date;a project, person or upcoming call: give me the name;a SharePoint or OneDrive file: select or attach it.Give me the most precise reference you have."After I answer, summarise what you identified in one sentence. Ask for confirmation only if the reference is ambiguous.SOURCES TO USEStarting from the selected reference, connect the relevant Microsoft 365 work content:relevant Outlook emails and threads;calendar invitations and related attachments;connected Teams meeting transcripts, notes and chats;relevant SharePoint or OneDrive project files.If you cannot retrieve a source directly, tell me exactly which email or file I need to open, select or attach. Do not fill gaps with assumptions.OUTPUTPrepare a concise professional briefing with:scope reviewed and sources found;context and recent history;decisions already made;open commitments with owners and deadlines;project status, risks and possible contradictions;priorities for the next conversation or decision;useful questions and next steps.RULESFor every important fact, provide the source, date and link when available.Clearly separate facts, inferences and missing information.Exclude unrelated content.Do not send messages, modify files or create calendar events.
# ROLE
Act as my briefing assistant in Microsoft 365 Copilot. Start from one concrete reference, reconstruct the connected work context and prepare an updated, verifiable briefing.
# REQUIRED FIRST STEP
Do not start searching immediately. First ask me this question and wait for my answer:
"Where would you like to start?
- an Outlook email or thread: give me the subject, sender or date;
- a project, person or upcoming call: give me the name;
- a SharePoint or OneDrive file: select or attach it.
Give me the most precise reference you have."
After I answer, summarise what you identified in one sentence. Ask for confirmation only if the reference is ambiguous.
# SOURCES TO USE
Starting from the selected reference, connect the relevant Microsoft 365 work content:
- relevant Outlook emails and threads;
- calendar invitations and related attachments;
- connected Teams meeting transcripts, notes and chats;
- relevant SharePoint or OneDrive project files.
If you cannot retrieve a source directly, tell me exactly which email or file I need to open, select or attach. Do not fill gaps with assumptions.
# OUTPUT
Prepare a concise professional briefing with:
- scope reviewed and sources found;
- context and recent history;
- decisions already made;
- open commitments with owners and deadlines;
- project status, risks and possible contradictions;
- priorities for the next conversation or decision;
- useful questions and next steps.
# RULES
- For every important fact, provide the source, date and link when available.
- Clearly separate facts, inferences and missing information.
- Exclude unrelated content.
- Do not send messages, modify files or create calendar events.OBIETTIVOAiutami a ricercare un argomento unendo la conoscenza aziendale con una ricerca web approfondita e aggiornata.BREVE INTERVISTAFammi due domande:1. Che cosa voglio comprendere, decidere o produrre?2. Quali file o riferimenti aziendali devo includere e quale mercato, periodo o tema deve coprire la ricerca esterna?Fai una terza domanda solo se indispensabile per definire il perimetro.RICERCACerca nei contenuti pertinenti disponibili in Outlook, Teams, SharePoint, OneDrive e nelle fonti che allego. Poi approfondisci online privilegiando fonti istituzionali, ricerca primaria, documentazione ufficiale e pubblicazioni recenti e autorevoli.Verifica le affermazioni importanti con più fonti quando possibile. Distingui sempre:ciò che emerge dalle informazioni aziendali;ciò che viene confermato o contraddetto online;le tue inferenze;le informazioni ancora mancanti.RISULTATOScegli liberamente struttura e livello di dettaglio più adatti al mio obiettivo e al pubblico. Metti in evidenza risultati, implicazioni, rischi, opportunità e raccomandazioni. Usa tabelle, timeline o grafici solo se migliorano la comprensione.Cita nome, data e link delle fonti. Concludi con le principali lacune informative e i prossimi passi consigliati. Non inventare informazioni.
# OBIETTIVO
Aiutami a ricercare un argomento unendo la conoscenza aziendale con una ricerca web approfondita e aggiornata.
# BREVE INTERVISTA
Fammi due domande:
1. Che cosa voglio comprendere, decidere o produrre?
2. Quali file o riferimenti aziendali devo includere e quale mercato, periodo o tema deve coprire la ricerca esterna?
Fai una terza domanda solo se indispensabile per definire il perimetro.
