She does not code, and she was done first
We both had the same problem to solve last month. Our ops teammate, call her Marina, spends part of every week reconciling two exports: one from a CRM, one from a marketplace dashboard. Names spelled differently, dates in different formats, a few rows that exist in one file but not the other. It was an hour of squinting at spreadsheets, twice a week.
I said I would write a script for it. I know Python. I know pandas. I had it on my list.
Marina, who has never written a line of code and does not want to, opened a chat with an assistant and started describing the job. Four days later she had a working script that took the two files and produced a clean matched list, plus a short report of the rows that did not line up. She was using it. I was still, if I am honest, at the "I will get to it this week" stage.
That stung a little. It also taught me the thing I keep repeating to people now.
The bottleneck was never the code
Here is what I assumed the hard part was: writing the loop, handling the fuzzy name matching, dealing with the date formats, catching the missing rows. Real work, sure. But it turns out a decent model can produce that code quickly once it knows exactly what "matched" means.
The part that was actually hard, the part Marina was better at than me, was saying precisely what the job is.
I would have started coding with a vague picture in my head and discovered the edge cases as I hit them. Marina could not lean on code, so she had to lean on words. And her words were good, because she does this task by hand every week and knows every ugly corner of it. Her first brief was something like: "I have two files. File A comes from our CRM, columns are these. File B comes from the marketplace, columns are these. Match a person in A to a person in B when the email is the same. If the email is missing, match on full name ignoring case and extra spaces. Show me the pairs that matched, and separately list anyone in A with no match in B and anyone in B with no match in A. Dates in File B look like this, please normalize them."
Read that again. There is no code in it. But there is a complete specification. Every rule an engineer would have to guess at, she stated, because she is the one who lives with the guesses being wrong.
What her iterations looked like
She did not get it perfect on the first try, and this is the part I want other developers to sit with. She iterated the way a good product owner iterates, not the way a coder debugs.
First version matched too little. She noticed some obvious same-people were not pairing up, looked closer, and realized one file had a leading apostrophe on some emails from an export quirk. So she added: "Some emails in File B start with a stray apostrophe, strip that before comparing." Second version over-matched two different people who shared a common name. She added: "If two people have the same name but different emails, do not match them, list both as unmatched so I can check by hand." Each round, she was not fixing code. She was making the description of the job sharper.
That is a skill. It is the same skill that makes a good bug report, a good ticket, a good design doc. Know the task cold, state the rules, name the edge cases, say what the output should look like. Marina had all of that from doing the work by hand. What she was missing (the syntax) turned out to be the cheap part to fill in.
What I took from it as an engineer
I have not stopped writing code. But I have changed where I think the value is.
The people around me who cannot program are no longer blocked from building small tools, as long as they can describe a job with the precision they would use to train a new hire. That is a big shift, and it moves the scarce skill from "can type Python" toward "can think clearly about a process and pin it down in plain language."
For me, it means two things. My own briefs to myself got sharper, because I saw how much faster clear specs move. And I started encouraging non-technical teammates to try the small stuff themselves instead of queueing it behind me, then bringing me in only when something needs real engineering (security, scale, touching production data). Marina's reconciliation script did not need me. It needed her clarity.
The bottleneck moved. Worth noticing where it went.
AGINE Academy is an independent product by AGINE AI (not affiliated with Anthropic). We teach building with Claude by doing the work, not watching lectures.
Top comments (0)