Color Code

Mid-Season Report

Source Code covered the front of the agentic pipeline — getting truth out of a person and into the machine. Object Code covered the back — signing for the work the agent handed you. This is the middle: turning a feature spec into the prompt sequence that yields the best output the model can give you, and the generous, competitive field building that craft in the open.

It's a season vantage from the booth. The field has run its plays in public — in commits, in issue threads, on the record — and a veteran who's played decades of toolchains is calling them: which earned their standing, which merely caught on, which is fading, and where the game is heading. The pleasure is real and so is the honesty. Nobody's won, the season has no final score yet, and no scoreboard yet settles whether any of these plays gets better code out of the agent than a careful prompt and no ceremony.

The house rule, kept from the siblings: a number, its source, and its limit, in the same breath. The method here is archival — read the commit, not the README; the live API, not the dashboard; the platform's own concession, not its marketing. Vendor and steward claims are named as such, and the optimistic ones are distrusted as hard as the alarmist ones. The author runs the home team (Superpowers), discloses it, and grades it hardest. No play in this collection has been shown to ship measurably better code — that absence is carried, not finessed.

01

A Star Is a Bookmark

The league leader has ~239,000 stars, and some unknown share of them are fake. Every adoption metric quits one step short of whether the output is any good — so you stop trusting the stat sheet and read the record instead.

02

The Plays the Game Wrote

When four teams reach for the same move and no single source explains it, you stop asking about style and start asking what forced their hand. The one play that earned its standing: get what the model can't be trusted to hold out of its head and onto durable ground.

03

The Plays the League Borrowed

Five teams run the same spec-first opening — and the paper trail shows they got it from each other, not from the game. Steal the structure under it; hold the heavyweight notation as an open question, not a settled one.

04

The Plays the Box Score Loves

Ten agents at once is the flashiest thing in the league, and the budget-controlled tape shows it buys speed, not quality. The progress was in the decomposition and the gates the whole time — the parallelism just borrowed their jersey.

05

The House Always Catches Up

The platform keeps shipping the community's best plays as built-ins, so what happens to the team whose signature move just became a button? Absorption isn't obsolescence — the button carries the mechanism, never the judgment of when it bites.

06

Does Any of This Win Games?

Every play is real, converged, and platform-blessed — and not one is proven to get better code out of the agent than a plain, careful prompt. That's not a knock; it's the score at the break, and the reason there's a veteran in the booth at all.