# RICERCA
Cerca nei contenuti pertinenti disponibili in Outlook, Teams, SharePoint, OneDrive e nelle fonti che allego. Poi approfondisci online privilegiando fonti istituzionali, ricerca primaria, documentazione ufficiale e pubblicazioni recenti e autorevoli.
Verifica le affermazioni importanti con più fonti quando possibile. Distingui sempre:
- ciò che emerge dalle informazioni aziendali;
- ciò che viene confermato o contraddetto online;
- le tue inferenze;
- le informazioni ancora mancanti.
# RISULTATO
Scegli liberamente struttura e livello di dettaglio più adatti al mio obiettivo e al pubblico. Metti in evidenza risultati, implicazioni, rischi, opportunità e raccomandazioni. Usa tabelle, timeline o grafici solo se migliorano la comprensione.
Cita nome, data e link delle fonti. Concludi con le principali lacune informative e i prossimi passi consigliati. Non inventare informazioni.ROLEAct as my data analysis partner in Microsoft Copilot Analyst. Help me understand the data, choose the right analysis, and produce a result that supports the decision.GET STARTEDAsk me:"Attach or select the files you want to analyse. Then tell me in one sentence what you want to understand, compare, improve, or decide. If you are unsure, write: 'Suggest the analysis.'"Wait until the files are available.If the objective is unclear, review the files and suggest up to three business questions that the data could help answer.Ask no more than two additional questions, one at a time and only when essential. Do not ask me to define technical metrics when you can infer them from the objective and the data.Before starting, summarise the question, data, and proposed analysis in no more than three lines. Ask for confirmation and proceed.CHECK AND ANALYSEIdentify the relevant files, sheets, periods, units, and fields.Flag only material issues: missing data, duplicates, inconsistent definitions, or incompatible datasets.Combine only compatible data.Select the metrics and comparisons that best support the objective.Identify relevant trends, anomalies, exceptions, and differences.Explain important transformations and assumptions in plain language.Connect every important finding to the relevant file, sheet, or field.Do not present correlation as causation without supporting evidence.FINAL RESULTChoose the structure, format, and level of detail that best fit the objective, audience, available data, and decision to support.Do not use a predetermined template. Decide whether the result needs a management summary, detailed report, tables, charts, anomaly review, recommendations, or a combination of these elements.Put the most decision-relevant evidence first. Include limitations, recommendations, next steps, and additional data needs only when relevant.RULESClearly distinguish data, interpretation, and recommendation.Do not invent missing values or facts.State when the evidence is insufficient for a reliable conclusion.
# ROLE
Act as my data analysis partner in Microsoft Copilot Analyst. Help me understand the data, choose the right analysis, and produce a result that supports the decision.
# GET STARTED
Ask me:
"Attach or select the files you want to analyse. Then tell me in one sentence what you want to understand, compare, improve, or decide. If you are unsure, write: 'Suggest the analysis.'"
Wait until the files are available.
If the objective is unclear, review the files and suggest up to three business questions that the data could help answer.
Ask no more than two additional questions, one at a time and only when essential. Do not ask me to define technical metrics when you can infer them from the objective and the data.
Before starting, summarise the question, data, and proposed analysis in no more than three lines. Ask for confirmation and proceed.
# CHECK AND ANALYSE
- Identify the relevant files, sheets, periods, units, and fields.
- Flag only material issues: missing data, duplicates, inconsistent definitions, or incompatible datasets.
- Combine only compatible data.
- Select the metrics and comparisons that best support the objective.
- Identify relevant trends, anomalies, exceptions, and differences.
- Explain important transformations and assumptions in plain language.
- Connect every important finding to the relevant file, sheet, or field.
- Do not present correlation as causation without supporting evidence.
# FINAL RESULT
Choose the structure, format, and level of detail that best fit the objective, audience, available data, and decision to support.
Do not use a predetermined template. Decide whether the result needs a management summary, detailed report, tables, charts, anomaly review, recommendations, or a combination of these elements.
Put the most decision-relevant evidence first. Include limitations, recommendations, next steps, and additional data needs only when relevant.
# RULES
- Clearly distinguish data, interpretation, and recommendation.
- Do not invent missing values or facts.
- State when the evidence is insufficient for a reliable conclusion.Personal Agent Challenge
Da un lavoro ricorrente al primo agente personale costruito e testato con Copilot.
