The Manager's Old Problem: Nobody Finishes Reading Pre-Meeting Materials

Anyone who has led a team knows this scenario well: there's a design review meeting on Monday, so on Friday evening the tech lead posts a thirty-plus-page technical proposal document to the group chat with a note saying "everyone, please take a look when you have time." Come Monday morning, the only person who has actually read it page by page is usually the document's author. The business stakeholders stare blankly at the architecture diagrams, finance frowns at the technical jargon, and the boss starts checking their phone by page three.

The problem isn't that the document is poorly written—quite the opposite: a proposal that runs to thirty pages is usually full of solid substance. The problem is that the information isn't layered: decision-makers need to know "choose A or B, how much each costs, where the risks are, and when it can go live," but the document buries these conclusions under layers of background, technical comparisons, and implementation details.

Managers compressing the material themselves also pays a price. Doing it personally costs an entire evening and easily introduces personal bias. Delegating it to a subordinate means the summary's quality depends on that person's communication skills and standpoint. Rushing it right before the meeting turns the one-pager into a mere formality.

This step happens to be one of the things AI is currently best at taking over: reading long documents thoroughly, reorganizing the information from the decision-maker's perspective, and outputting a consistently structured summary. Below is a workflow our team has been running for the better part of a year.

Workflow Overview: Three Roles, Four Steps

We break this pipeline down into three roles:

• Tech lead: provides the original proposal document and supplements key constraints (budget caps, timelines, non-negotiables);
• Manager (decision-maker): defines "what decision I need to make based on this one-pager"—this is the most overlooked input step, and the one that most affects quality;
• AI: handles the reading, distillation, structured output, and risk review.

The four-step process is as follows:

Step 1: Define the decision question. The manager writes one clear sentence stating what needs to be decided in this review, for example: "Within budget X, choose build vs. buy, decision needed within two weeks." This sentence goes directly into the prompt and serves as the criterion for what the AI keeps or cuts during compression. Without it, the AI's summary becomes an "all-encompassing recitation of the table of contents," which is far less valuable.

Step 2: Feed in the document and do the first-round compression. Give the complete proposal to the AI and use the prompt template below to generate a one-page decision summary. For long documents that exceed the context limit, feed them in segments: have the AI produce key points for each segment first, then merge and compress.

Step 3: Adversarial risk questioning. This is the step that best reflects the manager's perspective. After the first-round summary is done, issue an additional round of instructions asking the AI to review the summary in the role of a "devil's advocate": Which conclusions lack sufficient evidence? Which risks are downplayed? Which costs weren't accounted for? In this step, the AI can often dig up pitfalls that the document glossed over.

Step 4: Human finalization. The manager spends ten minutes verifying the numbers and conclusions, removes any occasional over-generalizations from the AI, and posts the final version to the group. Key numbers must be checked against the original document—the AI occasionally misattributes references during compression, and this one minute cannot be skipped.

A Reusable Prompt Template

你是一位资深技术管理顾问，擅长把冗长的技术方案压缩成供决策者使用的一页摘要。

【决策问题】{一句话说明这次要拍板什么，例如：在预算X内选择自建还是采购}
【关键约束】{预算上限 / 时间节点 / 不可妥协项}
【原文档】{粘贴或分批附上技术方案全文}

请输出一页决策摘要，严格包含以下板块：
1. 一句话结论：建议选哪个方案，核心理由（不超过50字）
2. 方案对比表：3个候选方案 × 维度（成本区间、上线周期、维护负担、扩展性）
3. 主要风险清单：每条注明影响程度（高/中/低）和文档中对应的缓解措施，若文档未提及缓解措施，标注「未覆盖」
4. 待决策事项：列出文档中未定、需要评审会上拍板的开放问题
5. 被压缩掉的关键信息：列出你判断为「实现细节、不影响决策」而省略的内容类别，供我确认

要求：
- 所有数字必须来自原文档，禁止估算或补充；找不到的信息写「文档未提供」
- 语言面向非技术决策者，术语首次出现时用一句话解释
- 总长度不超过一页A4

Item 5, "key information omitted during compression," was added after many rounds of iteration—it makes the AI's trade-off process transparent, so the manager can quickly confirm that "what was left out really shouldn't have been included."

Before and After Using AI

Efficiency: In the past, producing a solid one-page decision summary took the tech lead two to three hours to write, plus another hour for the manager to revise. Now the AI produces a draft in minutes, and human verification and finalization stay within half an hour. Preparing meeting materials has gone from "a Friday-evening burden" to "a Monday-morning routine."

Quality: Human-written summaries often carry a bias—the person who wrote the proposal tends to highlight its strengths. The AI outputs according to a fixed template, where the "risk list" and "not covered" annotations are mandatory, resulting in something closer to a neutral "audit perspective." Combined with the adversarial questioning in Step 3, many problems that used to surface only during execution now get raised at the review meeting.

Collaboration: The biggest change is where the meeting discussion starts. Before, the first twenty minutes went to "aligning on what the document says"; now, the meeting opens directly with the list of open questions. The tech lead also gets relief—they no longer struggle with the dilemma of "write it in detail and nobody reads it, or keep it brief and get challenged," since the full document is preserved and summaries are generated on demand.

Risk control: The boundaries of the AI approach deserve honest acknowledgment. The AI can occasionally misplace numbers or be overconfident in its conclusions, which is why the process mandates a "human number-checking" step. For content involving security compliance or external commitments, the summary serves only as an index, not a paraphrase. In one sentence: AI handles compression and structuring; humans are accountable for the conclusions.

Cost and Tool Selection

This workflow requires "long-context reading + structured output" capabilities, which all flagship models on the market can handle. If the documents don't involve sensitive content, cost-effective models are sufficient—check each platform's official pricing page for specifics. When rolling this out to a team, we recommend routing everything through a single API gateway to make it easier to control usage and costs (billing governance for that step is a whole other topic).

Final Thoughts

A manager's time should be spent on judgment, not on the manual labor of "condensing thirty pages into three." A well-defined prompt template plus a fixed verification process can ensure that every design review meeting starts with a one-page decision summary that people actually read.

If you're building your team's own AI workflow and need a stable model API gateway, check out this platform: https://api.thistoken.ai/register

---

Every example in this post runs with a single API key — get yours at https://api.thistoken.ai/register and start in minutes.