Why We've Always Done It This Way is Dangerous in the Age of AI
Written by Aravind Sakthivel, Author & Technology Leader
Aravind Sakthivel is a global technology leader with 23+ years in information technology (IT) and artificial intelligence (AI). Founder of London AI Studio and former Chief Information Officer (CIO) in the operating companies of Veralto/Danaher. He now serves as a Fractional CIO/Chief Artificial Intelligence Officer (CAIO), helping CEOs, boards, and investors turn IT and AI into measurable growth engines.
Some of the biggest barriers to artificial intelligence have very little to do with artificial intelligence. They are old processes that nobody questions, reports that continue because they have always existed, approval structures designed around constraints that disappeared years ago, and management habits that gradually became organisational rules.
"We've always done it this way" rarely begins as resistance to change. Most established processes once solved a genuine problem. The danger comes later, when the conditions change, but the process survives. AI is making that gap increasingly visible because it does more than offer a faster way of performing existing work. In some cases, it challenges the reason that work was organised that way in the first place.
When experience becomes a constraint
Experience is valuable because it helps organisations avoid relearning the same lessons. Standards, controls and procedures exist for good reasons. But a successful solution can gradually become an unquestioned assumption.
Many business processes were designed around limitations that once seemed permanent. Information was difficult to retrieve. Systems could not communicate easily. Analysis required specialist teams. Integrations were expensive. Reports had to be assembled manually. Managers often had to wait for scheduled reviews before they could see what was happening. Those constraints shaped organisational structures, yet many of them are now changing.
Research from the Organisation for Economic Co-operation and Development (OECD) suggests that a significant proportion of workers in developed economies are exposed to generative AI, particularly in occupations involving information processing, analysis, administration and management. Exposure does not mean that jobs automatically disappear. It means that substantial parts of how work is performed may change.
That distinction is important. AI may affect organisations less by eliminating entire roles and more by changing how tasks, decisions and responsibilities are distributed. When the underlying constraint changes, the process built around it deserves another look.
AI changes the question
Most organisations approach AI with a practical question: how can this technology help us perform an existing activity faster? Can AI summarise a document? Can it answer a customer query? Can it reconcile information? Can it prepare a report? Can it analyse a contract or assist with software development?
There is growing evidence that these applications can improve productivity. Research by Erik Brynjolfsson, Danielle Li and Lindsey Raymond examined more than 5,000 customer support agents using a generative AI assistant. Productivity increased by around 14 percent on average, with greater gains among less experienced workers. Other controlled studies of professional writing have also found that generative AI can reduce completion time while improving average output quality.
Those findings are encouraging, but they create another question: what exactly are we making more productive? If AI accelerates a process that no longer serves a useful purpose, the organisation has not necessarily transformed. It may simply be performing unnecessary work more efficiently.
This is where the leadership question should change from "How can AI improve this process?" to "If we were designing this process today, would we create it this way at all?" The difference is significant. One question leads to automation. The other can lead to reinvention.
Do not automate yesterday
Consider the familiar management report. Data is extracted from several systems, moved into spreadsheets, reconciled, turned into charts, discussed by analysts and finally presented in a weekly or monthly meeting.
Generative AI can clearly accelerate parts of that workflow. It can summarise results, draft commentary and prepare presentations. But before automating the report, ask why the report exists.
Which decisions depend on it? Which pages actually change management action? Why is it produced weekly? Why do managers wait until Friday to learn something that happened on Monday? Which sections remain because somebody requested them years ago?
Once those questions are asked, a different operating model becomes possible. Instead of improving the speed of report production, an organisation might continuously monitor the indicators that actually matter, identify exceptions and route relevant information to the person responsible for acting. The objective changes from producing the report faster to improving the speed and quality of decisions.
McKinsey's research into organisations adopting generative AI has highlighted workflow redesign as an important factor in moving from experimentation towards measurable business value. Simply placing AI tools on top of existing processes is unlikely to capture the full opportunity. AI can create efficiency within a process. Redesign can create efficiency across the process.
From assistants to agents
This becomes even more important as AI moves beyond generating content and begins taking actions. Traditional automation usually follows predetermined logic. Something happens, a rule is triggered and an action follows. Generative AI introduced much greater flexibility because software could interpret and create unstructured information such as language, images and code.
Agentic AI extends this further. An AI agent can combine a model with software tools that allow it to pursue an objective, gather information, perform actions and respond to changing conditions within defined boundaries. That is a significant shift.
An AI assistant might summarise a supplier problem. An agentic system could potentially investigate the issue across approved systems, compare contractual terms, assemble relevant evidence, recommend an action and escalate the case if it exceeds agreed limits.
The National Institute of Standards and Technology (NIST) has highlighted that AI agents also introduce new security and reliability risks because systems capable of using tools can potentially take unintended actions if they are poorly governed or manipulated. The right question is therefore not simply where AI agents can be deployed. It is where greater autonomy creates enough value to justify the additional risk.
Start with value
One of the easiest mistakes in a technology cycle is allowing the technology to define the problem. Organisations begin by saying, "We need generative AI," "We need Copilot," or "We need agents." Teams then search for enough use cases to justify the capability.
