Positive users praise analogies and effort-based human metrics as better ways to assess AI productivity than line counts, while negative users criticize the practice by noting AI's tendency to over-engineer code and require heavy deletions.
Most Activity
@svpino how about tokens used
@svpino I have a skill that tracks effort measured against human metrics - I reversed engineered tasks I perform and time taken and use a skill to track how much time the same task run by ai saves me.
Measuring AI productivity in "number of lines written" is a stupid mistake.
One Day, Everyone Will Have Always Been Against This.
@svpino measuring productivity by lines of code is the same energy as measuring a chef by how many ingredients they used
@svpino Junior writes 500 lines. Senior writes 50. Same job done. To a bad metric, junior's the rockstar.
That's how you hire mediocrity.
@svpino did you know meta measures tokens spent per employee? 😂
@svpino Lines of code was always a bad proxy. Now it's just more obviously wrong. The real measure is how fast you go from problem to working solution and how long it stays working.
@svpino I, as of lately, measure it in credit tokens burnt... one day I will have always been against this too...
@jima00 @svpino That's actually an interesting measure
@brankopetric00 @svpino It always beats me why companies run after effort metrics and not outcome metrics.
@lumitech_co But how do you quantify that?
@0xprotovox I love this analogy. On point.
@svpino Lines of code is the easiest metric to game. For enterprise AI, the better read is cycle time through review: how much human clarification, test repair, and rollback risk the model creates or removes before anything ships.
@RyanMorrisonJer this sounds... interesting :)
@svpino More code aint better code. Best engineers delete way more than they add
Since when was this ever an acceptable measurement? Any engineer worth their salt is typically unmoved when they learn about the number of lines of code.
The only atypical movement of the needle is if they suspect it's a low number for what it's doing, or too high a for what it's doing.
@svpino Lines of code have always been a terrible productivity metric
@svpino Lines of code was already a weak proxy before AI made it trivial
now anyone can generate volume instantly so the signal becomes meaningless the harder question is what actually replaces it as a quality signal
@svpino "ibm enters chat..." :)
The moment code generation became cheap, counting lines stopped telling you much.
What's really worth now is figuring out soon enough if those lines should exist, if they work, if they are worth the tokens spent and if you'll still want them six months from now.