AI Won't Save Your Broken Communication

Image Credit: Adobe Stock/vanzyst

“There’s a huge gap between what AI can do and what AI can do well.”

When I heard Sawyer Middeleer say this on the Chief of Staff Network podcast with Scott Amenta, I said “YES!” out loud in my car. And I kept saying it for most of the show.

I spent over 15 years living deep in the world of legacy code modernization as the CEO of Corgibytes. Teams would bring us in because the software was slow, brittle, or impossible to change. Almost every time we dug in, we discovered that the code wasn’t the real problem.

The code was actually the most faithful record of how the business process was broken.

That’s a big part of why we advocated against hard cutoffs and blank-slate rewrites. Revising code with intention on a compassionate, communicative, cross-functional team is one of the most effective ways to improve your operations.

Code problems are mostly operations and communications problems in disguise. If you only focus on the deployable aspects of your software, you’ll never achieve organizational effectiveness.

AI raises the stakes on this. If you have a bad process, automating it won’t fix it. You’ll just run what’s broken faster and at a larger scale. And, if your tooling creates too much abstraction from the source code, you’re also eliminating those durable artifacts that make it easier to find what’s broken so you can fix it.

Being excited about new tools isn’t new either. Each wave of technology tempts us to skip looking at the problem because the shiny new tooling tempts us towards only focusing on a narrow solution. That limits innovation and wastes valuable talent and time.

What I loved about Sawyer’s approach is how clear he is on the order of operations:

1. Pick a specific process, like resolving a support ticket, building a board deck, or triaging bugs.
2. Map the value it creates, not just the time it creates. Some work ties neatly to ROI, but some work, like strategic planning, doesn’t. An extra hour there might just be the most valuable hour of the entire quarter but you can’t necessarily put a neat metric on it.
3. Test AI against the people doing the work, and judge the results on quality as well as volume.
4. Where the output looks basically the same, let AI take it. Where a person brings something the AI can’t, keep the person.

He also named a failure pattern that I see fairly regularly, which is measuring everything by time saved. Sure it’s faster, but the quality is worse, and slowly the product, marketing, and customer relationships wear down into slop. You’ve probably experienced this yourself and just didn’t have a name to it.

That’s why mapping the value chain from end to end matters so much. Before you automate anything, you have to answer a harder question. Does the process we’re automating match the problem we have?

If you can’t answer that, AI can’t either.

Thanks, Sawyer and Scott for the great conversation. It’s well worth the listen: