You and Me and the Bot Make Three to Write

How to finally get the right things written down

Have you ever noticed that nobody ever writes the right things down? Maybe you’re on a small team, moving fast, your intel is buried in Slack threads and a Google doc from last year that might still be accurate. Or maybe you’re at a bigger company with a playbook that’s always outdated, a wiki nobody reads and way too much time spent in meetings about documentation. Either way, it seems like the only amount of documentation in any company is the wrong amount.

The problem always starts with a team size of one. If you’ve ever been the only person doing your job (really like 5 jobs) in your company, why bother to write anything down? You know what to do and you probably won’t forget. Okay, maybe you might, so you write some things down. Usually your passwords…on sticky notes…on your monitor.

At some point, though, your work will outgrow you and someone else will join. When two people start doing the same job, documentation starts to grow by accident, not intention. You might attempt to “start writing stuff down,” but mostly your direct messages, emails and texts are the only artifacts of your thought process. What’s interesting is that this transactional documentation tends to focus on exceptions – the Venn diagram of things that are in one of your heads and the latest edge cases. Valuable intel, but never enough to create a full workflow, just enough to survive this week.

That all changes when the third person joins. “Oh shoot, June starts next week and has to be trained! What do we do? How do we do it? We don’t have anything written down?” So you sheepishly half ass some training program for poor June. Ideally, June is curious in the face of uncertainty (#1 trait for early stage success) and starts asking questions. They start out innocent enough, but then become unintentionally threatening to your ego as June’s outside perspective reveals all the holes in your operations. Your instinct might be to defend your apparent lack of thought and commensurate documentation: “We were too busy! Writing things down is boring! We tried but it was too hard!” The dangerous part is that these defenses are correct.

Busyness is real, and failing to prioritize your focus is lethal. Boredom too, but it might just be a mask that frustration wears. You probably did earnestly try to write things down but the complexity of your evolving operations eat you alive. That’s not laziness, it is what happens when you try to formalize something you are still figuring out. So rather than trapping yourself in ego defense, embrace the opportunity that a third person brings to really start figuring things out.

For years I’ve given this exact advice to teams: get to 3 people and documentation will happen. I’d even encourage small companies to create “frankenteams” of loosely related roles just to get them to the magic number. This only rarely worked because while teams of 3 might create the necessary demand for great documentation, just having three people does not overcome the very real effort required to make great documentation.

Enter the Robot

Fortunately, modern technology has given us a solution to the third-person problem: enter the chat bot. LLMs or Large Language Models (right there in the name, folks) are great listeners. Add one to a meeting and it transcribes, summarizes, and makes sense of what was said. Invite it to a conversation and it engages your “I have to explain this to someone” energy without the ego threat that a new teammate does. Either way you have your third person. But it’s more than that. Because unlike a new hire who absorbs your documentation and then never touches it again, the LLM is extraordinarily good at the actual writing part, maybe better than any real person (except Dave Garrett). It doesn’t procrastinate or find documentation boring. You talk, it writes. You decide, it records. So the AI doesn’t just replace the third person. It supercharges what the third person was supposed to unlock. And for a moment, it feels like you’ve finally solved it.

So off you go, happily generating all sorts of meeting notes, maybe those policies you’ve been dreaming about. With documentation easier to create than ever, you are likely to run headlong into the next problem: how do you find what you need when you need it? Despite more hours than I care to count spent on playbooks, folder structures, wikis, and search tags, everything seems to end up in a dusty binder on a back shelf somewhere, or its digital equivalent. I’ve never really seen someone retrieve what they need, when they need it. So we do what we have always done — we find the nearest person who seems to know and ask them. Reliable in the moment, maybe. But not repeatable, and that’s no way to run a scaled, multi-site operation. And that’s the easy problem! (We are going to solve this one, I promise).

Setting the very real retrieval problem aside for a moment, you are still barreling headlong into the unavoidable truth about your complex operations: some situations were never going to fit in a Standard Operating Procedure (SOP) in the first place. Take a sadly common scenario: spend enough time clinically and you will be harassed, threatened, and/or assaulted by patients. If you are brave enough to raise the idea of “firing” said patient, policies (if they exist) are checked and always found wanting. The result is a “let’s give them another chance/not ruffle feathers/this is part of your job” flavored decision by the administration, out of fear of retaliation from their own bosses. They are not entirely to blame--these situations are nuanced and some clinicians want to fire patients for being 5 minutes late.

The failure, then, is not in drafting documentation, it is attempting to manage dynamic, ambiguous complexity with a static document in the first place. Imagine if you could access the reasoning from every previous time anyone had discussed this or any other complex scenario. Not just the decision, but the why behind it. Documentation then is not a necessary evil to poorly convey “the rules,” rather it is to capture your hypotheses, so that you can learn to make better decisions when things inevitably go awry. The SOP was never going to save you. But the accumulated reasoning might have.

What if you could superimpose all the hard conversations your organization has ever had in one place? How much more consistent could your service be? Better yet, what if instead of having to retrieve what you need, read it (ugh), and then try to interpret it for your idiosyncratic situation, you could just… ask? I mean, that’s why I built Schutzduck, you could stop reading this article right now and just go chat about it. This is the end-around the trap of scaling by dumbing down that Seth Godin warned us about — hiring people who are almost-as-good-as-you to replicate something that only works because of you. What is on offer here is a different bet: instead of replicating a person, you preserve the reasoning. Venture Capitalists have re-named this institutional knowledge “context graphs” (because of course they f-ing did) so that they can sell you a trillion dollar opportunity. But what if I told you that you can start building your version of the solution for way less, right now?

So enter the robot again, albeit a different incarnation: the oracle. Tools like Google’s NotebookLM allow you to load in everything: policies, procedures, meeting notes, transcripts, etc. wrapped in a chat interface. Your team doesn’t search institutional knowledge. They converse with it. They ask questions and get answers that synthesize across every document and every conversation you’ve ever fed it, with citations! The logic of a decision is preserved and queryable even if it was never formalized into a clean SOP. This matters enormously in high-judgment roles like sales or clinical work, where your front line now has instant, reliable access to the reasoning required to resolve their unique situation consistent with the mission of your organization. This is the smart use of AI you’ve been looking for.

Power Armor, not Drone Army

The LLM-as-third-person analogy has one important limit worth naming before a skeptic names it for you. The LLM pushes back on logic and structure whereas a human third hire pushes back on context and lived experience (the stuff the LLM takes at face value because it wasn’t there). The AI won’t tell you your SOP is wrong because it contradicts what actually happened last Tuesday. A teammate will. Which just goes back to my preference to think of these tools as Power Armor, not a drone army. You still need humans on the team. The bot helps your small team punch above its weight.

Small Is Still Beautiful

None of this changes the underlying truth that small teams are faster, tighter, and more aligned than large ones — Brook’s Law hasn’t been repealed. What these tools change is the capability ceiling. You no longer need to hire a third person to unlock documentation or a fifth to cover a capability gap. The question shifts from “do we need another person?” to “can we get away with a better prompt?” And this is the major difference I’ve seen between the newest companies (founded post ~2023) and those even slightly older: the latter are still asking how AI can make their processes more efficient and reduce headcount. The former are building something different entirely, and you can too.