TL;DR
- Communication is a loop. Explain your intent, listen or read carefully, check understanding, and correct the mismatch.
- Practice can help at any career stage. Many people I work with, or have worked with, communicate very well. Some benefit from focused practice, including experienced professionals.
- Agents increase the consequences. A misunderstanding can shape several steps of delegated work before someone notices.
- Practise deliberately. Use a short exercise, check specific feedback, and retry. The Communication Coach has five complete prompts you can use in ChatGPT or Claude today.
Many people I work with today, and have worked with in the past, are excellent communicators. I have also seen capable colleagues benefit from help explaining what they need, understanding a brief, or turning a discussion into a clear next step. I am sharing this as an invitation to practise, with respect for the people involved.
This includes people with four to six years of professional experience. They know their work, but their explanation may leave out essential context. A request may contain plenty of background without saying what decision is needed. Someone may hear an instruction and miss the condition that changes what they should do.
These observations do not define anyone’s ability or worth, and they are not a claim about an entire team, generation, or experience group. Communication also depends on context: the audience, time available, language, and clarity of the original request. They make me question how deliberately we develop these skills throughout our careers.
The growing use of agentic AI makes this practice more important for all of us, including people who already communicate well. We increasingly use conversation to delegate work. We also need to understand what comes back well enough to decide whether the work is useful, accurate, and complete.
The skill includes both expression and understanding
Clear communication combines several abilities: identifying the point, choosing relevant context, asking a useful question, listening, and checking what the other person understood.
Fluent English and an impressive vocabulary do not establish those abilities. A simple sentence can communicate a decision more accurately than a polished paragraph. An accent does not tell us whether someone understood the problem.
Reading and listening deserve equal attention. Recognising the words is only the beginning. We need to retain the conditions, distinguish facts from assumptions, and understand the next action.
Consider this fictional agent update: “The migration is complete in staging. Production remains unchanged. Three records failed validation.”
Reading only “migration complete” changes the meaning. The environment matters. The unchanged production state matters. The failed records matter. Understanding those details determines whether the next step is validation, correction, or a production change.
What research helps explain
Comprehension challenges are widespread. In the OECD's 2023 Survey of Adult Skills, 26% of adults, on average across participating OECD countries and economies, scored at Level 1 or below in literacy. This indicates difficulty with more demanding reading tasks, not an inability to read. It is not an estimate for India or for professionals with a particular number of years of experience. OECD findings.
Listening takes cognitive effort. Jonathan Peelle's review describes how challenging speech conditions recruit attention and working memory. Successful word recognition can coexist with poorer memory for the content. This is a reason to improve meeting audio and the listening environment as well as individual habits. Peelle, 2018.
AI use requires us to monitor our own understanding. Lev Tankelevitch and colleagues frame generative AI use in terms of metacognition: identifying goals, breaking tasks down, judging outputs, and adapting when an approach fails. Their framework helps explain why a useful interaction starts before the first prompt and continues after the answer. Tankelevitch and colleagues, 2024.
Confidence in the tool can affect scrutiny. A CHI 2025 study surveyed 319 knowledge workers, collecting 936 workplace examples. Higher confidence in generative AI was associated with less reported critical thinking. Participants described work shifting toward verification, integration, and oversight. These are self-reported associations; they do not prove that AI causes declining intelligence or comprehension. Lee and colleagues, 2025.
These studies address different populations and tasks. They do not demonstrate that experienced professionals generally lack communication skills, or that a single communication score predicts success with autonomous agents. My argument is a practical inference from this research and my own experience: understanding and explaining work deserve deliberate practice as delegation becomes easier.
Why agents make clarity more consequential
Take a fictional request: “Research our competitors and prepare a presentation.”
An agent still needs to resolve which competitors, which market, what sources to use, who will read the presentation, and which decision it should support. If those choices remain implicit, the resulting work may answer a different question from the one we intended.
A more useful brief would say: “Compare these three competitors for our India launch. Use their current public pricing and product documentation. Prepare five slides for the sales team covering pricing, integrations, and support. Cite sources, mark unknowns, and explain imperfect comparisons. Give me a draft before sharing it.”
The second request describes the outcome, context, constraints, success criteria, and authority boundary. Those are useful habits when delegating to a colleague too.
My concern with agents is how far an unresolved ambiguity can travel. It can influence source selection, analysis, slide creation, and delivery. An early clarification can prevent several steps of rework.
The responsibility is shared. An agent should ask useful questions, distinguish evidence from inference, and expose uncertainty. People should be able to begin with an incomplete thought and develop it through conversation. A better system helps us clarify what we want.
Comprehend, evaluate, then decide
There are three distinct questions to ask when reviewing an agent's response:
- Comprehension: What does it actually say, including conditions and exceptions?
- Evaluation: What evidence supports it, and does that evidence apply here?
- Decision: Given that, what should happen next?
A clear explanation can still be wrong. Before approving an important action, I would want to explain the proposal back in my own words and identify what remains unresolved.
There is experimental support for deliberate checks. In a study involving 199 participants, Buçinca and colleagues found that interventions prompting more analytical engagement reduced overreliance compared with simpler explainable-AI designs. The most effective designs also received less favourable subjective ratings. The result concerns that decision task, rather than every agent workflow, but it shows why the easiest interaction is not always the one that produces the best judgment. Buçinca and colleagues, 2021.
Ten minutes of daily practice
I would start with one exercise a day. Spend two minutes preparing, three minutes responding, two minutes checking feedback, and three minutes retrying with one deliberate improvement.
These are practical exercises, not a validated training programme. One useful ingredient is retrieval practice: recalling information before returning to the source. Karpicke and Blunt found benefits for meaningful learning, including comprehension and inference questions, in their science-text learning tasks. That supports the recall step without establishing this entire routine's effectiveness. Karpicke and Blunt, 2011.
Explain clearly
Describe a real, non-sensitive work problem in about a minute. Ask your practice partner to paraphrase the main point and request. Look for an ambiguity or unnecessary detail if present, then revise where useful. Preserve the essential conditions; a clear explanation need not be shortened just to meet a word target.
Listen and recall
Hear a short brief containing a goal, constraints, an exception, and an unanswered question. Explain what you understood before checking the transcript. Identify what you missed and what you assumed. With text, use the same method as reading and recall practice.
Read and interpret
Read a short passage, then state its main claim, supporting evidence, a limitation, and something it does not establish. Compare your answer with the exact sentences. This develops the habit of distinguishing the author's statement from your inference.
Clarify and delegate
Start with an ambiguous request such as “improve onboarding.” Ask up to five clarification questions. Write a brief covering the outcome, audience, constraints, success criteria, and what requires a decision. Check whether you invented any requirements.
Understand before disagreeing
Rehearse a discussion with a colleague who has a reasonable objection. Paraphrase their concern, ask a relevant follow-up question, and then respond. Check whether your recommendation addresses the concern they actually raised.
Use ChatGPT or Claude as a practice partner
You can use ChatGPT or Claude for these exercises. The Communication Coach page provides ChatGPT and Claude buttons that open each complete coaching prompt in a new chat, with Copy prompt as a fallback. Review the prompt before opening it, send it if the service has not already done so, start voice when available, and say “Ready” to begin; type “Ready” instead for text-only practice. The instructions include feedback and a retry before a suggested answer. The service may send linked prompts automatically; the buttons do not start voice. A quick practice table at the end of this article links directly to all five exercises in both services.
Use chat when you want to inspect wording and compare versions. Use voice when available to rehearse listening and responding aloud. See the official ChatGPT Voice guide and Claude voice guide. Account access, platforms, and limits can differ.
After you say or type “Ready”, the coach asks for optional context: your domain, role, audience, and preferred challenge. Try “retail shift lead, briefing colleagues, introductory” or “experienced finance analyst, executive audience, challenging”; say “use defaults” to start with a general workplace case. Difficulty should follow the task and your preference, rather than assumptions about seniority. Ask for plain language, shorter material, or written practice when useful. The scenario should supply any specialist facts you need. These prompts are designed to adapt across roles; they have not been validated for every domain or experience level.
Ask for “a new example” to change the situation. Ask to “try again” to keep the facts stable and improve your response. Varying the domain, audience, constraints, and follow-up questions makes it harder to memorise a model answer instead of practising the skill.
At the end, save a short summary of the exercise, scenario, strength, priority, and next practice. Paste it into the next chat if needed. This version has no publisher-run progress database or built-in reminders. Set an optional reminder separately in your calendar or a supported scheduling feature.
Communication Coach is one practice option. A trusted colleague, mentor, communication course, reading group, or a simple record-and-review routine can also help you practise. Choose an approach that gives you useful feedback and fits your circumstances. You do not need this particular tool to begin.
AI feedback is practice material, so check it against the original. It should not become an employment assessment.
A small investment in yourself
What interests me is why short, inexpensive practice is sometimes hard to start or sustain, even when we want to improve. Tools like these make me curious about what would help more of us turn that intention into a habit. Five or ten minutes spent explaining a point, checking our understanding, and trying again is time invested in ourselves.
I would not assume that someone who does not use a tool lacks interest in learning. They may not know it exists, may be unsure whether its feedback is useful, may feel uncomfortable practising, or may have competing demands on their time. Those are possible barriers to explore, rather than conclusions about any individual. Having access to a tool does not automatically create a habit.
Start with a small commitment that works for you: one exercise, one piece of feedback, one retry. Use a free option where available; no paid coach or subscription is required to practise with a colleague or review your own explanation. A short session is a starting point, with improvement checked over time rather than promised in a few minutes. As agentic AI makes explanation, comprehension, and judgment more consequential, this feels like a worthwhile part of ongoing professional development.
What I would ask managers to change
“Improve your communication” gives people little to work with. “I understood the background, but I could not identify the decision you needed from me” identifies a specific behaviour they can practise.
Give people opportunities to explain a recommendation, paraphrase a brief, and ask clarification questions before acting. Check whether instructions themselves are clear. A missed condition can reflect an overloaded brief as well as an inattentive listener.
Progress should show up in the work: a listener can paraphrase the request accurately, important constraints survive a summary, and questions resolve uncertainty before execution. Immediate improvement after feedback is useful, but a fresh task a week later is a better check of whether the habit is transferring.
I want people in my team, including experienced professionals, to feel comfortable saying: “Let me check that I understood.” As agents take on more work, that small pause can be one of the most useful parts of the conversation.