No matter how slick a business may look from the outside, there is chaos lurking. Workflows have a tendency to start off real simple, with complexity sneaking in over time, before it becomes a snarly mess of move this file to here, click this button then, restart that process, export here for there, type of thing, over a wide range of different systems from different vendors. It’s ugly, it’s messy, and most of time it is not documented.

This sounds familiar?

Grace starts every morning logging into the order system, reset the password because it always expires overnight, run the report, wait for the spinner, export the CSV, rename it to satisfy a naming convention nobody remembers inventing, then upload it into a second system that rejects it because someone put a comma in a product description. Fix that, reupload, hunt down a shared drive folder that three people have renamed without telling her, drop in a second file, and only then does she open her actual inbox. Nobody designed this process. It just accumulated, one export at a time, like a haunted house built entirely out of workarounds.

Every company has a version of this somewhere in some department. For all our modern day achievements our world is held together with nothing more than duct tape and bubblegum. Business leaders are getting all excited that this is suddenly going to be resolved by AI. Just point Claude at it, and Grace can go back to doing her normal job (and secretly hoping they don’t need Grace at all and get rid of another salary cost).

It’s naive at best and dumb at worst. You are not going to sit down and prompt “fix what Grace does”.

This is where AI can make the biggest impact on business, and everyone gets to keep their job, but one has to temper expectations and realize this is not a quick process.

Before you deploy a single AI agent to the job, you need to figure out what the problem is, break down each step into very small linear discrete units of operation, each with a logic gate to determine what success looks like. This is a standard business workflow process that we have been producing for decades.

Except we don’t.

We are relying on the muscle memory of Grace and because she has been doing it so long, she is pretty much on auto-pilot as to what she needs to do. Every quirk, failure, or misalignment, she has seen it, fixed it and not even thought about it. Is it documented? LOL of course not.

PwC’s US CEO Paul Griggs said this recently in a Business Insider interview: “If you’re layering it on top of messy processes, all AI is going to do for you is tell you how bad your messy process really is“.

AI has a little more tolerance to do things that traditional workflow engines struggled to do. It can make certain decisions on the fly if a gate doesn’t quite close properly, but it needs to still have a clear definition of what it is attempting to achieve and if it has completed it or not.

An LLM is fundamentally non-deterministic (because at heart it is just a sophisticated random generator), which means you ask it to reword or perform an operation, it will be different each time. It might be subtle, or dramatic, but it will be different.

Support Grace

Grace’s workflow here, if pushed to AI, maybe up to 10 different agents, running different models with different prompts/constraints, all working together, as they hand-off, with a human-in-the-loop gate somewhere to make sure they done what they were supposed to.

AI is like another employee, they need trained, they need guidance, they need a manual/playbook to which to operate to be successful.

The forums and news are filled with failed AI projects, no real ROI to point to. AI is a failure, the headlines say. Boards, PE partners, management are all rushing to throw AI at a problem they don’t understand. The reason they fail, is because they were never setup for success; no clear objective of precisely the problem they were looking to solve.

Do not underestimate what your Grace is doing to keep your organization healthy each and every day. She needs support and help. This is where the real ROI can come from and free up Grace to go after other areas that need her super power.

Start your AI project small, look for the real annoyances, and don’t bite off too much. You will fail.

Baz Luhrmann’s “Everybody’s Free (To Wear Sunscreen)” makes a point worth sitting with here. The real trouble in your business is rarely the thing keeping you up at night. It’s the thing you stopped noticing years ago, because Grace quietly handles it every single morning before you’ve had your coffee. Give the song a listen, it holds up both professionally and personally

Start your AI project small, look for the real annoyances, and don’t bite off too much. Or you will fail.