Prompt pronti all’uso
Act as an expert in organizational workflow design.Your job is to help me map out one of my real-world workflows.IMPORTANT: The example below does NOT represent my role. It only serves as an example of the STRUCTURE and FORMAT I want to achieve. Copy the SHAPE of the workflow, not the CONTENT.The final workflow must be built on the basis of:my role;my recurring activities;my real tasks.Example workflow:WORKFLOW 1: Design and delivery of training sessionsKey activity 1.1 Analysis of training needsKey activity 1.2 Content and activity designKey activity 1.3 Logistical and technical preparationKey activity 1.4 Session deliveryKey activity 1.5 Feedback collection and follow-upNow help me create my custom workflow.Before proceeding with the task, ask me the top 10 clarifying questions necessary to collect all the information I need to build the best possible workflow, really suitable for my role and my professional goals.Proceed with ONE question at a time.For each question:give me minimum 2 and maximum 4 numbered answer options, so that I can get back to you quickly;also leave me the option to write a personalized response if no option is perfect.The questions should help you understand:my role;my recurring activities;my main workflows;high-value activities;repetitive or cognitively demanding tasks;activities that could be supported by AI.
Act as an expert in organizational workflow design.
Your job is to help me map out one of my real-world workflows.
IMPORTANT: The example below does NOT represent my role. It only serves as an example of the STRUCTURE and FORMAT I want to achieve. Copy the SHAPE of the workflow, not the CONTENT.
The final workflow must be built on the basis of:
- my role;
- my recurring activities;
- my real tasks.
Example workflow:
WORKFLOW 1: Design and delivery of training sessions
Key activity 1.1 Analysis of training needs
Key activity 1.2 Content and activity design
Key activity 1.3 Logistical and technical preparation
Key activity 1.4 Session delivery
Key activity 1.5 Feedback collection and follow-up
Now help me create my custom workflow.
Before proceeding with the task, ask me the top 10 clarifying questions necessary to collect all the information I need to build the best possible workflow, really suitable for my role and my professional goals.
Proceed with ONE question at a time.
For each question:
- give me minimum 2 and maximum 4 numbered answer options, so that I can get back to you quickly;
- also leave me the option to write a personalized response if no option is perfect.
The questions should help you understand:
- my role;
- my recurring activities;
- my main workflows;
- high-value activities;
- repetitive or cognitively demanding tasks;
- activities that could be supported by AI.Great.Now that you've helped me create my workflow in the required format, let's continue the process of creating my custom advanced prompts/digital assistants.Now I want to show you an example of an advanced prompt.IMPORTANT:This example does NOT represent my specific role or job. It only serves as an example of the prompting STRUCTURE, LOGIC, and STYLE that we will use next.Think of this prompt as a reusable template.So:don't focus on the topic of the prompt;do not copy the content, instead, look at the invisible skeleton of the prompt;how it is built;how AI leads;how it organizes the flow of conversation;how it collects information;how it produces high-quality output.Your job for now is NOT to apply the prompt to my workflow.Your only task is to carefully study:the structure;the sections;the logic;the level of detail;how the prompt orchestrates the collaboration between user and AI.Below is the sample prompt for you to study:"You act as a recruiting expert with 20 years of experience: 10 years in consulting and 10 years in operations. In the last 5 years, you have specialized in the creation of job descriptions for strategic roles in large Italian companies.Your job is to support me in creating a job description for a new role in my company.Task 1: Initial interview to gather information. To ensure maximum accuracy, ask a series of targeted questions to gather all the necessary details about the role. Ask one question at a time and wait for a response before proceeding. If necessary, ask for clarification to dig deeper. Here are the questions:What is the official title of the role and to which department does it belong?What are the main responsibilities and daily activities of the role?What technical and soft skills are needed to excel in this role? Are there any preferred qualifications or additional experience desired?What performance indicators (KPIs) will be used to assess success in the role?Can you describe the type of work environment and company culture?Does the role require the use of specific tools, technologies or software?Are there any special requirements, such as availability to travel or flexibility in schedules?Task 2: Creating the job description. Using the information you gathered, write a detailed job description that includes:Title of the role and department.Role objectives: A brief description of the main purpose of the role.Key responsibilities: A list of daily tasks and strategic goals.Skills required: technical, transversal and preferred.Additional requirements: specific requirements such as travel, flexible schedules or technical knowledge.Description of the work environment: information about the company culture.Make sure the job description is clear, engaging, and adheres to industry best practices. The final structure should be well-organized, with clearly defined sections and professional language.Iterative FeedbackIf the job description doesn't fully meet expectations, propose changes or rephrasing based on the feedback received. For example, suggest improvements to the language or details to make it more targeted to the target audience.Final outputProvide a ready-to-use job description, which the HR manager can publish directly on recruiting platforms or share internally. Also attach a checklist summarizing the steps followed to create the job description."When you're done studying it, just say, "Structure learned."
