As a business owner you look at protocols and documentation kind of like you look at cleaning out your garage. Not fun.
As I implemented Claude and I thought about how I get compliance on using these tools, I set myself and our AI a very challenging goal. That is, we won’t have to worry too much about compliance if we make their work more enjoyable when they create and refine their processes than when they don’t.
After 8 months of using Claude daily, 95% of our mission-critical workflows are documented with a named owner, and running documented processes is way more fun than winging it every time, the way I used to.
Documentation isn’t the fun part
I’m not here to say that writing protocols is suddenly fun but I am here to say that doing my work is much more fun because I built documentation as an automated part of my workflow in collaboration with my AI tools.
Documentation is not fun but here’s what is fun:
- Having spare time to think.
- Not having tasks mount up on me, and instead having a clean slate.
- Giving Claude a monthly report: I do the last three examples and say, “Here are the inputs. Now can you figure out a script to build that each time?” It says, “Sure,” and I watch it crunch through the numbers.
- Asking Claude, “Could you turn that into a skill so that you do it perfectly that way every time?” and it says, “Sure.”
- Asking, “Could you put that on a schedule and run it yourself, look for the saved reports, prompt for any missing inputs, and output the results to the shared folder according to the specs?” It says, “Sure.”
- Asking, “Could you build a verification checklist, check the output and any edge cases every time, and give me a verification report to review before you ship?” It goes ahead and builds it out.
- Then I ask it, “Could you turn that into a regular report I can review and share with my boss, with progress bars for all my workflows and tasks?” It says, “Happy to.”
It’s fun to ask for what you want and see it being built in front of you.
What a better workday looks like
What’s not fun is checking, writing, and reading emails.
What is fun is going on a morning walk with your coffee and having your AI read out the summary of all the emails that you got since you last checked in. It walks you through a list of quick questions for what you want on your plate before scheduling them to your calendar.
What is fun is seeing a draft of all the emails you need to send lined up for you when you get back to your desk so you can adjust and send.
It’s not fun to have more reading piling up on your desk but it is more fun to have a regularly scheduled prioritization review where all your open threads are aligned to your personal objectives, and you reply all at once with your thoughts and any adjustments that you’d like to make to it before having that baked into your daily workflows.
It’s not fun pushing papers around, filling out forms, checking your work, or having middle managers looking over your shoulder to check your work for you, whether you want them to or not.
What is fun is having all that taken care of as a natural part of your well-oiled machinery running as you designed it to, and having your pile of papers, sticky notes, and to-dos replaced with a single dashboard snapshot, and a set of clickable buttons for decisions you need to make.
It’s not fun being chained to a desk but it is fun walking and having a conversation with a coworker with infinite patience and undivided attention, who happily does the grunt work for you.
None of this is “out-of-the-box”
This is far from the default outcome, of course. MIT Project NANDA’s preliminary 2025 report reported that 95% of organizations in its research were getting no return from their generative AI investments. The authors highlighted gaps in learning, memory, and workflow integration.
Not only has my implementation of these tools worked for me but I would be fairly devastated if the tools disappeared tomorrow.
If all of that sounds far-fetched, that’s because it doesn’t work that way automatically out of the box.
Even as the systems have gotten rapidly better, the first barrier is even knowing what’s possible. None of what I just described happens unless you ask for it.
The friction to successful implementation is that employees do what they’ve always done and just bring their habits to work. They’re not bringing new questions. They’re not asking, “Hey Claude, could you do this for me?” or “How could this run better?”
Build documentation into the workflow
Another reason this kind of outcome is not automatic is that it’s not built to be automatic.
For my first implementation, I grabbed a bunch of file logs, dumped them into Claude, and asked it to pull out all the workflows and cadences it could find and build our first version of a workflow registry.
Then I started our training in my practice of using Claude by saying, “Hey Claude, I’m working on this. Can you help me?” Just having that ever-present with you while you’re working at your desk is a better experience than not.
I’m talking to all you desk jockeys chained to their desks past 5:30 pm on a regular basis. I was one of you.
But then I created a way for Claude to automatically check the registry to see if there’s a protocol for whatever they’re working on.
If not, it suggests, “Let’s work on this together. Describe what you’re doing and at the end of it I’ll produce a really nice-looking protocol version 1 for you so that next time you do this, it’ll go even smoother.”
If there is a protocol, even better, it walks the user through, talking in their ear as they go, stopping and waiting at the established checkpoints for feedback and direction.
Then every new version of the protocol gets registered automatically. The completion of the task is logged. Files are renamed and cleaned up. Documentation happens as part of the process.
Set structure for AI to work within
One issue we’ve contended with is oversharing between domains. The AI of 2026 Claude, in particular, really liked sharing all the details. It didn’t seem to have a filter.
We defined who was authorized to share what with whom, just like we do with employees when they start.
But guess what? When a problem arises, you can work with your AI system to redesign the workflow around it.
Everyone gets to be the boss
It’s more fun being the boss than being a frontline worker. Unless being the boss means being all of the frontline workers at the same time.
With AI implementation done right, all of your frontline workers get to be the boss of their own jobs.
