I just wanted to be able to team up with an old coworker again for fun times and work towards something great. It’s never as simple as “We verbally agree to be cordial like we always have been.” because things change, feelings can shift, money can change things, and you want the friendship to survive because it’s worth more than all that. So we needed a contract, and the contract had to protect against state laws, business laws, precedents, lawsuits, employee tests, someone getting screwed over or not appreciated, or feeling under appreciated or not valued. Neither of us had any experience in that legal landscape, and neither of us could afford to hire anyone who did. In lieu of that AI was introduced, and I thought we were making awesome progress and WE WERE! But I also realized it had become two LLMs building a bullet proof contract between two strangers, not two friends.
It started in early September, the first night we walked through actual deal terms. Claude explained the state contractor test, that a contractor doing the core work fails most of it, and that some of the exposure could land on me personally. I stopped asking about mechanics and typed “How is this so fucking hard with the law, I just want to be able to share success with a friend, and talk about the company together to help it get better shooting ideas around, why is that impossible to achieve”. Claude’s answer was “Because the law cannot tell a friend from a scheme,” and then “the statute has no box for ‘we like each other.’” This is great, I was and am really frustrated that the business world has no avenue to keep friends in the process of ventures, and the law does not distinguish intent, only fact…
Two minutes earlier I had asked whether the contractor shape was safe at all. Claude said the state does not go looking at a two-person company, someone has to raise it. “In a two-person company that means [my friend] files a wage claim after a falling-out … So the realistic trigger is the friendship ending badly.” This was reassuring and helpful putting things in perspective. But the idea was something I could see taking off if we do it correctly, and I have no idea if money will change us or not if we ever make it to that future. I like to prepare, but how do you do that without offending a friend, and in an arena you know nothing about?
The first full proposal came out at 32 pages that same day. On the twenty-eighth my friend wrote back that the friendship mattered more than any of it, and that working all of this out now took away stress that could come later, then asked for time to read the real operating agreement before making a choice about ownership that could not be undone. So the rounds kept going. By the round that started October fifth the proposal had grown from 52 pages to 58, with a 24-page change table alongside it, and one question from my friend had added 18 entries to that table. This was a nightmare of never ending plugging holes that never really existed, we were planning for every world event instead of planning for the operating paths we wanted to work within. Neither is the right answer, but the business world doesn’t define a happy medium where everyone comes out content.
That night I asked Claude “How does the contract read, honestly, no holds barred. From a corporate, friend, and just perspective.” Part of what came back was “He asked one question about one sentence and received 58 pages and a 24-page table. Four rounds have each said ‘I think this might be the last of it.’ A friend starts to wonder whether the document is for him or for the drafting.” This was true, and I felt it. My friend would simply ask questions about certain clarifying terms or suggestions to be made, and when I got into the weeds of things I found large glaring contract holes everywhere I impulsively found the need to fix, and Claude was all too happy to do so. At this point Claude had already turned this into a legal wall of text and I was in the habit of simply tagging along for that ride and poking holes at it like a logic problem. I didn’t realize the wear and tear it could be having on the other side, and that worried me when I realized it.
I asked for a rebuild with a 20-page target and one condition, “Without reducing coverage, or anything that has been cemented so far (Because going backwards at this point is not ideal), reduce the language without creating new holes and issues (Hard I know).” It came back at 29 pages. It was an impossible ask of an LLM, which when looking at a legal document will see everything as important and very little that can go, and it was disheartening and frustrating to confront what I had created at that time. I asked one more question, “I am kinda surprised, most employment contracts aren’t event 12 pages, how did we get more complicated than that in a contractor agreement with profit sharing”. Claude said the contract had become several documents in one, and of the contractor agreement “On its own it would be the twelve pages you expected.” Then “Here no statute fills anything in, so every number and every anti-gaming rule has to be written.” I asked “Why cannot they be chained but separate contracts so the terms of one changing don’t constantly shift the other”. This was when I realized that I was on a never ending spiral, I set out to make sure that I could never hurt a friend, and all I was doing was giving him walls and walls of text to look at each conversation with no end in sight.