Great.
Now that you've helped me create my workflow in the required format, let's continue the process of creating my custom advanced prompts/digital assistants.
Now I want to show you an example of an advanced prompt.
IMPORTANT:
This example does NOT represent my specific role or job. It only serves as an example of the prompting STRUCTURE, LOGIC, and STYLE that we will use next.
Think of this prompt as a reusable template.
So:
- don't focus on the topic of the prompt;
- do not copy the content, instead, look at the invisible skeleton of the prompt;
- how it is built;
- how AI leads;
- how it organizes the flow of conversation;
- how it collects information;
- how it produces high-quality output.
Your job for now is NOT to apply the prompt to my workflow.
Your only task is to carefully study:
- the structure;
- the sections;
- the logic;
- the level of detail;
- how the prompt orchestrates the collaboration between user and AI.
Below is the sample prompt for you to study:
"You act as a recruiting expert with 20 years of experience: 10 years in consulting and 10 years in operations. In the last 5 years, you have specialized in the creation of job descriptions for strategic roles in large Italian companies.
Your job is to support me in creating a job description for a new role in my company.
Task 1: Initial interview to gather information. To ensure maximum accuracy, ask a series of targeted questions to gather all the necessary details about the role. Ask one question at a time and wait for a response before proceeding. If necessary, ask for clarification to dig deeper. Here are the questions:
What is the official title of the role and to which department does it belong?
What are the main responsibilities and daily activities of the role?
What technical and soft skills are needed to excel in this role? Are there any preferred qualifications or additional experience desired?
What performance indicators (KPIs) will be used to assess success in the role?
Can you describe the type of work environment and company culture?
Does the role require the use of specific tools, technologies or software?
Are there any special requirements, such as availability to travel or flexibility in schedules?
Task 2: Creating the job description. Using the information you gathered, write a detailed job description that includes:
- Title of the role and department.
- Role objectives: A brief description of the main purpose of the role.
- Key responsibilities: A list of daily tasks and strategic goals.
- Skills required: technical, transversal and preferred.
- Additional requirements: specific requirements such as travel, flexible schedules or technical knowledge.
- Description of the work environment: information about the company culture.
Make sure the job description is clear, engaging, and adheres to industry best practices. The final structure should be well-organized, with clearly defined sections and professional language.
Iterative Feedback
If the job description doesn't fully meet expectations, propose changes or rephrasing based on the feedback received. For example, suggest improvements to the language or details to make it more targeted to the target audience.
Final output
Provide a ready-to-use job description, which the HR manager can publish directly on recruiting platforms or share internally. Also attach a checklist summarizing the steps followed to create the job description."
When you're done studying it, just say, "Structure learned."Perfect. Now apply the SAME STRUCTURE as the studied prompt to my real workflow.IMPORTANT: Reuse the same structure and logic as the original prompt, but replace the content with my actual task.Start only with activity 1.1.Before creating the advanced prompt, ask me the top 10 clarifying questions necessary to understand this specific activity in depth and create the best possible advanced prompt, really useful for my professional goals.Proceed with ONE question at a time.For each question:give me minimum 2 and maximum 4 numbered answer options, so that I can get back to you quickly;also leave me the possibility to write a personalized answer.The questions should help you understand:the context of the activity;the desired result;the expected level of quality;any constraints;the tone;the success criteria;the mistakes to avoid;the most useful type of AI support.Once you've gathered enough information, create the full advanced prompt for activity 1.1 and then wait for my validation before moving on to 1.2.
Perfect. Now apply the SAME STRUCTURE as the studied prompt to my real workflow.
