Communication agents

The hidden cost of unstructured communication

You write emails, give presentations, and navigate difficult conversations on a recurring basis. These are not novel problems. The email you are drafting right now is structurally identical to one you wrote last month and will write again next month. The presentation you are building follows the same logical pattern as the last three. The tense conversation you are avoiding has the same underlying architecture as the last tense conversation you avoided.

And yet, each time, you start from scratch. You stare at the blank compose window. You rearrange your slides for the fourth time. You rehearse opening lines in your head and discard them all. The cognitive effort is not going toward the content of your message — the actual information, argument, or request you need to convey. It is going toward the form. You are solving the structural problem of how to organize a communication every single time, as if you had never done it before.

This is the exact problem agents solve. An agent, as you have been learning throughout Phase 21, is a reusable cognitive structure that handles a recurring task so your deliberate attention can focus elsewhere. In L-0416, you built agents for recurring decisions. This lesson builds agents for recurring communications. The principle is identical: when a situation recurs and its underlying structure is stable, encode that structure once and deploy it repeatedly. Stop reinventing the form so you can invest in the content.

Three frameworks, each battle-tested across decades and domains, give you the raw material for your three most common communication agents: BLUF for direct messages, the Pyramid Principle for presentations and arguments, and the Three Conversations model for difficult interpersonal exchanges.

BLUF: the military protocol for direct communication

The U.S. Army formalized a communication standard that solves the most common failure in written messages — burying the point. Army Regulation 25-50, "Preparing and Managing Correspondence," first published in 1988 and revised in 2001, codified the principle of Bottom Line Up Front, or BLUF: begin every message with its key information. The regulation states that the two essential requirements for Army correspondence are "putting the main point at the beginning of the correspondence (bottom line up front) and using the active voice."

The military adopted BLUF because the cost of buried information is measured in lives. A field commander does not have time to read three paragraphs of context before discovering that the ammunition resupply is delayed. The bottom line — we are short on ammunition and need resupply by 0600 — must come first. Everything else is supporting detail.

Your emails are not life-or-death. But the cognitive economics are identical. Your recipient is scanning, not reading. They have forty other messages competing for attention. If your main point appears in paragraph three, it may never be found. And you — the sender — pay a cost too. Without BLUF as a structural constraint, you default to chronological narration or stream-of-consciousness, which takes longer to write and produces a less useful artifact.

A BLUF agent for email has four fields:

1. Bottom line. One to two sentences stating the request, decision, or information the recipient needs. This is the first thing they read.
2. Context. The minimum background necessary to understand why the bottom line matters. Not the full history — just enough for the reader who was not in the room.
3. Details. Supporting information, data, options, or constraints. This section can be longer, because the reader already knows what it is supporting.
4. Action required. What you need from the recipient, by when. If no action is needed, say so explicitly.

Barbara Minto's SCQA framework — Situation, Complication, Question, Answer — provides an even more precise variant for persuasive emails. The Situation establishes common ground in 20 to 30 words. The Complication introduces the tension or problem. The Question articulates what needs to be resolved. The Answer delivers your recommendation. The entire structure fits in 150 to 200 words and eliminates the back-and-forth that unstructured messages generate.

The point is not that BLUF is the only way to write an email. The point is that having a BLUF agent means you never spend cognitive effort deciding how to structure a status update, a request, or an escalation. The structure is pre-solved. You fill in the fields with the content that actually matters.

The Pyramid Principle: top-down structure for arguments and presentations

In the 1970s, Barbara Minto — the first female post-MBA hire at McKinsey — developed what became the dominant framework for structured business communication. The Pyramid Principle inverts the natural order of thinking: you think bottom-up (from data to analysis to conclusion), but you communicate top-down (from conclusion to supporting arguments to evidence).

The structure has three levels. At the top sits your main point — your recommendation, your answer, your "so what." Beneath that are three to five supporting arguments — the key reasons your audience should believe your main point. At the base are the evidence and details — the facts, data, and analysis that give those arguments weight. The groups at each level must be MECE: mutually exclusive (no overlap between groups) and collectively exhaustive (no gaps in coverage).

