Same week, two rooms. In the first one I am moving a production MSSQL database to PostgreSQL — schema, indexes, the control plane that goes in front of it. In the second one I am at a dinner during Abu Dhabi Finance Week, explaining to people with capital why any of it matters. One dinner. Sixty-eight introductions came out of it.
Most engineers pick one room and defend the border. Most commercial people do the same from the other side. And every startup I have watched closely pays, monthly, for that border — it just shows up under other names in the books.
Translation Is the Silent Line Item
Here is what the border actually costs. A customer says something in a sales call. An account manager summarizes it for a product manager. The product manager writes a ticket. An engineer reads the ticket and builds what the ticket says — which is now three translations away from what the customer meant. Each hop is lossy. Each hop takes days. And the loss compounds precisely at the moments that decide whether the company lives: pricing, objections, the feature that would have closed the deal.
Specialization is supposed to be efficient. It is efficient — inside one function. Between functions, it manufactures a translation layer, and translation layers are where startups die. Not dramatically. Slowly, one slightly-wrong ticket at a time.
What the Sales Room Does to Your Engineering
Before a release I walked the checkout as a buyer would and moved price validation to the server: any price the client sends is recomputed, not accepted. Not a UI patch — a server rule. The browser proposes; the server decides.
I do not believe I make that call the same way if I have never sat in the room where money changes hands. An engineer who has only seen tickets sees a rendering bug: hide the wrong price, done. An engineer who has watched a client decide whether to trust a company sees what a wrong invoice actually costs — because the invoice is the product, in the only moment the client is paying full attention. The sales room is where engineers learn which failures are expensive. There is no ticket format that carries that information.
It works in the other direction too. My time in growth taught me what a number has to survive before you say it out loud: an audience’s skepticism, a competitor’s counter, a follow-up question. Engineering taught me how to make numbers that survive — provenance on every value, sanity bounds, sources that admit when they are stale. Growth without engineering rounds up. Engineering without growth builds things nobody asked for, beautifully.
The Price of Crossing
I will not romanticize it. Crossing costs.
You are underestimated in both rooms — too commercial for the engineers until they read your schema, too technical at the dinner until the third question. Context switching is real; a morning of index tuning and an afternoon of positioning are different sports played with the same brain. And you become hard to hire for, because job descriptions are written for lanes, and you are the lane violation.
The startups that figure it out stop writing the border into their org chart at the one stage where the border is pure waste. Under roughly twenty people, every translation layer is a subsidy paid to nobody. The role that should exist at that stage — the person who owns the system end to end and is also the one in the room when it gets sold — is rare not because the combination is impossible, but because both cultures teach you, from the first day, to stay in your lane.
Stay in your lane long enough and the lane becomes the ceiling.