IMPORTANT: Reuse the same structure and logic as the original prompt, but replace the content with my actual task.
Start only with activity 1.1.
Before creating the advanced prompt, ask me the top 10 clarifying questions necessary to understand this specific activity in depth and create the best possible advanced prompt, really useful for my professional goals.
Proceed with ONE question at a time.
For each question:
- give me minimum 2 and maximum 4 numbered answer options, so that I can get back to you quickly;
- also leave me the possibility to write a personalized answer.
The questions should help you understand:
- the context of the activity;
- the desired result;
- the expected level of quality;
- any constraints;
- the tone;
- the success criteria;
- the mistakes to avoid;
- the most useful type of AI support.
Once you've gathered enough information, create the full advanced prompt for activity 1.1 and then wait for my validation before moving on to 1.2.ROLEAct as my workflow interviewer and help me design a personal agent for Microsoft 365 Copilot Agent Builder.Your final result must be one complete prompt that I can copy and paste directly into Agent Builder.INTERVIEWAsk me one short question at a time. Ask no more than five questions and skip anything I have already explained.Start with:"What recurring result would you like Copilot to produce for you? Describe one real example of when you need it."Then understand:how I currently perform the work;which steps, decisions, and recurring difficulties are involved;which Microsoft 365 sources the agent should use, such as SharePoint, OneDrive, Outlook emails, Teams meetings, transcripts, chats, or calendar information;what the final result should contain and who will use it;which exceptions, sensitive situations, or quality problems require human review.Do not ask me to define technical configurations. Translate my answers into the appropriate Agent Builder instructions.Only design functions that can be performed inside Microsoft 365 Copilot Agent Builder. If the task requires scheduled work, automatic actions, or unsupported external integrations, explain the limitation in one sentence before continuing.FINAL OUTPUTAfter the interview, produce only one complete prompt ready to copy into Agent Builder.The prompt must clearly describe:the proposed agent name;its purpose and recurring outcome;the workflow it should follow;the Microsoft 365 knowledge sources it should consult;how it should handle missing or conflicting information;the expected format, tone, and level of detail;when it must stop and ask the user;how the user should review the result.Write the final prompt as direct instructions to Agent Builder.Do not add a separate explanation, checklist, test plan, or commentary outside the prompt.
# ROLE
Act as my workflow interviewer and help me design a personal agent for Microsoft 365 Copilot Agent Builder.
Your final result must be one complete prompt that I can copy and paste directly into Agent Builder.
# INTERVIEW
Ask me one short question at a time. Ask no more than five questions and skip anything I have already explained.
Start with:
"What recurring result would you like Copilot to produce for you? Describe one real example of when you need it."
Then understand:
- how I currently perform the work;
- which steps, decisions, and recurring difficulties are involved;
- which Microsoft 365 sources the agent should use, such as SharePoint, OneDrive, Outlook emails, Teams meetings, transcripts, chats, or calendar information;
- what the final result should contain and who will use it;
- which exceptions, sensitive situations, or quality problems require human review.
Do not ask me to define technical configurations. Translate my answers into the appropriate Agent Builder instructions.
Only design functions that can be performed inside Microsoft 365 Copilot Agent Builder. If the task requires scheduled work, automatic actions, or unsupported external integrations, explain the limitation in one sentence before continuing.
# FINAL OUTPUT
After the interview, produce only one complete prompt ready to copy into Agent Builder.
The prompt must clearly describe:
- the proposed agent name;
- its purpose and recurring outcome;
- the workflow it should follow;
- the Microsoft 365 knowledge sources it should consult;
- how it should handle missing or conflicting information;
- the expected format, tone, and level of detail;
- when it must stop and ask the user;
- how the user should review the result.
Write the final prompt as direct instructions to Agent Builder.
Do not add a separate explanation, checklist, test plan, or commentary outside the prompt.From Personal AI to Corporate AI
Risorse per osservare il lavoro, trascriverlo, trasformarlo in conoscenza operativa e definire un’opportunità misurabile.
Valutazione dell’opportunità
Cattura e trascrivi il lavoro
Scribe
StrumentoCattura un flusso svolto sullo schermo e lo trasforma in una guida operativa passo-passo.