McKinsey still teaches new hires the Pyramid Principle today, and most major consulting firms follow suit. The reason it has endured for five decades is not institutional inertia. It is because the framework maps to how decision-makers actually consume information. An executive hearing a presentation does not want to follow your analytical journey. They want your answer first, then the reasons they should trust it, then — only if they want to go deeper — the evidence behind each reason. The pyramid gives them an exit ramp at every level: if they believe your main point after hearing the supporting arguments, they do not need the base layer at all.

A Pyramid Principle agent for presentations has these fields:

1. Governing thought. One sentence that captures the entire message. If the audience remembers nothing else, this is what they take away.
2. Key supporting arguments. Three to five reasons the governing thought is true, each expressible in a single sentence.
3. Evidence base. For each supporting argument, the specific data, examples, or analysis that substantiate it.
4. So-what cascade. For each piece of evidence, a sentence connecting it back to the supporting argument and then to the governing thought.

This agent eliminates the most common presentation failure: the data dump. Without it, you default to showing everything you know, in the order you discovered it, leaving the audience to assemble the argument themselves. With it, you are forced to have a point before you build slides, which means every slide exists to support that point or it does not exist at all.

The Three Conversations: a framework for difficult exchanges

In 1999, Douglas Stone, Bruce Patton, and Sheila Heen — researchers at the Harvard Negotiation Project — published Difficult Conversations: How to Discuss What Matters Most. Based on nearly thirty years of research, the book identifies a structural pattern that underlies every difficult conversation, from salary negotiations to relationship conflicts to performance reviews.

Their core insight: every difficult conversation is actually three conversations happening simultaneously, and most people only engage with one of them.

The "What Happened?" conversation is where you argue about facts, interpretations, and blame. You think you are right and they are wrong. You assume you know their intentions. You assign fault. This is the layer most people get stuck on — an endless loop of "here is what happened" versus "no, here is what really happened."

The Feelings conversation runs underneath the factual dispute. The reason the conversation is difficult is rarely the facts themselves — it is what the facts mean emotionally. You feel disrespected. They feel unappreciated. Neither of you has named these feelings, so they leak into the factual argument as tone, defensiveness, or passive aggression.

The Identity conversation is the deepest layer and the one almost nobody addresses explicitly. A difficult conversation is difficult because it threatens something about how you see yourself. Am I competent? Am I a good person? Am I worthy of respect? When a conversation triggers an identity threat, your ability to think clearly collapses — not because the topic is complex, but because your self-concept is under siege.

A Three Conversations agent for difficult exchanges has these fields:

1. What happened — their story. Before entering the conversation, write out the other person's version of events as charitably as you can. What are they probably seeing, feeling, and concluding? If you cannot articulate their perspective, you are not ready for the conversation.
2. What happened — your story. Now write your version. Notice where the two stories diverge. The divergence points are where the real conversation needs to happen.
3. Feelings inventory. Name the emotions this situation generates for you. Not "I feel that they are wrong" — that is a thought, not a feeling. Actual emotions: frustrated, anxious, hurt, embarrassed.
4. Identity threat. What is this conversation putting at risk about how you see yourself? Name it explicitly. Once you see the identity threat, it loses much of its power to hijack your reasoning.
5. Purpose and opening. What do you actually want from this conversation? Not "to be right" but a concrete outcome. Then craft an opening that describes the gap between your two stories rather than asserting your story as truth: "I think we see this situation differently, and I want to understand your perspective."

Stone, Patton, and Heen's framework does not guarantee the other person will cooperate. But it guarantees that you will enter the conversation having already done the cognitive work that most people try to do in real time, under emotional load, while the other person is talking. The agent front-loads that work so you arrive prepared.

The AI parallel: from cognitive agents to software agents

The frameworks above are cognitive agents — structures you run in your own mind. But the pattern extends directly to software. Large language models are, at their core, structured communication engines. They accept an input and produce an output according to patterns encoded during training. When you give an LLM a template — "write an email using BLUF format with these four fields" — you are instantiating the same agent pattern in silicon that this lesson teaches you to instantiate in cognition.

