I have launched many businesses and built multiple teams. If there is one super power that I think helped me in my entrepreneurial journey, it is the subtle art of delegating. As I am now using AI at work and in personal life and building an AI-first product, I see some people simply not getting anywhere with AI. Quality outputs or meaningful interactions seem to escape them. At a closer look though, I discovered that AI success and effective delegation hinge on the same principles. In other words, some people struggle with poor AI results simply because of how they frame tasks, revealing that delegating to AI mirrors delegating to people in its demand for clarity and precision Empirical proof So I saw a curious Reddit post on prompt engineering. The public were actively discussing someone’s tweet on how true ‘AI whisperers’ have very little knowledge of true AI ‘jailbreaking’, as someone put it. Talking to AI, redditors agree, is not prompt engineering. But suppose you’re an SMB owner. Your marketing agency has just found traction, and you’re trying to eliminate the back and forth of client briefing. You open a Perplexity chat and you try to get it to analyze your client’s brief and suggest the best questions to ask in the next meeting to complete the brief. Do you need to do software engineering? No. But you have to literally talk to AI so it understands. Just as I was thinking about it, someone commented: “Good delegation skills will still apply. If you know how to delegate a task well, succinctly and accurately, then a lot of the tricks are not needed.” And I could not agree more! Even if you are an excellent coder and you know how big data works and all, you still very much need to – yeah, talk to AI to get it to do things for you. And if you’re a purpose-driven SMB owner, always on a tight budget, talking to AI is really all you can do. Where do effective conversations with AI begin, then? Delegation 101 Like I said in my other article, to remain AI-first and not AI-dependent, you need to treat it as a subordinate co-worker. That means you need to let AI do work, i.e. to delegate it some tasks. Delegation is distributing work and tasks among the team, as much in line with what they are good at as possible. Now, with AI, which admittedly excels in analysis and research, you don’t have to wonder about what it can do well, and you just need to—well, delegate! The science of delegation psychology explains that leaders may stumble on themselves and their negative self-talk when delegating. What you say in your head, as you talk to that co-worker, or the unconscious belief you harbor when crafting that email, have a visible impact on how successful your delegation is. Personally, these mantras reflect my thoughts at the dawn of my career as a CMO at a large bank branch, when I was given my first team: I trust my team. Their excellence and growth are part of my excellence and growth. This phrase is the opposite of what we all tend to secretly think: ‘Nobody is going to do it better than me’. The belief in one’s superiority undermines the confidence of people you work with. Delegating with a sigh is the beginning of the end. In contrast, delegating with a firm smile, a shoulder tap and some internal positive programming boosts personal responsibility and performance. Done better when delegated is far more valuable than done perfectly by me. That hurts, doesn’t it? Well, the truth is, you could spend hours on getting things done at 100%, but you could also spend hours on doing x2 and getting these things just done. Sadly (or thankfully), perfection is not a prerequisite for effective operations. Not delegating what I can do well means missing out on what only I can do. Mundane work is debilitatingly comfortable. It may simply be scary to distribute work across the team to get a task done because… you may not have a plan yet. Hiding behind ‘I know how to do it’ in the face of a real challenge is akin to betrayal, so to speak. One’s personal productivity is not in how much operational stuff one can crunch in an hour but how quick one is to learn. I trust my team. Their excellence and growth are part of my excellence and growth. This phrase is the opposite of what we all tend to secretly think: ‘Nobody is going to do it better than me’. The belief in one’s superiority undermines the confidence of people you work with. Delegating with a sigh is the beginning of the end. In contrast, delegating with a firm smile, a shoulder tap and some internal positive programming boosts personal responsibility and performance. I trust my team. Their excellence and growth are part of my excellence and growth. I trust my team. Their excellence and growth are part of my excellence and growth. This phrase is the opposite of what we all tend to secretly think: ‘Nobody is going to do it better than me’. The belief in one’s superiority undermines the confidence of people you work with. Delegating with a sigh is the beginning of the end. In contrast, delegating with a firm smile, a shoulder tap and some internal positive programming boosts personal responsibility and performance. Done better when delegated is far more valuable than done perfectly by me. That hurts, doesn’t it? Well, the truth is, you could spend hours on getting things done at 100%, but you could also spend hours on doing x2 and getting these things just done. Sadly (or thankfully), perfection is not a prerequisite for effective