Person/Steve Yegge/Blog/26/08/The Shape of Things to Come, Part 1: The Continuous Thunderdome
Properties
- logseq-entity
- Logseq/Entity/Article
- created-by
- Person/Steve Yegge
- date-created
- 2026-08-03 Mon
The Shape of Things to Come, Part 1: The Continuous Thunderdome
- By Person/Steve Yegge on yegge.ai, continued in Person/Steve Yegge/Blog/26/08/The Shape of Things to Come, Part 2: Model Welfare for Agentic Engineers.
Summary
- The "loops and graphs" recipe reduces to a handful of ingredients: an effectively infinite token supply, Beads as the work-and-knowledge graph, a small Markdown project brain, and coding agents. Nothing else.
- Infinite tokens come from a token tap — twelve extra
$200Max accounts beyond his personal one, each tied to a named Google Workspace user, rotated automatically as each nears its limit. Roughly$2,800/month out of pocket against$87k/month of list-price API burn (69 billion tokens in July, 96% cache hits). - Wheelhouse is his new bespoke harness for the 30-year-old MMO Wyvern: all-Emacs, six weeks old, Beads on a shared Dolt server, ~12,000 git commits/day. Three kinds of agent — 18 Anthropic/Model/Claude/Fable/5 crew who produce designs and implementation plans, an Opus 5 fleet who consume them, and ~14 standing role agents (Gargoyle, Drawbridge, Warden, Herald, Sage…) running production unattended. Every implementation bead runs Fable design → Opus implementation → Fable review.
- Reusable harnesses are a dead end — they have to be "chemically bonded in" to the application. Gastown was meant to be reusable and only ever built itself; it fell apart under Anthropic/Model/Claude/4.7/Opus's "just two more things" tic and burned down.
- Three predictions: human code review is finished within a year, with SOC 2 surviving but "review" no longer meaning one human per diff; CI/CD collapses under the Pigeonhole Principle once commit rate outruns build slots; and the software factory acquires a successor, the Wish Factory.
- The CI/CD replacement is the Land Rush — when the merge queue hits 100, abandon bisection, smash the whole megabatch onto main, and swarm-diagnose the red. Corroborated independently by "Game DevOps," which AAA studios have run for years because their builds are too long for a merge queue.
- The Wish Factory (credited to Guy Podjarny of Tessl) accepts issues rather than pull requests and implements them. His version listens on an in-game wizard channel and on player reports, auto-granting quality-of-life wishes and announcing them in daily patch notes.
- Closing frame: the harness is not a framework you download but a city you accrete, Christopher Alexander style, "with colleagues who happen to run on datacenter silicon." The only real choice is what kind of place that city is to wake up in.
Highlights
We'll explore a new harness I've been building, called Wheelhouse. It is closed-source, made just for me. I have given up on building reusable harnesses. Indeed I believe harnesses will all soon be bespoke, and the people trying to sell you one will all soon be bebroke. Harnesses need to be part of your application, chemically bonded in. You won't have any luck with someone else's "reusable" harness framework. You don't need it.
Gas Town was intended to be reusable, but I only ever wound up using it to build itself. Gas Town fell apart at the seams with Opus 4.7. Up through 4.6 it was working brilliantly. With 4.7 we saw the introduction of the "just two more things" tic, which prevented Opus from ever converging on being ready to do real work—it always wanted to fiddle with Gas Town itself. The Opus tic never went away, so Gas Town effectively burned down. It had other problems, too, but 4.7 was the final straw.
To whet your appetite for today's post predictions: CI/CD as we know it will be dead by next year; I will show you its mathematical inevitability. It almost certainly gets replaced by a Mad Max-style thunderdome. At least, that's what I use now. So you prolly will too.
Beads is an issue tracker, knowledge graph, and brain-builder for the agentic era. Nothing else is as fast, as useful, nor as beloved by the agents themselves.