Vai alla risorsa ↗
Word Transcribe
StrumentoImporta una registrazione e la trasforma in testo modificabile con speaker e timestamp.
Vai alla risorsa ↗
Parakeet Transcription
StrumentoApplicazione Windows del Microsoft Store per trascrivere file audio direttamente sul PC con NVIDIA Parakeet.
Vai alla risorsa ↗Struttura la conoscenza
LangExtract
RepositoryRepository open source per estrarre da testo grezzo decisioni, passaggi, eccezioni e criteri, mantenendo il collegamento ai passaggi della fonte.
Vai alla risorsa ↗
Docling Graph
RepositoryPipeline open source per trasformare documenti eterogenei in oggetti e knowledge graph strutturati, validati e collegati alla fonte.
Vai alla risorsa ↗Pronto all’uso
ROLEKnowledge extraction agent performing a session post-mortem.TASKConvert the attached transcript into a durable knowledge-base document.This is an EXTRACTIVE task, not an abstractive summary. Do not narrate the conversation.Target reader: a domain expert who did not observe the session and must resume the work cold.OUTPUTUse Markdown and this fixed order:1. Session card — actual objective, outcome, stack/versions. Max 3 lines.2. Verified know-how — per entry: trigger condition, exact steps/commands/params, causal explanation.3. Failure modes — attempt → root cause → early detection signal → correct alternative.4. Decisions & tradeoffs — choice, rejected options, decision criterion, falsifier (what would reverse it).5. Environment gotchas — versions, paths, undocumented limits, cryptic errors decoded.6. Heuristics — conditional rules: "IF <condition> THEN <action> BECAUSE <mechanism>". State scope and confidence.7. Operator preferences & model corrections — instructions the operator had to repeat, systematic model errors. Tag each as a candidate for: system prompt / persistent memory / project rules.8. Open threads — executable next actions only ("run Y on Z to verify W"), never "investigate X".9. Context primer — a 200-word self-contained handoff block to paste into a fresh session: state, constraints, closed decisions, next step.CONSTRAINTSGround every claim in the source document. Cite provenance: the utterance, command, or error that supports it.Do not inject prior knowledge, generic best practices, or plausible-sounding specifics. If a detail is absent, write "not specified" and abstain.Tag epistemic status per entry: [VERIFIED] executed with observed outcome | [ASSERTED] claimed, untested | [INFERRED] your inference.Entries must be atomic and self-contained — each one is retrieved in isolation. No anaphora, no "as noted above".Reject low-specificity entries. Recall > eloquence: prefer exact identifiers, numbers, literal error strings.Transcripts may be noisy (ASR errors, interruptions, filler). Normalize wording, never normalize facts: if a term is garbled, keep it verbatim and flag it as uncertain.If the source self-contradicts, the later state wins; flag the contradiction and its resolution as a decision.Empty section → "— none". No preamble, no closing offer. Start at section 1.Before emitting: "what would the operator have wanted to know at the start that only surfaced at the end?" If absent, add it.SOURCE[paste transcript here, or reference the attached file]
# ROLE
Knowledge extraction agent performing a session post-mortem.
# TASK
Convert the attached transcript into a durable knowledge-base document.
This is an EXTRACTIVE task, not an abstractive summary. Do not narrate the conversation.
Target reader: a domain expert who did not observe the session and must resume the work cold.
# OUTPUT
Use Markdown and this fixed order:
1. Session card — actual objective, outcome, stack/versions. Max 3 lines.
2. Verified know-how — per entry: trigger condition, exact steps/commands/params, causal explanation.
3. Failure modes — attempt → root cause → early detection signal → correct alternative.
4. Decisions & tradeoffs — choice, rejected options, decision criterion, falsifier (what would reverse it).
5. Environment gotchas — versions, paths, undocumented limits, cryptic errors decoded.
6. Heuristics — conditional rules: "IF <condition> THEN <action> BECAUSE <mechanism>". State scope and confidence.
7. Operator preferences & model corrections — instructions the operator had to repeat, systematic model errors. Tag each as a candidate for: system prompt / persistent memory / project rules.
8. Open threads — executable next actions only ("run Y on Z to verify W"), never "investigate X".
9. Context primer — a 200-word self-contained handoff block to paste into a fresh session: state, constraints, closed decisions, next step.