operations. Done better when delegated is far more valuable than done perfectly by me. Done better when delegated is far more valuable than done perfectly by me. That hurts, doesn’t it? Well, the truth is, you could spend hours on getting things done at 100%, but you could also spend hours on doing x2 and getting these things just done. Sadly (or thankfully), perfection is not a prerequisite for effective operations. Not delegating what I can do well means missing out on what only I can do. Mundane work is debilitatingly comfortable. It may simply be scary to distribute work across the team to get a task done because… you may not have a plan yet. Hiding behind ‘I know how to do it’ in the face of a real challenge is akin to betrayal, so to speak. One’s personal productivity is not in how much operational stuff one can crunch in an hour but how quick one is to learn. Not delegating what I can do well means missing out on what only I can do. Not delegating what I can do well means missing out on what only I can do. Mundane work is debilitatingly comfortable. It may simply be scary to distribute work across the team to get a task done because… you may not have a plan yet. Hiding behind ‘I know how to do it’ in the face of a real challenge is akin to betrayal, so to speak. One’s personal productivity is not in how much operational stuff one can crunch in an hour but how quick one is to learn. Let me get slightly personal here. As a seasoned CMO, I love to read reports. I can spend hours on Linkedin reading other people’s success stories and muse for ages on what we should do. But do I spend hours? No. I read reports to get the information I need fast. So I run reports through AI to crunch them into executive summaries. Why talk about personal beliefs behind delegation in an article on AI? The trick is simple: we tend to avoid delegation because of the backward beliefs that the mantras help combat. If we’re forced to delegate, we may not do it correctly - either because we are blinded or because we simply don’t know how to tell what we want. With AI, just like with humans, you need to have a clear image of what you expect before you talk to people. Effective Briefing Starts Here I loved this quote by Jarrad Roeder: As we practice framing better prompts for AI, we inherently sharpen our ability to communicate more effectively with humans. The subtle changes we experiment with—clarity, specificity, tone—carry directly into real-world conversations in a way that impacts understanding, persuasion, and empathy. It lends itself so well to the common pitfall people repeatedly encounter when delegating - or even sharing. The message you send out - your brief, your understanding of the end result - only exists in your head. Taking it out on paper - or in a message field in ChatGPT - takes some skill. In my journey to being a better manager, I was able to identify these 3 rules. Rule #1. Instruct, instruct, instruct Rule #1. Instruct, instruct, instruct The very word ‘instruction’ invites care for the person you delegate to. To me, to instruct means to equip for every step of the way: from ideation to deadlines to suggestions on actions to specific outcomes. What’s in it for AI? Just as instructing a human colleague means giving clarity at every stage, instructing an AI system requires the same intentional care, if not more so. Vague requests provide vague outputs. To achieve precision, you need to help AI build some background that covers your persona, the greater context and the intent. The process externalizes your thought and put a finger on assumptions that might otherwise be left implicit. To turn AI into an extension to your thinking process, you need to have a collaborative relationship with it. Rule #2. Don’t micromanage. Rule #2. Don’t micromanage. I’ve had a manager who would turn every task delegated into an agony of control. He would drive himself up the wall with his ‘I recommend’. Eventually, he’d snap with a nasty ‘Oh I should have just done that myself’. The end result? Anxious subordinates were unwilling to help and the manager was convinced we were sabotaging him. The problem with micromanagement is that it kills any initiative and creativity that comes with a different person on task. If you do it this way, they might do the other-and actually get things done better. Delegating without delegating is a sure way to undermine any work relationship. To put it into AI terms, the temptation to micromanage AI is different from micromanaging humans, but equally counter-productive. You micromanage when you treat every iteration as a failure or when you hurry to rewrite a prompt even before anything comes out of it. By doing this, you upset the very idea of micromanagement–to let another intelligence contribute to the process. Like when a human subordinate freezes under non-stop correction, so does an AI system. It restricts itself into narrow guardrails of your feedback and novel and creative outcome simply gets out of reach. The antidote here is to let AI create or fail creatively, on its own. Take the best out of what it does. Rule #3. The known unknowns. Rule #3. The known unknowns. Ah, that’s the trickiest part which is extremely easy to miss. Humans perform every task from a lifetime of accumulated context under their belt. People know the unwritten rules, the dynamics and the unseen do’s and dont’s. Every delegation to AI is a delegation to someone on their