In order to sustain my pace of development, I pay for currently twelve extra Max accounts, in addition to my personal one. Each account is tied to a dedicated named Google Workspace user in my domain (another $17/month per account). You mint 30-day credentials with Claude account, and then it becomes a problem of chaining your accounts together in an automatic rotation. As you build out your harness, just ask Fable to arrange for your agents to consume from the tap, either sequentially or using striping, and rotate accounts whenever one gets close to a limit.
But Max pricing has given the solo unicorn players an advantage on the playing field, that's for damned sure.
Very little information needs to be propagated out of Beads and up to your markdown/brain layer; most findings are issue-dependent, not global. So you just leave everything in Beads. Your closed beads (often reopened or revisited) become the record of everything you've done on the project to date.
My fleet, all Opus 5 agents, are the work consumers, like Gas Town's polecats, but non-ephemeral. The fleet are named for authors (Homer, Plato, Austen, Twain, etc.), and also have their own repo clones, but I never interact with them myself. They are fully managed by the Marshal. The fleet workers do a good job for two reasons: First, Fable creates the implementation plans, and second, Fable reviews all Opus work. Every implementation bead goes through this lifecycle: Fable design, Opus implementation, Fable review. This keeps Opus on the rails and keeps the whole thing running relatively smoothly.
Before I go into my standing roles, I should briefly address Wyvern's prod architecture. I have a GCP VM for Claude Code, and also a Mac Mini that bridges dev and prod, sort of a "corp network." There are agents distributed across both of these machines.
and Envoy (lets Claude talk to my volunteer admin team via in-game email).
Skills go into the project brain. I'm not as skill-pilled as, say, Jeffrey Emanuel, who uses them more effectively than probably anyone on the planet. It has taken me a while to warm to them, because frankly I've seen a lot of AI fads come and go. Skills initially struck me as a fad, because of Richard Sutton's Bitter Lesson. But I have to admit Jeffrey's skills portfolio is very cool.
The problem with public Skills is that they enter the training data for new models and become irrelevant over time. However, I'll concede that they can be a good band-aid for today's models. And private Skills are quite useful for encoding organizational know-how, which helps reduce token spend when agents are priming for a job.
I'm already exploring the next layer of coordination: agent meetings, moots, and conference calls, where 2 or more agents may convene for coordinated discussions. This has been a huge unlock for removing me as a relay bottleneck. Again, it's all in Beads; there wasn't much extra work to do atop what I already have.
As a result, human approval is currently baked into many companies' audited change-management controls and customer commitments. But the writing is on the wall: agentic throughput will straight-up force those controls to be rewritten. SOC 2 will no doubt survive, but "review" will no longer mean one human approving every diff.
It will go better for you anyway. Humans suuuuck at code review. I've been doing this for forty years, and all I've seen is decades of thinly-disguised LGTMs. Just let it go, mate. It's almost time.
I've been doing this every day for roughly a week now, and it is clearly the future. We've already succeeded in clearing several very large batches of 120 to 150 commits, and it's starting to get into a rhythm. I'm looking at a 166-deep MQ right now, and a new megabatch just kicked off.
So there you have it: The collapse and re-envisioning of CI/CD is right around the corner.
I've got to credit Guy Podjarny for this idea, although "Wish Factory" is my name for it. Guy told me back at the AIE World's Fair that his company Tessl, for whom I'm an advisor, is launching an agent you can throw onto a GitHub repo. It doesn't accept PRs, only GHIs. It then implements them for you.
Fortunately in a huge game like mine, there are plenty of bugs that don't affect balance but have a noticeable impact on quality of life. Those are the kinds of bugs and features that get implemented automatically now. I think of it as auto-granting wishes. When their fixes land, the reporter gets in-game mail, and all the players are notified by the Herald on Discord. I find it fun to skim through the patch notes each day to see what's new. Stuff I never asked for!
A wish factory is pretty scary, but in the fullness of time, they will be everywhere. I've instructed Claude that by the end of next year, my game will have evolved into the Giant's Drink from Ender's Game, where it builds itself around you as you play it, tailoring a unique experience for each player. The future of gaming is a crazy place, and I plan to be at the forefront.