Start from one real job

Put existing materials to use in the next professional delivery.

The customer suddenly needs an evidence-based proposal, opinion, or research summary. Materials are scattered across old files, meeting notes, and past communications. AI can first organize the relevant content into a draft; the scope of materials, applicable conditions, and final conclusions are still verified and confirmed by people familiar with the business.

You don't need to build a big system first. Start by clarifying what will be used this time and which content needs human sign-off, then see which preparation tasks are worth handing to AI.

Bring a piece of work to the conversation

A preparation draft that is easy to verify

ConfirmedThe problem the customer currently needs to solve, from page 2 of the meeting notes.

For referenceThe approach in the old solution applies to similar situations, but current constraints must be confirmed first.

To be confirmedWhether the latest rules have replaced the 2025 version; no final confirmation record has been found yet.

AI handles organizing and drafting; the owner returns to the materials to verify and makes the final judgment.

First, answer a common question

Is uploading files into an enterprise knowledge base enough?

No.You also need to know which version is valid, who has permission, how the materials relate to one another, and whether key statements in answers can be traced back to the source. AI prepares first; the owner verifies and makes the final judgment.

ScopeVersionSourceConfirm

Continue: Why do enterprise knowledge bases and RAG still give wrong answers?

First, put the materials in the right place

Even with the same amount of material, the way to prepare is not the same.

Someone needs to find that sentence they remember seeing faster; someone needs to put several materials and applicable conditions together; someone needs to distinguish the latest version from historical practice; someone needs to see in a draft where each key statement comes from.

Professional work inherently requires people to make trade-offs and confirmations. Only by arranging materials, rules and confirmation points clearly can AI truly take on the preparation work.

Having to dig through materials all over again every time

You remember that key information was written in meeting notes, old proposals, or customer emails, but when it's time to actually deliver, you still have to dig through files from memory. AI will also retrieve a few relevant passages, but misses the most critical constraints or context.

What to do first:Clarify important information along with its original location and the project or topic it belongs to, so it can be retrieved by question.

Every document is correct, yet the conclusions don't connect

The current requirements are in the project records, past approaches are in old proposals, and the conditions that can't simply be copied are in yet another supplementary note. Laying several documents side by side doesn't mean you've shown whether they can be used together.

What to do first:Put the relevant materials, applicable conditions and intermediate evidence into the same verifiable workspace, rather than letting AI fill in the blanks directly.

You found the basis, but are not sure whether it can still be used now

The same project has an old plan, a revised draft, temporary exceptions, and the latest confirmation. Experience that worked in the past may not apply today. AI doesn't inherently know which version takes priority.

What to do first:See clearly where materials come from, when they were formed, what scope they apply to, and who should confirm when conflicts arise.

The answer flows smoothly, yet no one can explain what justifies it

In professional work, writing fluently isn't enough. You also need to know whether the original text supports the statement, whether contrary materials were omitted, and whether it's a fact, an inference, or a judgment still pending confirmation.

What to do first:First have the AI deliver a reviewable draft: key statements can be traced back to the source, and anything without supporting evidence is clearly flagged.

The same piece of work will step by step raise new requirements

Not everything needs to be built into a complex system.

First timeStart by producing a source-backed first draft that can be checked manually.

Second timeNo need to re-explain background, definitions and criteria, or format — you can reuse approaches that have already been confirmed.

As materials growFind the materials that truly apply this time, not just similar-sounding words.

When multiple people work togetherKeep reviews, handovers and version changes in the work, not in someone's head.

Some work only needs one clear instruction; some work requires organizing materials first; only when it truly happens repeatedly, involves multiple people or is prone to errors is it worth continuing to turn it into a reusable workflow.

How Andy Ren Xu participates

Don't rush to pick tools, first do this piece of work once.

Bring a piece of work you repeatedly have to deliver, the related materials, and the results you're getting now. Andy will work with you to clarify: what materials are needed this time, what can be organized by AI, and what must be confirmed by the person responsible.

Then decide whether to first organize materials, improve search methods, fill in the relationships between materials, or organize validated effective practices into a long-term usable workflow.

Define this delivery firstClarify what is to be delivered, who uses it, and which content requires human judgment.

Define the scope of materialsIdentify which materials are usable and which need updated versions, conditions, or original-text support.

Complete one small section togetherFirst produce a reviewable result so the owner can check whether it matches actual work.

then decide whether to scaleOnly when needed do you continue to handle collaboration, permissions and long-term maintenance.

First conversation: establish the starting point first

What will you get first?

A narrowed scope of workClarify the inputs, the expected results, and the parts that AI should not decide on its own.

One material scopeKnow which materials to review first and which content to leave out for now.

A few key confirmation pointsSee clearly where versioning, facts, external commitments, and human judgment need to happen.

A next step that can be checkedTry a small piece first, or make clear that it isn't worth further complicating right now.

For one-off low-risk work, or a problem that a single prompt can already solve, there's no need to build more just to "use AI." What's truly worth investing in is the work that happens repeatedly, always requires looking up, always risks being done wrong, and always depends on one person remembering how to do it.

Start with one piece of work

Come with one specific delivery; you don't need to decide on the technical solution first.

Three sentences are enough to prepare: what you are currently delivering, what materials you need to find and organize each time, and which item must be confirmed by a person. Also bring a sample or a materials description that does not involve customer confidential information. Andy will work with you to decide whether to first organize materials, improve search methods, fill in the connections, or organize the approach into a complete workflow.

The first time, clarify the delivery scope and verifiable starting point. Content involving legal opinions, medical judgments, investment conclusions or formal external commitments should still be reviewed and confirmed by the appropriate owner.

Come to the discussion with one specific task, relevant materials and current resultsYou don't need to understand the technology first, or prepare a complete set of requirements.Contact Andy and describe this workYou can start like this:
"I am working on ____, each time I need to prepare ____, and what I most need to confirm is ____."

Want to understand the research method behind it

If you care about "why several materials don't connect," you can read IndexRAG original paper. It discusses a method for organizing in advance the connections scattered across different materials.

The point of this article is not to make you choose a particular technology, but to help you first smooth out the work that keeps recurring. Technical details will be judged later, around specific materials and ways of working, when they are truly needed.