In the winter of 1609, Galileo Galilei pointed a telescope of his own grinding at the Pleiades and counted. Six stars were visible to the naked eye, seven on a good night. Through the glass he found thirty-six more. He turned to the Milky Way and found that the pale band across the sky was made of individual stars, thousands upon thousands of them, too faint to have been seen by anyone in the whole of human history. He published the drawings in 1610 in a short book called Sidereus Nuncius, the starry messenger, and the sky was never the same size again.
For the next two centuries the atlas-makers kept up. Bayer in 1603 had drawn some twelve hundred stars, each one placed inside a constellation figure engraved with the care of a Renaissance portrait. Hevelius in 1687 added new constellations of his own and kept the figures. Flamsteed in 1729 added precision and kept the figures. Bode in 1801 drew more than seventeen thousand stars, many of them telescopic, invisible to the eye that was meant to look at the chart, and he still kept the figures. Orion still had his club. Cassiopeia still had her chair.
Then the atlases stopped.
THE LAST FIGURE
The reason was arithmetic. Between 1852 and 1859 Friedrich Argelander at the Bonn Observatory catalogued 324,198 stars in the northern sky, one at a time, by eye, through a small telescope. The Bonner Durchmusterung was published as a set of charts, and the charts had no figures on them. The engravers were as skilled as ever. There was no longer room. You can draw a hunter around forty stars. You cannot draw one around a hundred thousand. The figure that had given the sky its meaning became a decoration in the way, and the decoration was removed.
Elijah Burritt's atlas of 1835 was the last major one to keep the constellation pictures. After that, star charts became what they are today: precise, exhaustive, and unreadable to anyone who has not been trained. The data won. The art that had made the data legible for two hundred years was the price.
Astronomers did not choose this. The telescope chose it for them. Every improvement in the instrument produced more stars than the previous generation had known existed, and at some point the number crossed a line past which the old way of seeing could not follow.
THE NEW INSTRUMENT
A contribution graph was designed for a particular density of work. A developer commits a few times a day on a good day. A very productive one, on a very good day, might commit twenty. The colour scale on GitHub runs to four shades of green, which was enough to cover the range of human output.
That range is changing. In February 2026 Andrej Karpathy declared that the new default for a developer was to write no code at all: to orchestrate agents that write it, and to act as oversight. We called this the seventh age of the developer, in a profession that has been declared finished six times before. Stripe's agent system produces more than a thousand merged pull requests a week. A single developer running a swarm of coding agents can generate in an afternoon the commit volume that used to take a quarter.
The graphs are getting brighter. We see it on the atlases we render. The star fields of 2026 are denser than the star fields of 2023 in a way that has nothing to do with people working harder.
This is the telescope problem, arriving in a different sky. The instrument now finds more stars than the old figures can hold. The archetypes we wrote about, the Steady Orbit, the Supernova, the Fade, were drawn around a human density of work. When every day is a supernova, the Supernova stops meaning anything. When an agent commits through the weekend, the weekend voids that made a graph look like a person's life fill in. The figures dissolve, the way Orion dissolved under Argelander's hundred thousand points.
WHAT A STAR WAS COUNTING
The question underneath this is what a star on the atlas was ever measuring.
The GitHub graph counts commits, pull requests, reviews, issues. It always has. But nobody who ever framed a contribution graph was framing a commit count. A commit is a unit of bookkeeping. What the graph recorded, accidentally, was attendance: the days a person sat down and decided that something should exist that did not exist yet. The count on any given day was a proxy for that decision, and a rough one.
Agents break the proxy. They do not break the thing it was standing in for. An agent that merges four hundred pull requests overnight has decided nothing. Somewhere, a human read the ticket, judged the approach, set the quality gate, looked at the output and said yes or no. That day of judgement is one day. It is one star.
Codex Stellarum already scales brightness relative to your own most productive period rather than to a fixed count, which means inflation cancels itself out. A day with four thousand agent commits and a day with forty are both bright if they are both the brightest you had. What the atlas cannot yet do, and what we will say plainly rather than fake, is tell a human commit from an agent's. The API does not distinguish. Neither can we.
THE MAGNITUDE OF SHOWING UP
Argelander's catalogue made the old way of drawing the sky impossible. The sky itself lost nothing. Within a generation, astronomers had found another way: photographic plates, then spectra, then the classification of stars by their light rather than their position, work done largely by women at Harvard whom history called computers. The figures were gone, and a deeper structure appeared in their place.
Something similar is available here. The commit count was always the crudest signal on the graph. The rhythm underneath it, the days on and the days off, the seasons, the gaps, the return after a long silence, was the thing worth keeping, and none of that is produced by an agent. An agent does not take August off. An agent does not go dark for six weeks after a burnout and come back lighter. An agent does not stop for a new baby. The pattern of a human life around the work is exactly what the agents cannot generate, and it is exactly what an atlas is drawn from.
The tools will keep finding more stars. The question for anyone making charts is whether the human shape can still be read through them. We think it can, if the chart is honest about what it is counting.
After Bode, the art drained out of astronomy. It does not have to happen twice.
See what your sky looks like this year. Enter your GitHub username at codexstellarum.com and preview your atlas - free to try, prints from £25.