The review rounds on that rebuild turned up about 150 findings, and fixing them kept opening new edge cases. Claude kept rewriting the file after every message, and I lost it, verbatim, “JESUS CHRIST WOULD YOU STOP EDITNG AFTER EACH MESSAGE AND JUST TALK THROUGH SOMETHING”. A couple of minutes later I shut it down. “We will need to pick this up tomorrow. At this point I think you have essentially poisoned the well, we may need to start from scratch with the frankenstien you have created harvested for IDEAS to build upon. And this time, with simplicity in mind.” This was my breaking point that night before I went to bed. I set out to work with a friend and had created a monster without realizing it when I wanted a safety net. And the tool I trusted to help make my ideas into a reality failed me, and I failed it in how I approached the task. Every time I watch Claude epically fail at something it is like staring into a mirror, because it can only do what I tell it I want, and only with the resources and instructions I provide, so its end results that dissatisfy me are typically a direct reflection of my inability to get to the goal I desire.
The next morning I opened with “Okay, so.. it is morning, I have a fresh perspective. Lets TALK through this, no writing anything.” This was my redemption, my comeback. I spent all night tossing and turning dreaming about this stupid conversation, waking up every hour only to forget what the perfect prompt would be to get Claude to do what I wanted. I know that sounds deranged and desperate but that is how frustrated I was. It was my responsibility to correct my mistakes, and I had to get Claude aligned with my goals and in a state of mind to help facilitate what I want, not solution the next output. What we landed on was v4 draft at 7 pages, containing everything that I thought mattered, talking through everything we could, deciding what should change and what we can wait on with negotiation clauses. Overall, I went from v1 to v4 in 24 hours of complete and utter structure and content changes in the process to try and make it right. I think v4 is the right path at distilling everything I think is meaningful into 7 pages, but I still don’t fully know the answer. That afternoon I wrote my friend. “I went back to what the contract was ever for. Two things: we can work together without the state deciding you are my employee, and you cannot get screwed once real numbers are involved. Everything else in those 58 pages was one of us asking an LLM ‘what could go wrong’ and it answering with a paragraph (At least that’s what I was doing…).”
I sent both versions, the 58 pages and the short one. “It came out at seven pages. I think it is okay for the two of us to build some of this on trust rather than spelling every branch out, but if you want it explicit the whole way down that is equally fine with me, it will just read long.” Then, “A few things in the short version are different from the 58 pages, and some of them are not in your favor, so here they are first.” I never want to sugar coat something that is obviously disadvantageous for the other side, but again, this is a friend. It is hard to tell a friend that you are putting in constraints to actively work against their interests for either your own or the company’s, that is why friendships and businesses rarely mix, but I like to think that we are different and we can meander through it all. Neither of us is cut throat, both are empathetic and compassionate and respect one another. I think that if we can get past the contract theory and land on something we can both accept then we can get through any rough patches we may come up against. The email closed with “Nothing is final and I know I have just been arbitrarily changing things each email as I come up with shit but absolutely feel free to simply say what you want or don’t want, I will take zero offense!” Honestly, my friend is TOO nice, he would let me take way more than I have any right to without batting an eye or raising a complaint, which is why I tried to be as fair as I can in the contract, and try to be more generous in certain areas than most business contracts would be. Because with a friend, sometimes he won’t stand up for himself, and a real friend would stand up for him, even if they didn’t come out on top in something in the process.
One of the changes I listed was “Cause” for ending his role, which now covered publicly known conduct that would hurt the company’s standing with the people it serves. I told him “Claude didn’t like the ‘[my friend] is like a Teddy Bear and won’t hurt a fly’ reasoning I guess.” There are some things that I really hated to write into that contract, things like removal for cause, or publicly known conduct that would hurt the company. That sentence sucked to write and I resisted writing it in a lot of ways. Claude only thinks in risks, it can emulate emotions in conversation but at the end of the day it will always hedge on the side of caution, and it made a valid point. Even if my friend never did anything wrong, and I doubt he ever would, if he was associated or involved with something that he couldn’t control that could reflect back and tear down the social standing of the company, what would we want to do? The answer is I still don’t know, contracts are binding and it’s hard not considering everything.
I think I found the issue, I think I have a solid proposition for the solution, I have sent everything off and will see what he thinks. Contract theory is a nightmare and I hope I haven’t burned him out in all this. It has been a learning process, and learning by fire in some regards, but overall learning is good, and if it leads to being able to work with him again it will be worth it.