Most Jobs Are Just Legacy Code
A short argument for redesigning roles before you automate them.
Here is a quiet truth about most organizations: the majority of roles and processes are not designed. They accumulated. Someone built a workflow fifteen years ago, added steps to it over time, and now nobody remembers why half of those steps exist. Technology changed, regulations changed, customer expectations changed — but the work itself mostly stayed the same.
That is why so many AI projects feel underwhelming. Companies skip the redesign and go straight to automation. They ask: "How do we make this process faster?" when they should be asking: "Does this process still make sense?"
The customs broker example
Take customs brokerage. To many people, it looks like a specialist role that requires years of experience, deep product knowledge, and judgment. And in some cases it does. But a large portion of the day-to-day work is HS code classification: looking up what code applies to a product, checking whether it falls into multiple categories, and selecting the most accurate one.
This is not a judgment problem. It is a structured search problem. Build a robust HS code lookup that can identify products correctly and suggest alternatives when a product fits multiple categories, and you have removed a huge slice of repetitive work. The broker does not disappear. They move up the value chain: handling exceptions, complex rulings, disputes, and client relationships. The automation does the part that never needed a human in the first place.
The mistake is to assume the role is a single unit that must either be automated or preserved. Most roles are bundles. Some parts belong to a human. Some parts belong to a system. The opportunity is in separating them cleanly.
The wrong kind of automation
I once saw a team build a tool to auto-fill PDF forms for CBSA submissions. It saved time compared to typing the same information by hand. But the real problem was not typing speed. The real problem was that CBSA already accepts electronic data submission, and the team was still treating the PDF as the source of truth.
They automated a workaround instead of fixing the workflow. A better question would have been: "How do we submit this information directly to CBSA and skip the PDF entirely?" The same pattern shows up everywhere: tools that auto-fill templates instead of capturing clean data, chatbots that answer repetitive questions instead of fixing the underlying confusion, dashboards that report problems instead of preventing them.
Another common version: building templates to collect missing information from customers, when the real fix is designing the intake workflow so incomplete entries cannot happen in the first place.
Automation applied to a broken process just produces broken output faster.
A simple rule for what stays human
Not everything should be automated. My rule of thumb: if an action can lead to two very different outcomes, and the right choice cannot be resolved by historical data or an agreed protocol, a human should be involved.
For example, choosing whether to terminate an employee is not something you hand to an agent. There is no clean dataset that tells you what the right call is. It depends on context, risk, values, and judgment. But gathering the relevant documents, checking policy references, and drafting a neutral summary of the facts? That can absolutely be delegated.
The goal is not to remove people. The goal is to stop wasting human attention on work that does not need it.
What to do with this
Pick one role or process in your organization and ask three questions:
- Which steps exist only because of an older system or older rule?
- Which decisions are truly judgment-based, and which are just lookups or matching?
- If you were designing this workflow from scratch today, what would disappear?
You will usually find that the work can be split into smaller pieces than you assumed. Some pieces go to AI. Some pieces go to humans. Some pieces go away entirely. That decomposition is the real value — not the tool you buy to speed up the old process.
The organizations that get ahead in the next decade will not be the ones with the most AI tools. They will be the ones that looked honestly at how work is structured and redesigned it for the technology that already exists.