A stronger approach begins with a business outcome. Where are customers experiencing friction? Which decisions are slow because information arrives too late? Which activities consume large amounts of employee time? Where do errors repeatedly occur? Which processes become more expensive every time transaction volumes grow?
Only then should the organisation decide whether AI is the appropriate solution. Sometimes generative AI will be the answer. Sometimes conventional automation will be cheaper and more reliable. Sometimes the real problem is poor data, fragmented systems or an outdated policy. Sometimes the best solution is simply to remove a process. There is little value in using sophisticated AI to perform unnecessary work exceptionally well.
Find the old constraint
A useful way to challenge an established process is to identify the constraint that originally created it. Perhaps five approvals were necessary because information was difficult to verify. Perhaps analysts manually combined reports because systems could not exchange data. Perhaps customers were transferred between departments because different teams controlled different information. Now ask whether the limitation still exists.
This is particularly relevant because AI increasingly affects cognitive work. OECD research shows high exposure among managers, analysts, software professionals and other occupations that rely heavily on knowledge. This makes the current wave of automation different from earlier assumptions that technology primarily affects repetitive physical work.
The sensible response is not to remove every organisational boundary. It is to check whether the reason for the boundary remains valid.
Decide where humans matter
Human oversight should also be designed according to consequences rather than applied uniformly. An AI system drafting an internal meeting summary does not require the same controls as one making a financial commitment, altering customer rights or operating critical infrastructure.
NIST's AI Risk Management Framework emphasises clear accountability, human oversight and continuous monitoring. This suggests that organisations should think in levels of autonomy.
In one process, AI may retrieve information. In another, it may recommend an action. In an environment with low risk, it might perform a reversible action within defined limits. Decisions with significant consequences may remain subject to human approval. The important point is to decide deliberately where human judgement creates value and where human involvement exists simply because the old process required it.
Measure outcomes, not activity
AI projects are often assessed by how much time they save. That is useful, but it is not enough. Suppose an AI system reduces the time required to produce a report by 80 percent, but nobody makes a better decision because of it. The process has become more efficient without becoming more valuable.
Metrics should therefore reflect the outcome the process exists to support. Depending on the context, that could mean faster customer resolution, lower error rates, reduced operating cost, shorter decision cycles, improved service quality or increased employee capacity.
The customer support research by Brynjolfsson and colleagues is useful because it measured issues resolved per hour rather than simply typing speed. The focus remained on the operational outcome. That is a better standard for AI investment.
Run an assumption audit
Leadership teams can apply this thinking without launching another major transformation programme. Choose one recurring process that consumes meaningful time or creates obvious friction and examine the assumptions underneath it.
Ask what outcome the process exists to create, why information moves through certain people, why approvals occur in a particular sequence and why ownership sits with a particular department. Then identify which constraints still exist and which have disappeared.
Could information move directly between systems? Could AI interpret material that currently requires manual review? Could an agent monitor for exceptions continuously? Could actions with low risk be automated while unusual cases are escalated?
Then compare the redesigned process with the existing one. Does it improve speed, quality, customer experience, cost or risk? If it does not, newer technology may simply be creating newer complexity.
The real legacy system
Businesses often describe old software and infrastructure as legacy systems, yet some of the most persistent legacy systems are assumptions.
"This needs a monthly meeting." "Finance has to prepare this manually." "A manager must approve every case." "Customers expect this process." "We have always reported it this way."
Any of those statements might still be correct. The problem begins when they are accepted because they were correct in the past. AI does not make experience irrelevant. It makes regular reexamination of experience more important.
Organisations should not challenge processes simply because they are old, nor automate them simply because the technology now allows it. They should force inherited assumptions to compete with current evidence.
Reinvention before automation
Digital transformation has shown how easy it is to modernise technology without modernising an operating model. Paper forms became electronic forms. Spreadsheets became dashboards. Manual processes became digital workflows, but the underlying logic often remained untouched.
AI creates the same risk at a much greater scale. An organisation can add a chatbot to a poor customer journey, generate an obsolete report faster or use an AI agent to navigate an unnecessarily complicated approval process. That may look innovative while changing very little.
The greater opportunity is to examine the process before introducing the technology. Ask what problem it solves, which assumptions shaped it and whether those assumptions remain valid.
The next time someone says, "We've always done it this way," the phrase should not automatically be dismissed as resistance. It should become the beginning of an investigation.
Why was it done this way? Which constraint created it? Does that constraint still exist? If the organisation was starting today, knowing what AI now makes possible, would it design the process the same way?
Sometimes the answer will be yes. But when the answer is no, the organisation may have discovered something more valuable than another opportunity for automation. It may have found an opportunity for reinvention.
The most important AI question may therefore be the simplest one: are we using AI to make yesterday's organisation run faster, or are we using it to design a better one?
Read more from Aravind Sakthivel
Aravind Sakthivel, Author & Technology Leader
Aravind Sakthivel is a global technology leader and entrepreneur with over two decades of experience in enterprise IT, AI, and digital transformation. He served as Chief Information Officer at Esko Graphics and now leads London AI Studio while advising as a Fractional CIO and Chief AI Officer. Aravind has delivered complex M&A integrations, global ERP rollouts, and cloud transformations while driving measurable growth and resilience for CEOs, boards, and investors.