# CONSTRAINTS
- Ground every claim in the source document. Cite provenance: the utterance, command, or error that supports it.
- Do not inject prior knowledge, generic best practices, or plausible-sounding specifics. If a detail is absent, write "not specified" and abstain.
- Tag epistemic status per entry: [VERIFIED] executed with observed outcome | [ASSERTED] claimed, untested | [INFERRED] your inference.
- Entries must be atomic and self-contained — each one is retrieved in isolation. No anaphora, no "as noted above".
- Reject low-specificity entries. Recall > eloquence: prefer exact identifiers, numbers, literal error strings.
- Transcripts may be noisy (ASR errors, interruptions, filler). Normalize wording, never normalize facts: if a term is garbled, keep it verbatim and flag it as uncertain.
- If the source self-contradicts, the later state wins; flag the contradiction and its resolution as a decision.
- Empty section → "— none". No preamble, no closing offer. Start at section 1.
- Before emitting: "what would the operator have wanted to know at the start that only surfaced at the end?" If absent, add it.
# SOURCE
[paste transcript here, or reference the attached file]ROLEKnowledge extraction agent performing a post-mortem on THIS session.TASKConvert everything we did in this conversation into a durable knowledge-base document.This is an EXTRACTIVE task over your own context window, not an abstractive summary and not a recap for me. Do not narrate what we discussed.Target reader: a domain expert with zero context who must resume this work cold.SCOPEThe entire session from its first message, including work done through tools, files, and code — not only the recent turns.OUTPUTUse Markdown and this fixed order:1. Session card — actual objective (the one we converged on, not the one I opened with), outcome, stack/versions. Max 3 lines.2. Verified know-how — per entry: trigger condition, exact steps/commands/params, causal explanation.3. Failure modes — what we attempted → root cause → early detection signal → correct alternative. Include your own wrong turns and dead ends, not just mine.4. Decisions & tradeoffs — choice, rejected options, decision criterion, falsifier (what would reverse it).5. Environment gotchas — versions, paths, undocumented limits, cryptic errors decoded.6. Heuristics — conditional rules: "IF <condition> THEN <action> BECAUSE <mechanism>". State scope and confidence.7. Operator preferences & model corrections — what I had to restate or correct, where you drifted or misread intent. Tag each as a candidate for: system prompt / persistent memory / project rules.8. Open threads — executable next actions only ("run Y on Z to verify W"), never "investigate X".9. Context primer — a 200-word self-contained handoff block I can paste into a fresh session: state, constraints, closed decisions, next step.CONSTRAINTSGround every claim in what actually happened in this session. Cite provenance: the turn, command, tool call, or error that supports it. Do not reconstruct from what a session like this usually contains.Recency bias is the main failure mode here: the earliest turns carry the framing decisions. Weight the whole session, not the last exchanges.If part of the session was compacted, truncated, or is no longer legible to you, say so explicitly under section 9 rather than filling the gap.Do not inject prior knowledge, generic best practices, or plausible-sounding specifics. If a detail is absent, write "not specified" and abstain.Tag epistemic status per entry: [VERIFIED] executed with observed outcome | [ASSERTED] claimed, untested | [INFERRED] your inference.Entries must be atomic and self-contained — each one is retrieved in isolation. No anaphora, no "as noted above", no references to "this chat".Reject low-specificity entries. Recall > eloquence: prefer exact identifiers, numbers, literal error strings.Where we contradicted ourselves, the later state wins; flag the contradiction and its resolution as a section 4 decision.Empty section → "— none". No preamble, no closing offer, no summary of this instruction. Start at section 1.Before emitting: "what would the operator have wanted to know at the start of this session that only surfaced at the end?" If absent, add it.
# ROLE
Knowledge extraction agent performing a post-mortem on THIS session.
# TASK
Convert everything we did in this conversation into a durable knowledge-base document.
This is an EXTRACTIVE task over your own context window, not an abstractive summary and not a recap for me. Do not narrate what we discussed.
Target reader: a domain expert with zero context who must resume this work cold.
# SCOPE
The entire session from its first message, including work done through tools, files, and code — not only the recent turns.
# OUTPUT
Use Markdown and this fixed order:
1. Session card — actual objective (the one we converged on, not the one I opened with), outcome, stack/versions. Max 3 lines.
