September 2026 · 4 min
Good Work Is Half the Work
It speaks, until you're somewhere that can't hear it.
Good job. Well done. These are the words you can find stickered or written in red ink at the top of a school worksheet. The message is clear. It’s good to be right. It’s good to complete the task. This kind of encouragement is often all I had to motivate me as a kid. It instilled a maxim into my psyche: do good work.
This maxim confirmed itself for me repeatedly: I got more attention and resources because I could do good work. I got into Harvard not because I wrote a polished application or knew how to best present myself. I had no idea about any of that. All I had was my record of work and perseverance, and there was someone who could appraise that.
So the maxim held for years, until it didn’t.
I had a mentor, a senior staff member hired onto a team I joined as an intern. His charge was to make machine learning work for our company. He handed me a copy of Hands-On Machine Learning, and I read it religiously. I would learn a concept and see it in motion in the cash flow model I was building. He would say that making it work is the first part. Code that runs isn’t code that’s done: you prune it, you type it, you test it, you keep the place where you configure a thing separate from the place where you call it. He sharpened me as an engineer, gave me responsibility beyond my expertise. So I grew.
He was a good work speaks kind of man, and so was I. When something was wrong his answer was always the same: be better engineers, be more skilled, do better work. Neither of us had ever needed another answer.
The team was good. Everyone was improving, chemistry was building, timelines were shortening. He grew with it: as the team got bigger he became a manager, and he was good at that too. From the inside, it looked like the investment was paying off. We replaced a vendor’s safety model with one we built in house. It performed better and cost less. We migrated off another vendor’s modeling platform to containerized pipelines. They charged him with machine learning and got dev ops for free.
Then machine learning became somewhat of an esoteric term. Language models were AI now. Our company, like everyone, wanted to get the most out of it. But AI wasn’t a new product market to us, it was a set of tools. He was a craftsman, and you don’t use a hammer to fasten a bolt. He looked at what those models could actually do at that time, on our data, and the answer wasn’t much yet. He was an expert, but it didn’t matter: terms of art were now buzzwords.
Our company had a reorg. Our team didn’t survive. The team building models and software on top of data dissolved, at the exact moment the market decided models on data were everything.
It’s frustrating, but I never felt concern, truly. Good work speaks. Simultaneously my maxim was invalidated and a source of consolation. You can invest correctly in people, and get exactly what you invested in, and it can still be invisible. The room had no instrument that could see us. A company can count heads and costs and revenue per employee. It can’t count the judgment that would have saved it years. There’s no column for that on the spreadsheet.
You can invest correctly in people, and get exactly what you invested in, and it can still be invisible.
Anyone who has built models will tell you the hard part isn’t the algorithm, it’s whether the input was ever captured. Companies run on the same constraint. If nobody recorded it, it can’t be weighed, and quality isn’t a field anyone fills in. We spent our time on the work. Not on selling it, and not on justifying it.
That’s only a mistake in some places. Good work speaks inside institutions built to hear it. An admissions committee is one: appraisal is its entire function, and it pays people to read carefully and to communicate what would otherwise be obscure.
So do most companies, including that one. They properly appraise the skills their core business depends on, because those are existential. So the question was never how good our company was, or how much it valued its people. It was whether the skill sat on the path to the money. Where it does, the instruments already exist, and they were built out of self-interest. Where it doesn’t, there are only proxies: title, tenure, visibility, who speaks well in a meeting, and cost.
Which leaves two ways to be seen. Either your function is aligned with what the company is actually trying to do, and the seeing is automatic. Or it isn’t, and somebody has to market you inside the building.
Our failure was the second one. It wasn’t silence: our director talked to the executives constantly. She spoke the way the rest of us did, measured and precise and careful not to claim more than she could support. She reported. Nobody sold.
My mentor’s words cycle through my head often, the way any good mentor’s do. Especially now, when I build with the tools he was right to doubt back then. The lesson isn’t to work less carefully. It’s that good work is half the work, and to get the most from it you either need to be somewhere that appraises it properly or you need to find yourself a spokesperson. You should probably figure that second half out first.