Biggest lesson from 2025: systems are messy because people are messy (and that needs to be OK).
If you spent any time at all on LinkedIn in 2025, you couldn’t be blamed for believing we have entered the golden age of perfect data and intelligent systems. The narrative from enterprise AI vendors was relentless: clean your data, unify your back end and the algorithms will deliver frictionless, automated bliss.
I saw this firsthand when I visited AWS in Austin, Texas in September — the agentic promised land. It was a compelling vision.
Photo by Google DeepMind on Unsplash
But after spending the last 12 months in the trenches working with businesses locally and globally — from agile startups to established enterprises and everything in between — I’ve walked away with a very different lesson.
The reality of business is messy. And that’s perfectly fine — it needs to be.
The biggest lesson I’ve learned this year is that the pursuit of clinical, perfectly systematised workflows is often a trap. We need to stop treating human idiosyncrasies as “bugs” in the system and start seeing them for what they are: the glue that actually holds operations together.
The myth of the unified back end.
It goes without saying: every CTO will tell you they want a “Single Source of Truth”. In practice, I have rarely seen a back end that isn’t a jigsaw puzzle of legacy databases, modern SaaS tools and Excel spreadsheets doing some very heavy lifting.
We are taught that this fragmentation is a failure of strategy. We are told that teams aren’t following Standard Operating Procedures (SOPs) because of a lack of discipline. But often, that fragmentation exists because it simply works for the speed at which the business is moving. There are only so many people to do so many things, and ensuring every new data pipeline aligned with every old one was simply too time consuming to validate — instead, the team committed to finding a way to make it work.
New software is onboarded to solve a specific, immediate pain point. It rarely integrates perfectly with the ERP implemented five years ago. Ecosystems evolve organically, not architecturally. When we try to force strict standardisation over the top of this organic growth, we often break the very agility that allowed the business to survive in the first place.
And here’s the thing — unless they are reaching a decisive inflection point, these businesses are surviving. They are making do.
Systematisation and the individual.
One of the most fascinating observations from 2025 is that systematisation works incredibly well at the individual level, but often falls apart when scaled.
I’ve worked with operators who have incredibly complex, unique workflows. To an outsider (or a systems architect), it looks like chaos. But to that individual, it aligns perfectly with their personality and the “wiring” of their brain. They know exactly where that file is. They know exactly which client needs the high-touch email and which one just wants the invoice.
When we try to flatten these individual nuances into a rigid, standardised process for the sake of “optimisation”, we often see a drop in performance — even feelings of alienation in the employee. We strip away the cognitive shortcuts that experienced staff have built up over years. We trade effectiveness for uniformity, and it’s rarely a welcome trade.
The “garbage in, garbage out” fallacy.
Enterprise AI businesses love the phrase “garbage in, garbage out” because it puts the onus on the client to have perfect data before they can unlock value.
But here is the truth the tech giants ignore: businesses have always found ways to cater for garbage.
They do it through human logic. They do it through manual effort. They do it by working long hours to bridge the gap between what the system says and what reality is.
I’ve seen countless scenarios this year where a human “fixer” steps in to interpret messy data that an AI agent would have rejected. They apply context that doesn’t exist in the database. They make a judgment call based on a phone conversation they had three weeks ago or because this is what happened that one time in 1979 or simply because “that’s the way it’s always been done”.
If we wait for perfect data before we start our digital transformation journey, we will never start. The systems we build need to be robust enough to handle the mess, or at least smart enough to hand it off to a human who can.
The psychology of being needed.
This is a whole other conversation for someone smarter than I, but bear with me. In my view, the most overlooked aspect of the 2025 tech conversation is the psychological need to work.
Tech leaders and consultants (myself included) are often obsessed with productivity and efficiency. If we link X to Y so that Z happens whenever A, B or C are triggered, Geoff will have so much more time to do the important stuff. We look at a manual process and think, “I can automate that, save you 10 hours a week, and you’ll be happier.”
But we often miss the fact that people take immense pride in their work — even the manual, “inefficient” parts. Sometimes that is the important stuff. There is a deep satisfaction in being the person who understands the complex workaround. There is value in feeling needed, in being the only one who can untangle the knot. There is a boost to Geoff’s ego when he gets that annoying call on holiday to help with something because he is the only person on Planet Earth who knows how to do it.
When we automate everything, we risk stripping away the sense of purpose that many employees derive from their daily grind. A perfectly optimised role can feel sterile. If a machine does 90% of the thinking, the human is left wondering, “What is my value here?”
Revolution or evolution?
So, where does this leave us as we head into 2026?
The lesson for me is that the human element remains the single most critical factor in successful digital transformation. We need to stop aiming for all out revolution, the tear-down-and-rebuild approach that promises a utopia of perfect automation.
Instead, we need to embrace a phased approach with the human at its centre. We need to build systems that include room to take a step back if needed. We need digital transformation strategies that account for:
Messy data: Building workflows that don’t break when the input is imperfect.
Human nuance: Allowing flexibility for individual working styles rather than strictly enforcing rigid conformity.
Emotional buy-in: Understanding that automation should respect the user’s need for purpose.
Systems are messy. People are messy. And as we continue to integrate AI into our businesses and, especially, as agents do work for us, success will come from designing resilient, human-centric architectures that can survive in it.
Have a safe and happy holiday period. See you in 2026.