first day, with no corporate memory, no sense of your priorities beyond what you explicitly state, and no intuition about what matters in your specific context. This creates a deceptive trap: AI is so articulate, so confident, and so capable at pattern-matching that it feels like it understands your world. But it doesn't. It's operating on the surface of what you've told it, extrapolating from its training data, without the embodied knowing that comes from actually living within a system. The concept of unknown knowns can also be viewed from a different perspective. We, humans, know the shape of our own ignorance. For example, I know I lack expertise in corporate finance, so I’d go ask someone if I had to. This knowing of your unknowns creates a feedback loop which AI cannot do reliably. It generates outputs with the same level of confidence regardless of whether it's reasoning from solid ground or vague common knowledge. The logical consequence is, AI doesn't recognize its limits and cannot signal when it's operating beyond its actual competence. This is why delegation to AI requires you to maintain epistemic vigilance. The user must be clear about the boundary between what the AI can reliably do and where it hallucinates confidence. Takeaways I want every manager preparing to delegate to AI and hoping for an effective and honest conversation with it, to understand these things: Vagueness in, vagueness out. When your request sounds like ‘make good social media posts’, the result is going to be less than satisfactory. Human performance doesn’t benefit from ambiguous task-setting. Generic, underwhelming prompt returns are only the logical outcome. Just like with people, effective delegation to AI requires clarity and quality instructions. Put on your own mask first, then assist others. I think this phrase you hear on airplanes is a perfect wording for my next thought. To delegate and achieve results, YOU need to spend time on the task first. See the process through and through, every stage and milestone. Before you blame the other person for their misunderstanding the task, make sure your instruction was adequate covering every aspect of work. Vagueness in, vagueness out. When your request sounds like ‘make good social media posts’, the result is going to be less than satisfactory. Human performance doesn’t benefit from ambiguous task-setting. Generic, underwhelming prompt returns are only the logical outcome. Just like with people, effective delegation to AI requires clarity and quality instructions. Put on your own mask first, then assist others. I think this phrase you hear on airplanes is a perfect wording for my next thought. To delegate and achieve results, YOU need to spend time on the task first. See the process through and through, every stage and milestone. Before you blame the other person for their misunderstanding the task, make sure your instruction was adequate covering every aspect of work. My Prompt Writing Checklist Task description should be lengthy, concrete and detailed. Write a comprehensive product comparison guide for fintech payment solutions targeting freelancers in Eastern Europe. The guide should compare at least 5 platforms (Wise, Payoneer, Revolut, PayPal, and Stripe Connect), evaluating each across 8 specific dimensions: transaction fees for international transfers, withdrawal speeds, supported currencies, compliance with local regulations in CIS countries, user interface complexity, mobile app functionality, customer support responsiveness, and fraud protection measures. For each platform, include concrete fee examples showing the cost of transferring $500, €500, and ₽50,000. The output should be approximately 2,500 words and structured with an executive summary, detailed comparison tables, and individual platform deep-dives. Provide sufficient context and the viewpoint to adopt. Point out specific considerations for the task. You are a content strategist for a B2B SaaS HR tech company targeting mid-market enterprises (100-500 employees) in Germany and Austria. Write a LinkedIn post about remote work compliance challenges. Adopt a thought-leadership perspective that positions our audience as forward-thinking HR leaders. Emphasize: (a) the complexity of multi-country payroll regulations, (b) the business risk of non-compliance, and (c) how automation reduces administrative burden. The tone should be confident but not preachy. Acknowledge that HR directors are already stretched thin, so frame our perspective as a partner solving real problems, not a vendor pushing solutions. Target post length: 180-220 words for optimal LinkedIn engagement. AI models learn from provided data. You can add prior data sets that explicitly state that the agent can use these artifacts A and B as references and artifacts C and D as anti-references. Generate 5 subject lines for an email newsletter about AI automation for freelancers. Use these as positive references—study their effectiveness: [INSERT: 3 high-performing subject lines from past campaigns with open rates >35%]. Use these as anti-references—avoid this tone and structure: [INSERT: 2 subject lines from low-performing campaigns with open rates <15%]. The new subject lines should feel urgent and benefit-focused like the reference examples, but must be original. They should be 5-8 words maximum, include a power verb, and speak directly to pain points (time-saving, income growth, or compliance