Build to Be Replaced

Last week Anthropic shipped ten ready-made agent templates for financial services work: pitchbooks, KYC screening, valuation reviews, month-end close. JPMorgan, Goldman, and Citi were all in the room at the launch. One 400-person hedge fund says every employee uses Claude Code. If you are an analyst watching that unfold, the fear is not irrational: white-collar work is being packaged into reproducible workflows. The difference from the thousands of consultants and agencies already positioning to 'AI-ify' organizations is that this time the source is entering the market directly. Anthropic is shipping managed services and implementation support alongside the templates.

The Instinct Is to Grip Tighter

So analysts are doomed? Maybe. They are likely the first of a long line of expert operators to follow. The fear is real. The instinct is to grip tighter. Make the spreadsheet more complicated. Gatekeep the dashboard, one-off data sources. Become the only person who knows where the bodies are buried in the data model. Insulate yourself. Build a moat out of process knowledge.

The problem is, that does not work. It has never worked. AI just makes the failure faster.

The Operators Who Last

The operators I respect, and learned from, have always done the opposite. They solve a problem space. Then they build for their own succession through people, automation, and now AI. Then they move to the next problem space. Sometimes that next problem space is at a new company. More often it is in the same building, in a different role. The reason most people misread this philosophy is that they hear "work yourself out of a job" and assume the next move is a resignation letter, and sometimes it is, but more often it is a conversation with the same manager or executive in the same company.

At Zappos, I started running email. One channel, one coordinator, one set of problems. She and I optimized the channel and built it into something she could own, develop, build her own team around. Then we moved on supporting more divisions, automation, lifecycle, reporting, analysis... Same company, different problem spaces. The pattern has repeated across every role since. That is not accidental. It is a design choice I adopted early on in my career.

When I resigned my first position after 6 years in the same small engineering office, moved 6,000 miles (~10,000 km) to start a new life, his parting words were: "I wish you lots of success. While it won't be easy to find a replacement for you, everybody is replaceable." [paraphrased: the actual conversation was in Swiss-German] He was no-nonsense, one of the most principled individuals I have encountered to date. His words stuck with me, not as a warning but in the spirit of how they were delivered. 'Do your best, provide value, make sure succession is ensured, and you will always find places that need you.'

The pattern is the point.

Workflows, Not Jobs

Now apply it to AI. The workflow-versus-job distinction is what most of the fear discourse misses. AI replaces workflows. So did the junior hire, the offshore team, the SaaS tool, and the consultant before it. Every wave of operational change has compressed the labor cost of a workflow. The white-collar professionals who lost ground were the ones whose entire identity was a single workflow they refused to let go of. The ones who kept moving up were the ones who treated their current workflow as a problem to be solved and then handed off.

What AI changes is the speed. The hand-off used to take quarters. Now it can take weeks. That is uncomfortable, but it is not new.

The mental-trap is mistaking the workflow for the job. The job is to find the next problem worth solving. The workflow is what you build to free yourself to do that. People who confuse the two end up defending the workflow when they should be moving past it.

What Does Not Get Replaced

This requires a different relationship to work. You have to be willing to lose the comfort of being the only one who can do a thing, give up the illusion of being the only solution to a problem. You have to learn to like the moment after the hand-off, when you have nothing on your calendar except the open question of what is next. That moment is uncomfortable for a lot of people. It still is for me sometimes. It is also the most valuable moment in an operator's career.

What does not get replaced is experience, pattern recognition, and judgment. The ability to look at a business and see which problem is worth solving next. Which technology will work and which will not. Which person on the team is ready for more. Which workflow is a moat and which is just chasing the next shiny thing. That ability comes from doing the work, getting your hands dirty, watching what works and what does not, and accumulating the kind of taste and judgment no model can shortcut.

The operators who thrive right now are the ones plugging AI into the execution layer and themselves into the problem-finding layer. Not because they are more technical than everyone else. Because they figured out a long time ago that workflow ownership was the wrong thing to grip.

The Two Strategies

If your career strategy is to be the only person who knows how something works, AI is going to be your replacement.

If your career strategy is to solve the problem and move on, AI is the biggest advantage right now.