Part VII: Too slow to benefit
Why internal friction kills AI before it delivers anything useful.
This is Part 3 of 3 in the Execution & Pitfalls section of the 10-part Pragmatic AI Thinker series. We’ve covered the problems with chasing the wrong use cases and treating AI like a one-off project. Now we’re going deeper: why even good AI efforts stall inside organizations that simply can’t move fast enough.
TLDR;
Your AI isn’t blocked by the model. It’s blocked by your systems.
When execution breaks down, it’s rarely a tech problem. It’s the organization around it that can’t move, decide or adapt fast enough.
Step 1: Request access. Step 2: Request to request access.
It’s the environment
You’ve got a working AI, but nothing’s changing. Why?
I’m not going to tell you AI is overhyped or underhyped. That’s not the point.
What I’ve seen over and over again isn’t that applying AI is the problem, it’s everything around it. The blockers are internal: unclear systems, slow feedback, risk-averse culture and outdated decision structures.
You can plug a smart system into a frozen org and get… nothing. Or worse, you get output no one can use, trust or act on.
I saw this at a large mobility enterprise. The organization was already slow, held back by sluggish software and human systems. I watched teams try to build AI on top of scattered systems, copying data between tools, no clear source of truth and no shared trust in what was real. The ambition was there. But the foundation wasn’t.
Why it gets stuck
Here's what stalls progress even when the AI is technically working:
Hierarchy and redundant reporting lines. By the time approvals happen, the window for action is gone.
Lack of experimentation. Every idea needs a deck, a steering group or a sponsor.
Overwhelm. AI feels big, messy and intimidating. So nothing happens.
Slow systems. Legacy tooling, broken handoffs and no automation.
Bad data. Poor data quality and unclear ownership means the output is questionable before the AI even starts.
None of this is about the capability. It’s about the conditions it walks into.
I’ve seen this even inside companies known for innovation. At Spotify, some areas moved fast but others were slowed down by layers of manual processes and cautious cultures. You couldn’t experiment freely without long delays and approvals. Innovation wasn’t blocked by tech. It was blocked by inertia.
How to make it work
The teams I’ve seen make real progress don’t just build good AI, they clear the runway for it.
They reduce internal latency with clear scope, fewer layers and fast feedback. They test in smaller loops, skipping the decks and pitch theatre in favor of shipping and learning. They invest in the boring stuff: better data pipelines, tighter interfaces and clean handoffs between data producers and consumers. And they make AI part of the process; not a layer, not a lab. Just embedded.
It’s not about speed
I’m not saying move fast and break things. I’m saying: move smart and reduce the drag.
When you reduce friction, AI starts working the way it’s supposed to: in the background and inside workflows, improving decisions and shortening loops.
It’s not magic! It’s the result of systems designed to evolve.
Closing thought
AI won’t fix a slow organization. But it will expose one.
And once it does, you have two options: Redesign for speed. Or get left behind.
📩 Enjoyed this? Subscribe for more pragmatic AI insights straight to your inbox.