There’s a comforting version of AI adoption where the technology sweeps in and makes everything faster. Leaner development, quicker delivery, less friction. It’s a good pitch. It runs straight into a wall in most organisations, and the wall is the technical foundation nobody wanted to deal with.
AI can’t fix a broken deployment system. It can’t repair shaky infrastructure or untangle undocumented dependencies. What it can do is find all of them, quickly, and usually in front of an audience.
The debt doesn’t go anywhere
Significant technical debt, immature pipelines, systems held together with workarounds that have quietly become the system. None of that disappears when AI arrives. It becomes the first thing the AI hits.
That’s not a failure of the technology. It’s the accumulated result of a stack that’s been neglected, deferred, and worked around for long enough that the workarounds are load-bearing. AI doesn’t create that situation. It just makes it impossible to keep ignoring, because it can’t move through a broken process any faster than the process allows, and when it tries, the cracks show.
How it shows up
The signs are familiar to anyone who’s worked in delivery. Lead times that have been high so long nobody questions them. DevOps initiatives that were attempted and didn’t stick. Continuous delivery that exists in some parts of the system and not others. Testing environments that don’t match production. A technical debt backlog with items on it that have been there for years, for reasons everyone has stopped asking about.
Then a pilot works beautifully in controlled conditions and falls over the moment it meets the full production system. That’s not bad luck. That’s the foundations announcing themselves.
Why AI makes it worse before better
Adding AI to an immature pipeline doesn’t accelerate delivery. It adds traffic to a bottleneck that was already the constraint. There’s research showing AI coding tools correlating with drops in throughput and stability in teams with weak delivery foundations, which is the opposite of what the tool was sold to do. The tool is fine. The pipeline it’s pouring into can’t take the volume.
What this actually asks of you
The uncomfortable part is that the fix isn’t quick. Some technical debt carries timelines measured in years, not months. You can’t AI your way past it, and pretending otherwise just moves the failure downstream to a more expensive moment.
The firms making real progress are treating AI as the reason to finally deal with the foundations. The ones treating it as a reason to work around them faster are building on the same ground that’s failed them before, and they’ll find out the same way, just quicker this time.