simplification). Avoid clickbait and vague language. AI models may not grasp the desired output format from the task description. To be on the safe side, you can actually point out the desired outcome. Create an SEO competitor analysis for our fintech blog. Return the output as a structured markdown table with 6 columns: [Competitor Name | Primary Keywords (top 10 by volume) | Content Pillars (main topic clusters) | Publishing Frequency | Estimated Monthly Traffic | Content Gap Opportunities]. Include 4 competitors. Below the table, add a 200-word executive summary highlighting: (1) which competitor owns the most valuable keywords, (2) content topics none of them cover, and (3) our recommended priority topics. Format the summary in 3 labeled sections with bold headers: 'Keyword Dominance Analysis,' 'Content Gap Opportunities,' and 'Strategic Recommendations’. Task description should be lengthy, concrete and detailed. Write a comprehensive product comparison guide for fintech payment solutions targeting freelancers in Eastern Europe. The guide should compare at least 5 platforms (Wise, Payoneer, Revolut, PayPal, and Stripe Connect), evaluating each across 8 specific dimensions: transaction fees for international transfers, withdrawal speeds, supported currencies, compliance with local regulations in CIS countries, user interface complexity, mobile app functionality, customer support responsiveness, and fraud protection measures. For each platform, include concrete fee examples showing the cost of transferring $500, €500, and ₽50,000. The output should be approximately 2,500 words and structured with an executive summary, detailed comparison tables, and individual platform deep-dives. Task description should be lengthy, concrete and detailed. Write a comprehensive product comparison guide for fintech payment solutions targeting freelancers in Eastern Europe. The guide should compare at least 5 platforms (Wise, Payoneer, Revolut, PayPal, and Stripe Connect), evaluating each across 8 specific dimensions: transaction fees for international transfers, withdrawal speeds, supported currencies, compliance with local regulations in CIS countries, user interface complexity, mobile app functionality, customer support responsiveness, and fraud protection measures. For each platform, include concrete fee examples showing the cost of transferring $500, €500, and ₽50,000. The output should be approximately 2,500 words and structured with an executive summary, detailed comparison tables, and individual platform deep-dives. Write a comprehensive product comparison guide for fintech payment solutions targeting freelancers in Eastern Europe. The guide should compare at least 5 platforms (Wise, Payoneer, Revolut, PayPal, and Stripe Connect), evaluating each across 8 specific dimensions: transaction fees for international transfers, withdrawal speeds, supported currencies, compliance with local regulations in CIS countries, user interface complexity, mobile app functionality, customer support responsiveness, and fraud protection measures. For each platform, include concrete fee examples showing the cost of transferring $500, €500, and ₽50,000. The output should be approximately 2,500 words and structured with an executive summary, detailed comparison tables, and individual platform deep-dives. Provide sufficient context and the viewpoint to adopt. Point out specific considerations for the task. You are a content strategist for a B2B SaaS HR tech company targeting mid-market enterprises (100-500 employees) in Germany and Austria. Write a LinkedIn post about remote work compliance challenges. Adopt a thought-leadership perspective that positions our audience as forward-thinking HR leaders. Emphasize: (a) the complexity of multi-country payroll regulations, (b) the business risk of non-compliance, and (c) how automation reduces administrative burden. The tone should be confident but not preachy. Acknowledge that HR directors are already stretched thin, so frame our perspective as a partner solving real problems, not a vendor pushing solutions. Target post length: 180-220 words for optimal LinkedIn engagement. Provide sufficient context and the viewpoint to adopt. Point out specific considerations for the task. You are a content strategist for a B2B SaaS HR tech company targeting mid-market enterprises (100-500 employees) in Germany and Austria. Write a LinkedIn post about remote work compliance challenges. Adopt a thought-leadership perspective that positions our audience as forward-thinking HR leaders. Emphasize: (a) the complexity of multi-country payroll regulations, (b) the business risk of non-compliance, and (c) how automation reduces administrative burden. The tone should be confident but not preachy. Acknowledge that HR directors are already stretched thin, so frame our perspective as a partner solving real problems, not a vendor pushing solutions. Target post length: 180-220 words for optimal LinkedIn engagement. You are a content strategist for a B2B SaaS HR tech company targeting mid-market enterprises (100-500 employees) in Germany and Austria. Write a LinkedIn post about remote work compliance challenges. Adopt a thought-leadership perspective that positions our audience as forward-thinking HR leaders. Emphasize: (a) the complexity of multi-country payroll regulations, (b) the business risk of non-compliance, and (c) how automation reduces administrative burden. The tone should be