Riders
People who have publicly described running or building for agent swarms. We list fleet size or token spend only when they said it themselves, and every entry links to the post.
21 lit particles, one for each riderPoint at a row to find its particle
| Rider | What they’ve said | Stated scale | Type | Date | Source |
|---|---|---|---|---|---|
@matthuang · Co-founder, Paradigm; founder, Tempo |
Coined “swarm riders”: people who command large agent swarms to productive ends, with productive token spend he puts at 1,000–10,000x a median software engineer. Says many of the riders he knows are at Tempo and Paradigm. | Not stated | Voice | 2026-11-10 | x.com |
@jacobeverly · Product, Boundless |
Says he coordinates swarms toward larger outcomes, is rolling the capability out to his team, and that taste and defining verifiable success become the scarce skills. | 8–16 agents · ~600M tokens/day (aiming for 1B) |
Practitioner | 2026-11-10 | x.com |
@jsnnsa · Founder, Spawn |
Runs Spawn with ten people and a permanent fleet of named agents that fix user-hit errors and ship to production. Says the company spends 30x payroll on AI tokens. | 6.5T Claude Opus tokens in 30 days |
Practitioner | 2026-10-18 | x.com |
@lxe · AI engineering, Pocket Model |
Steered a swarm of Codex and Claude agents, each owning one design lane, to #7 of about 1,270 teams in the Muse Spark Open Kernel Challenge. Wrote up nine lessons (reset agents often, steer the supervisor, watch costs). | 6 parallel agent lanes on 2–3 spot H200s |
Practitioner | 2026-11-04 | x.com |
@Vtrivedy10 · Applied research, LangChain Labs |
Runs a “24/7 research factory”: persistent agents own workstreams and launch frontier swarms against well-defined briefs, while he talks to one or two orchestrators. | “Billions of tokens a day” across subscriptions |
Practitioner | 2026-11-08 | x.com |
@ivanzhouyq · AI research, Databricks |
Says his research team moved from running experiments by hand to a swarm of agents that provisions GPUs and runs training jobs, so the humans focus on goals and reviewing outcomes. | Not stated | Practitioner | 2026-10-16 | x.com |
@bread_ · MegaETH; Modern Market show |
Built a T3 Agents dashboard for watching subagents and is testing it at swarm scale. Described swarm riders as “a new class of prompter” who maintains many agents at once. | Testing ~16k active subagents |
Practitioner | 2026-11-16 | x.com |
@gakonst · GP & CTO, Paradigm; eng lead, Tempo |
Building Nanocode, an SDK for cloud agent swarms on Durable Objects with a Rust harness compiled to wasm. Also asked what people want to see when a swarm is running on a topic. | Not stated | Builder | 2026-11-11 | x.com |
@emilianobonassi · VP Engineering, Boundless |
Launched AmphoraDB, an experiment in serverless Postgres for agent swarms: no signup, the wallet is the login, paid per statement over MPP on Tempo. | Not stated | Builder | 2026-11-13 | x.com |
@MiguelriosEN · Founder & CTO, Grep / AgentRun |
Wrote up how Grep built swarms: a manager agent that ticks asynchronously, code for scheduling and budgets, and agents with inboxes. Argues a swarm is coordinated work, not a batch over a list. | Swarms of up to ~1,000 agents |
Builder | 2026-10-28 | x.com |
@PranjalAggarw16 · PhD student, CMU LTI |
Co-built SwarmChat, an open-source Slack for agents, then ran swarm-scaling experiments on ProgramBench. Reports 16-agent swarms matched a single agent’s score 6x faster. | ~0.5T tokens of experiments |
Builder | 2026-10-29 | x.com |
@aaronjmars · Builder, MiroShark & Aeon |
Argues we’re in “the swarm era”: the goal is finding problems where throwing billions of tokens keeps improving the output, and renting GPUs for open-model workers can beat per-token pricing. | Not stated | Builder | 2026-11-13 | x.com |
@nxt3d · Co-founder, Unruggable |
Building ID Agents, a “swarm OS” for running Claude Code, Codex and Cursor agents from one terminal, and writing about reputation and trust inside swarms. | Not stated | Builder | 2026-10-16 | x.com |
@mattshumer_ · Investor; ex-CEO, HyperWrite |
Says his goal is to be “the single best allocator of tokens in the world” and that he will use as many tokens as it takes. His bio reads “Commander of the Swarm.” | Not stated | Practitioner | 2026-11-09 | x.com |
@noah_z_zhang · Solo founder |
Shares his token budget openly as a solo founder running long coding tasks on Devin and frontier models. | 500M–1B tokens/day (later 1–2B) |
Practitioner | 2026-10-29 | x.com |
@illscience · Investor, a16z |
Notes from conversations: swarms aren’t just brute-force parallelism, mixing frontier and cheaper models may set a new cost curve, and swarms will be far more token-hungry than agents. Running his own experiments. | Not stated | Voice | 2026-11-16 | x.com |
@levie · CEO, Box |
Says swarms will clearly separate teams that adopt them from those that don’t, wherever value is bound by research, decisions, data or code, and predicts a new field of FDE work wiring them up. | Not stated | Voice | 2026-11-11 | x.com |
@musalbas · Co-founder, Celestia |
“Being a swarm rider is quietly the most valuable skill at any organisation right now.” | Not stated | Voice | 2026-11-10 | x.com |
@Leothebayesian · Pseudonymous markets account |
Argues research speed in trading now comes down to how many frontier agents you can coordinate in parallel, and is bearish on quant shops without massively parallel agent infra. | Not stated | Voice | 2026-11-16 | x.com |
@vamsibatchuk · Member of technical staff, Google DeepMind |
Ran Google Antigravity’s /swarm: 20+ builder agents produced a tactical FPS browser game (30k+ lines of code) while 36 blind visual-judge critics reviewed the renders and sent fixes back. | 20+ builder agents · 36 critic agents |
Practitioner | 2026-11-17 | x.com |
Sr. Principal Engineer, AWS |
Wrote that thousands of autonomous AI agents wrote every line of code in AWS Emulator, and that his team hasn’t written or reviewed any of the code. Agents write tests, find issues and fix them; a merge gate decides what lands. | Thousands of autonomous agents |
Practitioner | 2026-11-11 | aboutamazon.com |
Nothing matches that search. Know one we’re missing?
Missing someone or something? Submit it with a public source.