Lawyer Lin Shanglun’s article: Impressive BD—the product being sold is AI integration capability

Business development in the AI era: if you can evolve in an end-to-end way based on the pain points of your own team, you can do it even if you’re not a full-time developer. This kind of mindset should be something every position must be prepared to have.
(Background: Lin Shanglun, a lawyer’s featured article》Do professionals who understand AI become winners in the AI era? )
(Additional background: Lin Shanglun, a lawyer’s featured article》From the Cai Agga incident, fear of AI: you’re not angry at AI—you’re afraid of being outdone ）

Table of contents

Toggle

• The thing he built actually hits the pain points pretty well
• Talking about how he puts AI into it
• But I think the Output can be even more
• From a lawyer’s perspective, there’s something that must be done: bring hallucinations under control
• A few functions I think his BD Agent could add
  • Automatically generate contract revisions from meeting minutes
  • Adjust the strength of the contract depending on negotiation positions
  • Compliance review—block risks upfront
• Finally

At a tech expo, I met a business development (BD) manager from a company that’s about to become a major one. But more specifically, he personally rolled up his sleeves and built a set of internal systems for their company’s BD using AI. He was very generous—he pulled out his phone and showed me his prototype. We just chatted while looking at it.

Because I usually also work on AI development, and because I’m a lawyer, when I looked at his work I couldn’t help but think a few layers deeper. This article is basically me整理 the conversation from that day—what I saw, and what directions I discussed with him that could be continued to play with.

The thing he built actually hits the pain points pretty well

Right from the start, he mentioned a very real problem: their BD team’s outside development workload was so big that there was no way for a supervisor to digest everyone’s daily reports. What supervisors could give were only very rough instructions—they couldn’t lay out precise tactics for each store, for each location. And after BD went out to execute, it was very easy for people to bump into each other’s prospects. Synchronizing everyone’s progress was also extremely inefficient.

So he built an app. After logging in, you get an admin interface with many layers. The most core layer visually maps all the development points opened externally by BD and the deal status. The benefits are very direct:

First, everyone no longer needs to sync verbally with each other. Before each person kicks off a new round of development, they can see at a glance what others have done and what stage each location has progressed to. Second, from his perspective as a supervisor, he can directly see the entire pipeline—where each store is stuck at a glance.

Honestly, you can tell at a glance that this was made by someone who really understands BD. The logic of that interface isn’t something engineers could come up with out of thin air—it’s clearly designed based on having run business out in the field.

Talking about how he put AI into it

I asked him out of curiosity: where exactly is the AI used? He said he connected an AI that整理s into a database all the knowledge BD uses for external development, past contracts, and cooperation case histories.

He gave a very practical example: sometimes BD develops into chain brands, but the way to negotiate with chain brands is completely different from negotiating with individual stores. Chain scale is large and their influence is strong—there are often利益 exchanges to be made. But in training, regular BD often can’t explain clearly how “what level the other side is” matches “what leverage we can offer in response.” So he delegated that part to the AI to answer. I think this entry point is spot on.

I asked him how he built the database. He was honest that it’s basically deployed in a Google Sheet, reading it directly with JS. But he emphasized one key point: for each BD, what they looked up and what the AI returned—everything is recorded, everything is traceable. He also wrote a “merchant definition” in advance: the AI runs a definition first to judge what type a new customer is, and then goes back to compare with cooperation cases.

There’s one small detail I especially want to mention. I told him that the free version of Gemini Flash has a pitfall—token limits might cause results to be less than expected. He had already thought of this. He pre-planned every question’s path and hard-coded it: for each type of problem, it only reads the specific pages needed. This not only prevents missing information, but also saves time and tokens. This approach—don’t cram everything in; judge first and then retrieve the needed data—really gets the point.

But I think the Output can be even more

After getting to this point, I told him something pretty blunt: “Your thing is truly great, and it really fits the way you use it. But—there isn’t enough of an AI flavor.”

I’m not picking on it. I just think there’s still room to play. What he’s doing in essence is turning BD’s past work tracks—things that were opaque, uncontrollable, and hard to get back—into something transparent, visible, and manageable. The input side has already been done solidly.

But from a developer’s perspective, I think he’s stuck on output. Right now, his output is almost entirely concentrated in the ChatBox and in organizing information. ChatBox is definitely a good thing, but I feel his output could go to more places—for example:

First, directly generate company-format documents, like MOUs and reports—rather than making people read the conversation and then organize it themselves. Second, use different interfaces and workflows to handle different tasks, instead of stuffing everything into the same chat box. The underlying thing is still the LLM, but you can design different Agents for different scenarios—one task, one workflow.

From a lawyer’s perspective, there’s one thing that must be done: bring hallucinations under control

I talked to him about this from a lawyer’s perspective. Because what his system touches—contracts and commercial terms—has less tolerance for errors than people generally think.

My personal habit is: AI-generated output must not invent anything out of thin air. So I suggested he add two things:

One is citations and traceability. Add a citation behind every output result; when you click it, you can see which file and which section the statement comes from. This is actually especially important for him—if BD uses the negotiation talking points generated by the AI to fight on-site, and they can’t verify the source afterward, then they’re taking a risk with unverified information.

The other is lowering the temperature. In scenarios that require precision and can’t be allowed to run wild, you should suppress the model’s randomness. With these two working together, the space for hallucinations becomes much smaller.

A few functions I think his BD Agent could add

Their company has a legal department and lawyers, and most contracts use standard templates and don’t change much. It sounds like AI has no room to do anything, but I actually think this kind of “looks stable” scenario is exactly where there’s something to play with. I casually chatted with him about a few directions:

2. Automatically generate contract revision drafts from meeting minutes

After BD discussions, the most tiring part is turning two hours of meeting talk into formal contractual clauses. AI is good at this: take the standard template, then dump the entire meeting minutes into it—it generates a new version based on the conclusions. Sections that were deleted or modified are marked with tracked changes, so the legal team can immediately tell what was changed.

3. Adjust contract strength depending on negotiation positions

I think this one is the most interesting because it can connect directly to the logic in his system. The problem with a standard template is that it can’t reflect the difference in negotiation leverage each time. When negotiating with a big company, if our position is weaker, the terms should be softer; when negotiating with a smaller counterpart, if our position is stronger, the terms can be harder. No contract is one-size-fits-all—each one should be customized. If his Agent can automatically generate versions with different strengths based on the “size of the other side,” wouldn’t that tie right into his system’s idea of “matching a level to the right weapon”?

4. Compliance review—block risks upfront

BD-facing external content often carries compliance risks. AI can help check sentence by sentence: which words are too exaggerated, which rules are violated, and whether to delete something or add a disclaimer. For a BD team that runs outside every day, this kind of “block it before it goes out” capability is actually pretty useful.

I also reminded him of one more point: if the company has a few clauses it particularly wants to keep, it’s not a conflict at all. The wording can change, the tone can change, but in substance retaining our rights and interests is fully achievable technically.

Finally

After chatting that day, I left with a deep impression. He spent three weeks, and he even deliberately used his own budget—before the prototype was done, he didn’t touch company resources. He turned a real pain point into a usable product, and employees were already using it, and the boss also nodded in approval.

For more than two months, he’s been doing the same thing: find pain points, and find moments where AI can be plugged in. I think that’s the mindset someone who wants to use AI should have—not “AI is cool, let’s try it,” but “This is painful—can AI come and stop the pain?”

In the end, the most useful AI usually isn’t the flashiest one. It’s the one that understands business best and is closest to real pain points. Seeing a BD person like this start thinking through his own work using AI logic, I actually find myself quite looking forward to what he’ll build next.