Software engineering calories: what would cal.ai say about your product?
Most teams have no idea what they're putting into their product
For most of 2025, I worked for myself, at home. My wife and I are not very good at resisting tasty food (opening the fridge just to see what’s inside is a habit of mine), so we just don’t buy junk. A side benefit from working from home was that I actually ate pretty healthy. I also trained for a marathon, so I was in pretty good shape and happy with my body.
(Getting to engineering in a second, promise)
At the end of last October, I started a new hybrid job. There are many benefits in my workplace, one of them being a crazy breakfast. And I’m the breakfast-kind of guy. When I’m in hotels with buffets, I eat so much that I’m barely hungry at dinner.
It was fun at first, but very quickly I started to gain weight - 3 kg in the first month(!). I tried setting up rules for myself: not touching the snacks drawer, limiting myself to one plate at breakfast, not eating sweets.
But nothing really changed, which didn’t make any sense! I’m eating one plate of breakfast - some salads, cheese, and bread - regular lunch, and a light dinner. I’m running almost every day, and I STILL somehow keep gaining weight.
One day during breakfast, I complained about my first-world problems, and a coworker suggested the very sensible idea of just tracking what I eat, something I’ve never really done (I’m 30, so I felt there was still time until I had to fight nature).
I first tried by opening a WhatsApp group with myself, but after a couple of days, I downloaded cal.ai to give it a try (not connected to them in any way).
I found the 2 culprits:
My breakfast plate was way more calorie-heavy than I expected. The cheese, salads with mayo, bread, etc, added up to 1000+ calories. Shocking, right?
I was snacking on salty things much more than I expected (I love those damn salted cashew nuts). It was not sugar, so I felt like I’m not ‘violating my rules’.
Now, to engineering.
Engineering Bodies
The math of our bodies is simple: eat more calories than you burn → gain weight. Burn more than eat → lose weight.
Imagine your product was a human body. Your input is code, and your output is feature adoption. And like me with my “healthy” breakfast, most companies have no idea what they’re actually putting in (more on that in a moment).
Whether you need to gain or lose weight depends on where you are. Some companies (like young startups) are skinny teenagers, they genuinely need more input. But most companies are just getting fatter - every CEO wants to move faster, pushes for more code, more features. More weight should mean more strength, like in judo. A 90kg fighter will crush a 60kg one. But if you reach 400kg, a 50kg kid will beat you (even if it’ll take them a while). Some products would genuinely be better if half their features were deleted.
No matter where your product sits, it can only absorb a certain amount of calories while staying fit. Google does this very nicely. You rarely see new features in Gmail or Calendar - but that’s how I want them. It’s a battle-tested fighter that knows not to gain weight to win. I don’t see anything replacing Google Calendar anytime soon.
But for companies that do need to gain weight, the question becomes: how do you even know how much you’re eating?
So what are engineering calories?
Yes, the code is your input.
But measuring lines of code is like counting bites. There might be some correlation between the number of bites you took in a day and your calories, but it’s super low. You can bite empty air and it’ll count.
A bit better approach is the number of PRs. I’d say it’s like counting the number of servings you ate, more servings usually means more calories. This is better than nothing, but what’s the actual caloric value of each PR? A one-line config change and a full architecture refactor are both “one PR,” but they’re not the same meal.
That’s what Weave does - using ML to estimate the real caloric value of each piece of work.
For newer readers: I consult for Weave, so I’m not objective here. But I’d encourage you to plug it into your GitHub - it takes 2 minutes, and the data is often surprising.
Now, I know that some of you are thinking this is about managers hiding in dark rooms figuring out who to fire. It’s not. Obviously coding is just a small part of an engineer’s work, and a lot of it is invisible.
Look at how PostHog does it (from their public operating docs):
Engineers can log in to see their numbers and those of other engineers.
We understand that all the work an engineer does can’t be properly represented in a tool. […]
We don’t use Weave to:
Make a call on how valuable someone is - some people with low output are very valuable to the company, and we’re 100% aware that things like heavy support load can impact output
They give ALL their engineers access to data about everyone else. If someone produces less input, it doesn’t mean they’re a bad developer. Maybe they’re reviewing everyone’s code. Maybe they’re mentoring. Maybe they’re stuck in meetings explaining to everyone else how to work better.
Sometimes, as an Engineering Manager, you feel like things are moving well. Meetings are happening, engineers are busy, PRs are opening. But when you look back, you feel like you barely delivered anything this quarter. The problem in 99% of the cases is not your engineers, but the systems your company has in place.
The goal isn’t a perfect measure. Cal.ai couldn’t see fat cheese hidden under a salad leaf, but it was at most 10-20% off, which is much better than hunches.
When you actually measure the inputs to your product, you see what’s holding you back:
If a specific repo has much lower input, maybe it needs some love.
If your strongest engineer has lower-than-expected numbers, they’ll be happy to explain what shit they’re dealing with.
If someone is super productive, they’re the ones who can share their system with others.
A note of caution
Most of you reading this work in companies that don’t actually need to gain weight. They need to start by exercising more - increase current feature adoption, or delete useless features.
For those who actually need to gain weight, remember that not all calories are created equal. 1000 junk food calories are very different from 1000 healthy meal calories. You might not notice it in the short term, but over time, your body will feel the difference.
The same goes for code. If the input to your system is tons of garbage AI-written code, it’ll come back to bite you (even if you don’t notice the difference yet).
I’m still taking photos of my meals, by the way. I look ridiculous doing it, but I’m down 4kg :)
Discover weekly
Revenue addiction kills companies by Elena Verna. It was always a bad habit. But in the AI transition, it has become deadly (in all weight categories).