What I stopped saying in career talks

I used to answer career questions too quickly.

A student would ask whether to choose research or industry, a large company or a startup, graduate school or a job. I knew the usual considerations and could give a polished answer. The answer often described the options better than it helped the student decide.

After repeated talks at universities across several countries, I began removing advice that sounded complete before the person’s actual constraint was known.

I stopped saying “follow your passion”

Passion can identify work worth exploring. It is weak as a complete decision rule.

Students often have several interests and little evidence about the daily work behind them. They may love machine learning coursework and dislike running experiments. They may enjoy solving algorithmic problems and dislike the incentive structure of a particular research path.

I now ask for the activity, environment, and rate of learning the person wants. Which work would they willingly repeat? Whose feedback would make them better? What can they try before making an expensive commitment?

Interest should produce an experiment. The result can then inform the choice.

I stopped saying “go where the smartest people are”

Talent density matters. Access matters too.

A student can join an exceptional team and receive narrow work, weak feedback, or no view of the decisions that shape the product. Another team may have less prestige and give the student direct responsibility with a strong reviewer.

I ask who will inspect the work, how often, and what the student will own after three months. Impressive people nearby matter when the role creates a working relationship in which their judgment reaches the student.

I stopped treating prestige as a proxy for option value

A well-known institution can open doors. It can also delay the work a person actually wants to test.

Option value depends on what the person will learn, produce, and be able to demonstrate. A recognizable name with generic responsibilities may create fewer useful next choices than a smaller role with a shipped system and a credible owner who can explain the person’s contribution.

I still include reputation in the decision. I no longer let it stand in for scope, mentorship, or evidence of work.

I stopped asking for a five-year plan

Long plans often force false precision onto choices with weak information.

I prefer a shorter claim: after this role, which decision should become easier?

A research assistantship may reveal whether the person likes open-ended investigation. An infrastructure role may test whether they enjoy operating systems under production constraints. A startup may reveal whether broad ownership energizes them or scatters their attention.

The next step should buy information as well as experience. A path can be coherent without being predictable.

I stopped telling people to keep every door open

Meaningful work closes some doors for a while.

The attempt to preserve every option can produce shallow commitments: a little research, a little interviewing, several clubs, and no work carried far enough to change the person’s capabilities.

I ask which uncertainty deserves a real test now. The commitment should be long enough to receive hard feedback and short enough to revise if the evidence changes.

Keeping one decision reversible is useful. Keeping every identity provisional can prevent a person from learning what sustained ownership feels like.

I stopped trying to tell a perfectly clean story

My own path includes research, large-company engineering, quantitative finance, startups, teaching, and competitive programming. A clean retrospective can make each transition look planned.

It was more useful to preserve the decision record. What was I trying to learn? Which constraint had become limiting? What evidence made the next move reasonable at the time?

Students sometimes reject a good opportunity because it complicates the narrative on a résumé. The work should produce the story. The story should not overrule the work.

The decision record I use now

When someone brings me two options, I ask them to write one page.

The table does not choose automatically. It makes disagreement useful. A parent may be optimizing stability. A professor may be optimizing preparation for research. The student may be optimizing speed of responsibility. Those are different objectives, not different estimates of one score.

Career advice becomes useful when it changes a live decision. I ask fewer questions about the identity a student wants to claim and more about the work they are prepared to inspect closely.