We often talk about “technical debt” on our podcast and in this newsletter. Technical debt is shorthand for all those slow, bulky software systems and tools that you’ve been living with for years that are just too expensive to replace. 

Let’s coin a new phrase: Procedural debt. Procedural debt has nothing to do with technology. It’s all those lousy, inefficient, bad habits that your company lives with today. You’ll find it in government, non-profits, for-profits—just about any organization that’s been around for more than five years. One thing about habits is if you enact them enough times, you get good at them. And when you get good at something, you’re much less inclined to change.

Software that’s rushed through organizations often doesn’t help—instead, it just sort of formalizes and further acknowledges procedural debt. Tools can actually be a bad-habits enabler, further entrenching less-than-ideal ways of working. 

Good software makes things work a little faster. Great software changes how you work. When people talk about “enterprise software solutions,” that’s the actual hope and promise, and why it costs so much money. It will change the way you work. It will abolish your technical debt—and your procedural debt.

Historically, implementing enterprise software solutions would involve hiring a ton of consultants and paying a lot for software licenses, and then, over 36-96 months, you’d roll out the new tools—slowly, with lots of training and painful migrations.

Today, we’re headed for a world of AI, and the general vibe is that this will be replaced—all those consultants and coders—with “agents.” They really love their automation on the West Coast, and they think typing into a box is too much work; the next wave is just telling an LLM to do things for you, autonomously, and report back. Agents can roam around the web, and also around your data and workflows, then suggest improvements, fix your grammar, tighten your legal language, and let you know that inventory is running low. They can do this all day and night.

More ambitiously, they can autonomously execute multi-step, complex processes that used to take all sorts of orchestration and coordination in the past. No more running that approval form down the hall! And they’re getting smarter and better every day.

Now, if this was a more traditional marketing newsletter, I might say: Agents are a powerful tool for paying down your procedural debt. You’ll be able to spin up agents to enact processes, and they’ll accelerate everything you do. Procurement? Agents. Governance? Agents. Compliance? Agents.

But I don’t believe that. Agents are amazing at producing artifacts; they’re autonomous and incredibly motivated to work all day. And this makes them the ultimate enabler of procedural debt. They have no interest in taking a step back and offering up ideas about how to change how you work. They’re not going to challenge assumptions or revisit the real underlying conditions that led to that oddball 18-step approval process.

You are going to be sold the idea that agents can do everything better and faster, and make you pretty reports at the beginning or end of the week. But the truth is that agent-based software is so easy to make, so fast to deploy, and so useful that I expect it to codify and enforce really bad processes. It’ll encode that approval process. It’ll keep that PDF-based workflow intact. It’ll read through the thousand resumes without really thinking about why you have so many generic submissions instead of a few useful, targeted pieces of inbound with thoughtful cover letters.

I would like to propose a different approach: Let AI observe your work. Call it an “AI Advisor.” Don’t let it try to fix anything. Describe how you do things, upload the PDFs, share the government forms. Let it present you with a deck on how you can do better—what process could be simplified, what apps could shorten workflows.

We use agents in our products, we love the technology, and we love this new ecosystem. But we also believe that how people work together is…well, human. Our processes tend to be there for a reason, and it’s important to know and understand those reasons before you engineer them away. LLMs can’t understand, so that’s still on us, at least for a while.