How AI Is Changing Communication: Weighing Efficiency with Efficacy

Alison Beard and Adi Ignatius introduce communication as one of the central ways generative AI is changing executive work. They note that leaders and employees are using AI for emails, speeches, internal reports, client pitches, and presentations. Ignatius, who writes professionally, says he is uncomfortable having AI write for him but welcomes it as a research assistant; he has also seen it provide substantial writing and organizational help to people who are not natural writers. Beard agrees that AI can help, but warns that overreliance can prevent people from engaging with the work, sounding human, or even making sense despite producing polished prose. Stanford professor Matt Abrahams says AI can help leaders synthesize huge amounts of information, organize it meaningfully, and identify themes and trends that would otherwise take a long time to uncover. But he also says it can jeopardize authenticity and hinder empathy. Zapier CEO Wade Foster similarly sees major benefits and major etiquette problems. AI can summarize information from Slack, email, meeting notes, project-management and CRM tools, and even a company’s codebase, giving him a more comprehensive view than he could obtain by talking through layers of management. Yet inconsistent data or speaker attribution can make summaries unreliable, and employees may simply paste AI output into communications without checking whether it expresses the right idea. The result can be a runaway volume of low-effort, low-thought communication.

Asked whether leaders should use AI for speeches, emails, board decks, or client pitches, Abrahams recommends using it to overcome the blank page, develop an initial structure, and turn scattered ideas into themes. He also uses it after a meeting to review questions and discussion for recurring themes, then checks whether the communication addressed them. AI can help leaders think through alternative structures for different audiences, but Abrahams warns that blindly asking it to create a board deck, client pitch, or all-hands presentation can produce trouble if the leader never evaluates the result. Foster says he uses AI across these communication tasks, but returns to a principle from his cofounder Brian: “You can delegate the work, but you can’t delegate the accountability.” AI-assisted communication is still the executive’s responsibility, so the leader must inspect the quality and make the final judgment. He argues that AI can produce good or bad work, just as humans can, and leaders must model responsible use by setting expectations. Abrahams recommends that companies explicitly define AI as a thought or sparring partner, require people to review and synthesize the final product, and discuss AI use during hiring and performance reviews. At Zapier, Foster says the author should put more effort into a communication than the reader, and should disclose whether AI was an assistant, co-author, or barely reviewed first draft. He also says expectations should vary by medium: customer communications require more responsibility than an automatically generated status update or Jira ticket.

Foster says he would never fully delegate company all-hands remarks, kickoff speeches, or other communications carrying significant emotional weight. He may use AI to prepare, but he wants to be personally present. Abrahams makes the broader recommendation that leaders should never outsource empathy. AI can help a person consider how to tailor a message to someone’s characteristics, but repeated delegation of difficult conversations can cause “empathy atrophy.” Negative performance reviews, constructive feedback, and attempts to resolve team conflict require the communicator to be fully invested and able to respond authentically. Beard observes that AI is good at polished prose but not necessarily at logical structure, and says it often needs extensive prompting to provide useful coaching or rehearsal rather than merely make a leader sound more polished. Abrahams adds that AI can create a “comprehension gap”: communication is the end result of critical thinking, and people who outsource too much of that thinking may become anxious when asked questions or required to explain their own message. Foster acknowledges seeing this in himself and his staff. He says responsible AI use can make people better, but users must recognize when they have delegated so much that they no longer understand the subject. The appropriate amount of delegation depends on the task’s stakes and the consequences of getting it wrong.

Beard worries that AI can make executives sound like average corporate bots, full of smooth but interchangeable jargon. Abrahams calls this an “authenticity crisis”: audiences may not know whether a person’s words are genuinely theirs or were supplied by a chatbot. He recommends deliberately signaling humanness through occasional filler words or hesitations, sharing one’s thought process, acknowledging that a question is difficult, and connecting an idea to a personal experience. These details can show that a real person is thinking rather than presenting an artificially perfect answer. Abrahams has trained an AI system on episodes of his Think Fast, Talk Smart podcast so it can reproduce some of his colloquialisms. Foster has built a “Wade bot” using large bodies of his writing, but says it reaches only 80% or 90% of the way to sounding like him and must be updated as models change. He finds that newer models can be strong at coding while becoming less effective and more personality-free at communication. When AI edits his writing, Foster often rejects suggestions that make him sound more polished because they sand away his “raw edges.” He says people want to hear what Wade thinks, “warts and all.” Abrahams is especially concerned about younger or newer workers who may use AI as a crutch to sound confident. He urges senior leaders to reinforce that employees’ own voices and ideas matter, and Foster says he calls out writing that sounds like “AI slop” by asking the author what they actually think.

The guests say organizations may need to redesign situations so people demonstrate their own thinking instead of relying on hidden chatbots. Abrahams supports spontaneous speaking and even mentions the return of handwritten “blue books” as a way to assess thought directly. He says in-person settings may sometimes be more valuable than remote ones because they make it harder to outsource answers and can improve discussion and ideation. AI can still help people prepare: Foster uses voice-to-text and found that reviewing transcripts exposed habits in his own speech and made him want to write better. For difficult conversations, he asks AI to help identify talking points, then uses only bullet-point checkpoints rather than reading polished chatbot language. Abrahams compares AI rehearsal to athletic sparring. Leaders can ask an AI to play a skeptical stakeholder or an experienced critic and practice answering questions without memorizing scripted responses. In his Stanford classes, students practice Q&A with an LLM to get repetitions. At Zapier, scheduled bots in Slack are openly identified as bots and receive direct feedback. Customer-support bots operate in more structured workflows with human review, CSAT measurements, and internal evaluations. Foster says personal agents that post in Slack as if they were their owners are more problematic; clearly labeling an agent as posting on someone’s behalf helps preserve trust. Abrahams warns that human language is already becoming more homogenized and frequent, with fewer signs of individual variation, as people adapt to communicating with LLMs.

The discussion turns to whether persuasion and charm will become more important as AI makes writing and content creation widely available. Abrahams agrees that communication is fundamentally about connection—the root of “communication,” he notes, means “to make common.” AI can assist with relevance and persuasion, but people still need to listen, paraphrase, reflect, and build empathy. He hopes those relational skills remain difficult for AI to master. When Beard asks what AI should deliberately not optimize, Abrahams says it should sometimes prompt users to pause and reflect, especially when a request concerns an intense personal relationship or could cause serious damage, rather than immediately supplying the most efficient answer. Foster says the models’ goal-seeking design can make them overly eager to please, producing what they think the user wants rather than an objectively sound recommendation. He changes his system prompts to ask AI to debate him, challenge assumptions, and push back. For the one human skill leaders must preserve, he chooses curiosity and judgment: people should not turn down the part of their minds that cares about the work and investigates it deeply. His practical AI recommendation is to connect the model to the organization’s actual communication systems and data, because company-specific context produces better guidance than generic coaching. Abrahams’s corresponding recommendation is to never automate empathy and to use AI before high-stakes communications to analyze the audience and suggest ways to make the message more relevant, tight, and specific.