confident but not preachy. Acknowledge that HR directors are already stretched thin, so frame our perspective as a partner solving real problems, not a vendor pushing solutions. Target post length: 180-220 words for optimal LinkedIn engagement. AI models learn from provided data. You can add prior data sets that explicitly state that the agent can use these artifacts A and B as references and artifacts C and D as anti-references. Generate 5 subject lines for an email newsletter about AI automation for freelancers. Use these as positive references—study their effectiveness: [INSERT: 3 high-performing subject lines from past campaigns with open rates >35%]. Use these as anti-references—avoid this tone and structure: [INSERT: 2 subject lines from low-performing campaigns with open rates <15%]. The new subject lines should feel urgent and benefit-focused like the reference examples, but must be original. They should be 5-8 words maximum, include a power verb, and speak directly to pain points (time-saving, income growth, or compliance simplification). Avoid clickbait and vague language. AI models learn from provided data. You can add prior data sets that explicitly state that the agent can use these artifacts A and B as references and artifacts C and D as anti-references. Generate 5 subject lines for an email newsletter about AI automation for freelancers. Use these as positive references—study their effectiveness: [INSERT: 3 high-performing subject lines from past campaigns with open rates >35%]. Use these as anti-references—avoid this tone and structure: [INSERT: 2 subject lines from low-performing campaigns with open rates <15%]. The new subject lines should feel urgent and benefit-focused like the reference examples, but must be original. They should be 5-8 words maximum, include a power verb, and speak directly to pain points (time-saving, income growth, or compliance simplification). Avoid clickbait and vague language. Generate 5 subject lines for an email newsletter about AI automation for freelancers. Use these as positive references—study their effectiveness: [INSERT: 3 high-performing subject lines from past campaigns with open rates >35%]. Use these as anti-references—avoid this tone and structure: [INSERT: 2 subject lines from low-performing campaigns with open rates <15%]. The new subject lines should feel urgent and benefit-focused like the reference examples, but must be original. They should be 5-8 words maximum, include a power verb, and speak directly to pain points (time-saving, income growth, or compliance simplification). Avoid clickbait and vague language. AI models may not grasp the desired output format from the task description. To be on the safe side, you can actually point out the desired outcome. Create an SEO competitor analysis for our fintech blog. Return the output as a structured markdown table with 6 columns: [Competitor Name | Primary Keywords (top 10 by volume) | Content Pillars (main topic clusters) | Publishing Frequency | Estimated Monthly Traffic | Content Gap Opportunities]. Include 4 competitors. Below the table, add a 200-word executive summary highlighting: (1) which competitor owns the most valuable keywords, (2) content topics none of them cover, and (3) our recommended priority topics. Format the summary in 3 labeled sections with bold headers: 'Keyword Dominance Analysis,' 'Content Gap Opportunities,' and 'Strategic Recommendations’. AI models may not grasp the desired output format from the task description. To be on the safe side, you can actually point out the desired outcome. Create an SEO competitor analysis for our fintech blog. Return the output as a structured markdown table with 6 columns: [Competitor Name | Primary Keywords (top 10 by volume) | Content Pillars (main topic clusters) | Publishing Frequency | Estimated Monthly Traffic | Content Gap Opportunities]. Include 4 competitors. Below the table, add a 200-word executive summary highlighting: (1) which competitor owns the most valuable keywords, (2) content topics none of them cover, and (3) our recommended priority topics. Format the summary in 3 labeled sections with bold headers: 'Keyword Dominance Analysis,' 'Content Gap Opportunities,' and 'Strategic Recommendations’. Create an SEO competitor analysis for our fintech blog. Return the output as a structured markdown table with 6 columns: [Competitor Name | Primary Keywords (top 10 by volume) | Content Pillars (main topic clusters) | Publishing Frequency | Estimated Monthly Traffic | Content Gap Opportunities]. Include 4 competitors. Below the table, add a 200-word executive summary highlighting: (1) which competitor owns the most valuable keywords, (2) content topics none of them cover, and (3) our recommended priority topics. Format the summary in 3 labeled sections with bold headers: 'Keyword Dominance Analysis,' 'Content Gap Opportunities,' and 'Strategic Recommendations’. In the end, a good prompt is a lengthy message that treats the model as a co-worker and equips it for quality output. AI models indeed outpace the human brain in terms of speed: faster analysis and synthesis, sharper focus on discrepancies and similarities. Yet AI—at least for now—can’t read your thoughts. Where something may be obvious to you, it is not to an outside, whether a machine or another human. Explaining your thoughts in a structured way is a prerequisite for effective communication and meaningful collaboration with AI.