2. Verified know-how — per entry: trigger condition, exact steps/commands/params, causal explanation.
3. Failure modes — what we attempted → root cause → early detection signal → correct alternative. Include your own wrong turns and dead ends, not just mine.
4. Decisions & tradeoffs — choice, rejected options, decision criterion, falsifier (what would reverse it).
5. Environment gotchas — versions, paths, undocumented limits, cryptic errors decoded.
6. Heuristics — conditional rules: "IF <condition> THEN <action> BECAUSE <mechanism>". State scope and confidence.
7. Operator preferences & model corrections — what I had to restate or correct, where you drifted or misread intent. Tag each as a candidate for: system prompt / persistent memory / project rules.
8. Open threads — executable next actions only ("run Y on Z to verify W"), never "investigate X".
9. Context primer — a 200-word self-contained handoff block I can paste into a fresh session: state, constraints, closed decisions, next step.
# CONSTRAINTS
- Ground every claim in what actually happened in this session. Cite provenance: the turn, command, tool call, or error that supports it. Do not reconstruct from what a session like this usually contains.
- Recency bias is the main failure mode here: the earliest turns carry the framing decisions. Weight the whole session, not the last exchanges.
- If part of the session was compacted, truncated, or is no longer legible to you, say so explicitly under section 9 rather than filling the gap.
- Do not inject prior knowledge, generic best practices, or plausible-sounding specifics. If a detail is absent, write "not specified" and abstain.
- Tag epistemic status per entry: [VERIFIED] executed with observed outcome | [ASSERTED] claimed, untested | [INFERRED] your inference.
- Entries must be atomic and self-contained — each one is retrieved in isolation. No anaphora, no "as noted above", no references to "this chat".
- Reject low-specificity entries. Recall > eloquence: prefer exact identifiers, numbers, literal error strings.
- Where we contradicted ourselves, the later state wins; flag the contradiction and its resolution as a section 4 decision.
- Empty section → "— none". No preamble, no closing offer, no summary of this instruction. Start at section 1.
- Before emitting: "what would the operator have wanted to know at the start of this session that only surfaced at the end?" If absent, add it.Knowledge graph: quale scegliere
Cognee
RepositoryTrasforma file, trascrizioni, URL e altre fonti in una memoria condivisa basata su grafo e ricerca vettoriale. È la scelta più adatta per costruire una knowledge layer aziendale da materiali eterogenei.
Vai alla risorsa ↗
Graphiti
RepositoryCostruisce un knowledge graph temporale che conserva fatti, relazioni, fonti e cambiamenti nel tempo. È più adatto a decisioni, processi e contesti cliente che evolvono continuamente.
Vai alla risorsa ↗Agentic Governance
Riferimenti per misurare il valore, assegnare responsabilità e ridurre l’esposizione dei dati prima della produzione.
Linee guida
Microsoft · Measure and report value
FrameworkGuida per confrontare baseline e risultati del prototipo attraverso valore, costo, qualità, adozione e rischio.
Vai alla risorsa ↗Microsoft · Roles and responsibilities
FrameworkRiferimento per distinguere responsabilità business, revisione AI, governance e passaggio al team IT.
Vai alla risorsa ↗Strumenti per oscuramento e anonimizzazione
Open Redact
StrumentoStrumento nel browser per oscurare contenuti nei PDF prima di condividerli con servizi esterni.
Vai alla risorsa ↗
OpenAI Privacy Filter
RepositoryProgetto open source per rilevare e trasformare localmente informazioni personali nei documenti.
Vai alla risorsa ↗
Microsoft Presidio
RepositoryFramework configurabile per identificare, anonimizzare e proteggere dati personali in testo e immagini.
Vai alla risorsa ↗Regulations
Guide ufficiali accessibili e strumenti pronti all’uso per un primo controllo su GDPR e AI Act.
AI Act
European Commission · Navigating the AI Act
Fonte ufficialeFAQ ufficiali e accessibili su ambito, categorie di rischio, obblighi, responsabilità e calendario.
Vai alla risorsa ↗EU AI Act · Full text
Fonte ufficialeTesto completo ufficiale del Regolamento (UE) 2024/1689 sull’intelligenza artificiale.
Vai alla risorsa ↗