As I continue to explore AI, I’ve been thinking about the question “where is AI slowing me down?” I’ve watched a few companies go all-in on AI this past year, trying to automate everything they could get their hands on. I’ve also watched half of them get stuck. They didn’t actually understand the process they handed off in the first place.
David Sinkinson wrote about this in his newsletter, When AI Slows You Down, and puts into words something worth thinking about.
Here’s what I took away:
- Somebody still has to make the call. Automating a process doesn’t make the hard decisions go away. It just changes who’s making them and when, usually to a version of you with less context and less time, dealing with it after something’s already gone sideways.
- Do it manually first, a lot, before you automate it. You have to run a process by hand enough times to understand its edges, the exceptions, the weird cases that don’t fit the pattern. Skip that step and you’re automating a guess.
- Sometimes the answer is “don’t automate this yet.” It’s common to start building automation around something and realize halfway through that parts of it don’t automate well at all, or that the whole problem needed to be rethought first. That’s not a failure, that’s the process working the way it should.
- A confusing process stays confusing at any speed. Handing something off to a tool doesn’t sharpen it up. Whatever wasn’t nailed down before still isn’t nailed down, just moving quicker now.
- Complexity automated is scaled complexity. If you try to automate or scale something that is complex and inconsistent, you get more complexity. Garbage in, garbage out.
A tool is really good at doing exactly what you tell it, which is a problem if you were never that precise about what you meant in the first place.
At CRE OneSource we often say: People, Process, Technology… in that order! You cannot talk about technology if you don’t have the process nailed down precisely. And you can’t nail the process if you don’t have the right people.
Now I’m thinking “Do I actually understand this well enough to trust something else running it?” instead of “Can this be automated?” Most of the time the honest answer tells me exactly where to start. Value before autonomy.
Charlie Coppola
When AI Slows You Down
The teams that automated the most aren’t always moving fastest. Here’s the failure mode nobody talks about:
The Automation Slowdown
Adopting AI was supposed to compress timelines. For a lot of teams, it did.
But there’s a pattern emerging on the other side of early adoption that nobody’s really talking about yet.
Some of the companies that moved fastest on AI are now moving slower than they were before. Not because the tools stopped working. Because of what happened when they did.
The Over-Automation Trap
Here’s the thing: when AI makes execution cheap, the temptation is to automate everything you can get your hands on.
So teams do. They build workflows for content, analysis, reporting, outreach, research — you name it. They chain agents and automations together until the whole process runs without them. It feels amazing. Look at us go.
Then a decision point shows up that the automation wasn’t built for. An edge case. An exception. A situation that needs context the system doesn’t have. Basically, the real world.
And suddenly the team is slower than they were before, because now they have to override a system, diagnose why it produced the wrong output, and manually correct a process that used to be (you know) a human making a call. Oof.
The automation didn’t remove the judgment requirement. It just buried it under a layer of process. And it turns out unearthing buried judgment is a lot more expensive than just exercising it in the first place.
Speed Requires Clarity, Not Just Automation
The teams moving fastest with AI aren’t the ones that automated the most. They’re the ones who were most deliberate about what they automated.
There’s a meaningful difference between:
-
Automating execution inside a well-defined process
-
Automating decision points that were never clearly defined in the first place
The first one makes you faster. The second one creates technical debt in your workflow that compounds every time something breaks, the situation changes, or a customer does something unexpected.
Clarity has to come before automation. If the human version of the process was fuzzy, the automated version will be faster and fuzzier. Congrats, you’ve scaled the confusion.
What Slow Actually Looks Like
The tricky part is that at first, it feels like productivity.
Outputs are shipping. Dashboards are updating. Reports are generating. The system looks like it’s running.
But underneath that, the judgment calls that used to happen in real time are now backed up. Someone has to audit the automation. Someone has to manage the exceptions. Someone has to decide when the system is wrong and what to do about it. That “someone” is usually your best people, quietly drowning in a queue nobody’s tracking.
That overhead is invisible until it isn’t. And by the time it shows up, teams have often built more automation on top of a foundation that was already fragile. Now you’re not just cleaning up one bad automation — you’re cleaning up the three you built on top of it. Cool. Not.
The Fix
Before automating any workflow, the question worth asking isn’t “can AI do this?”
It’s “do we understand this process well enough that a system following our instructions would produce the right output in every scenario we’re likely to face?”
If the answer is no, the automation will eventually produce the wrong output. And the cost of catching and correcting it will eat right through the efficiency gain you were counting on.
Chris and I had a version of this rule at AppArmor. Before we automated anything — sales collateral, RFP responses, customer onboarding — we had to be able to describe the process end-to-end on a whiteboard in a few minutes. If we couldn’t, we didn’t have a process. We had a habit. And habits automate badly.
The companies compounding the fastest right now are treating AI deployment like engineering work. They define the process first, they identify where judgment is genuinely required, and they automate the parts that are actually repeatable and unambiguous.
That’s slower up front. It’s considerably faster over 12 months.
The Bottom Line
AI doesn’t make you faster by default. It amplifies whatever’s already there.
Clear processes get faster. Unclear ones get more complicated.
The teams struggling with AI right now are struggling because they automated before they understood. And now they’re managing the outputs of a system built on a foundation they skipped.
Speed is a byproduct of clarity. AI just makes the absence of it a lot more expensive.
You got this. But before you automate the next thing, take twenty minutes and see if you can actually explain the process end-to-end. If you can’t, that’s your real first project.
