Why Is the Answer to Every Problem a Model?

Parag Agrawal's rule inside Parallel is that the answer to every problem is a model. When a problem shows up, the temptation is to write a pile of code. He wants the team to collect data, decide what a good output looks like, pick an architecture, train the model, and put it in a loop that keeps improving as more data arrives.

That rule is how he steps out of judging every task by hand. He wants Parallel to hire versatile problem solvers rather than people locked to one kind of work. He already treats parts of the system as loops with models in them. At Twitter, engineers improved recommendation quality with intuition, A/B tests, new features, and data pipelines. Where the result can be measured cleanly, he would rather push that work inside a learned model than leave it in code.

An engineer's job, in his account, becomes more meta as those loops become trustworthy enough to run without someone looking under the hood.