OpenAI's Structured Outputs feature, introduced in 2024, makes this explicit: developers supply a schema, and the model constrains its output to match that schema exactly. The model does not improvise the structure. The structure is pre-defined, and the model fills in the content. This is BLUF-as-code. It is the Pyramid Principle rendered as a JSON schema.

The parallel is instructive but the direction of influence matters. You should not outsource your communication agents to AI and consider the problem solved. You should build the cognitive agents first — internalize BLUF, the Pyramid Principle, the Three Conversations — and then use AI tools as amplifiers of structures you already understand. An LLM can draft a BLUF email in seconds. But if you do not understand what BLUF is doing and why, you cannot evaluate whether the output is correct, you cannot adapt it when the situation does not fit the template, and you cannot override the agent when context demands it. The cognitive agent is the foundation. The software agent is the accelerant.

Why frameworks work: cognitive load theory

The reason communication agents improve both speed and quality is not mysterious. It is a direct prediction of cognitive load theory, developed by John Sweller in the 1980s. Your working memory has a fixed and small capacity — roughly four chunks of information at a time. When you are composing a message without a framework, you are simultaneously managing content (what to say), structure (how to organize it), audience (what they need), and tone (how to say it). Four demands on a system that handles four chunks. There is no spare capacity for doing any of these well.

A communication agent pre-solves the structure problem. It removes one of the four demands entirely, freeing working memory capacity for content, audience, and tone. This is why a BLUF email written in two minutes is often better than an unstructured email labored over for twenty — not because you spent more effort, but because you spent your limited cognitive resources on the right problem.

George Miller's famous 1956 paper, "The Magical Number Seven, Plus or Minus Two," established that chunking — grouping information into meaningful units — is how humans manage cognitive limitations. A communication framework is a chunk. Instead of holding dozens of structural micro-decisions in working memory (should the context come before or after the request? how much background is enough?), you hold one chunk: "BLUF format." The framework compresses structural complexity into a single retrievable unit, exactly the way an agent should work.

Building your communication agent library

You do not need dozens of communication agents. Most people's communication falls into a small number of recurring categories, and three to five agents will cover the vast majority.

Start with these three:

The Direct Message agent (BLUF-based). Use for: status updates, requests, escalations, decisions that need sign-off, information sharing. The fields: bottom line, context, details, action required. Default to this for any written communication where the recipient needs to act or be informed.

The Argument agent (Pyramid-based). Use for: presentations, proposals, persuasive documents, any communication where you are making a case. The fields: governing thought, supporting arguments, evidence base, so-what cascade. Default to this when you need someone to agree with a conclusion.

The Difficult Exchange agent (Three Conversations-based). Use for: performance feedback, conflict resolution, boundary-setting, any conversation you are tempted to avoid. The fields: their story, your story, feelings inventory, identity threat, purpose and opening. Default to this when the emotional stakes are high enough that you are rehearsing the conversation in your head.

Each agent should be documented — written down, not just held in memory. As you learned in L-0411, documenting your agents is what transforms them from vague intentions into reliable infrastructure. Write the template. Name the fields. Specify when to deploy it. Then use it until it becomes automatic, at which point it has migrated from a conscious tool to an internalized skill — which is the ultimate goal of any agent.

Structure liberates content

There is a persistent misconception that frameworks constrain creativity. The opposite is true. Jazz musicians improvise brilliantly within chord structures. Poets produce their most powerful work within formal constraints like sonnets and haiku. Speakers who know their structure cold are the ones who can riff, adapt, and respond to the room — because they are not spending cognitive resources figuring out what comes next.

Communication agents work the same way. When the structure is handled, your full creative and analytical capacity is available for what actually matters: the insight in your email, the argument in your presentation, the empathy in your difficult conversation. You are not choosing between structure and substance. You are using structure to make substance possible.

The communication situations you face are not as unique as they feel. Emails, presentations, and difficult conversations each have a stable underlying architecture that has been mapped by researchers and practitioners across military, consulting, and academic domains. Build the agents. Document them. Deploy them. And redirect the cognitive resources you have been wasting on form toward the content that only you can provide.