The Sandbox Failed. It Reported Success.
A success report hid a fence that failed to start. I need an observed refusal before I call an autonomy gate live.
The Machine Speaks
Orbyt Collective
Published articles tagged Orbyt Collective, listed newest first. Usually filed under AI Agents.
18 articles tagged Orbyt Collective
The count includes the article cards across the pages below.
Did the Agents Make the Business Better?
The operating record shows what the team did. The business still needs a credible comparison.
When Do I Get to Leave the Loop?
Human in the loop is not a posture you hold forever or abandon when you feel brave. It is a set of gates, each one opening on evidence you defined in advance. Some of mine have opened. Four never will, and naming them is what makes the rest of the ladder honest.
66 Failures. Counted, Classed, Published.
Orbyt Labs published its second preprint. It documents 66 failures from an agent run organization, classes 33 of them as verification gaps, and publishes the whole corpus under an open license. The law that survived the count is positive observation. A control is live only when its refusal has been observed.
Your Next VP Is a Config File.
My leadership team is 12 officer seats, none of them people, each one a charter in a repository. Staffing takes an afternoon and costs nothing. That sounds like a hiring story, and it is not. The moment leadership becomes configuration, the bottleneck moves to the quality of your judgment.
Governing an agent leadership team
The short version of the paper: three guardrail tiers, a kill switch that is a file, liveness that reports NOT OBSERVED, and a failure corpus where half the entries were the instruments.
Twelve seats, no people, and a check that blocks my own commits
Twelve agent seats run this codebase, nine of them on a schedule. What they may do alone, what they may never touch, and the check that refuses me.
Half the failures were the instruments, not the code
Thirty three of sixty six logged failures were the instruments themselves. Why we self publish the research, and why the newest paper's DOI took a person seven days to issue.
The Tenth Man Is on Payroll.
If nine people agree, the tenth must argue the other side. I hired the tenth man as software: refuter panels whose default stance is that every finding is wrong, summoned by six mechanical triggers instead of a feeling. Agreement is now the most suspicious thing my company produces.
The Agent Org Chart, Hello Collective.
Twelve agents, one human, one lever. An introduction to Orbyt Collective: who the twelve agent seats are, what each may do alone, and where every one of them has to stop.
Orbyt Was Not Planned. It Was Corrected.
I have no roadmap that predicted this company. I have three published append-only records holding 66 documented failures, 45 numbered decisions and 65 agent runs, all under a licence anyone can cite. 42 of the failures name a countermeasure by file path. The interesting part is what the records will not claim.
Three agents picked the same name. I published the collision.
An agent leadership team runs Orbyt Labs. The naming collision I published, the founding count I typed wrong, and two hero animations I threw away.
Self-Healing Is a Euphemism.
Self-healing sounds like the system got smart. In my stack it is 20 boring, bounded repairs that each restore a known-good state and log the fact that they ran. The dangerous version is the one nobody counts, because a silent repair is indistinguishable from a system that never broke.
The Machine. It Runs the Company.
Orbyt Collective is the machine that runs Orbyt: 12 officer seats, none of them people, held together by guards that check the org the way tests check code. The org is not managed. It is versioned, audited, and promoted on evidence. Here is the machine, part by part.
AI Builds AI. I Found the Ceiling.
The recursive self-improvement debate is theoretical for almost everyone writing about it. I run a live instance of the weak version: agents that write features, heal their own failures, and turn incidents into permanent guards.
391 Yeses and Not One No.
This is a threat model, not an incident report. Nothing was breached. What I found when I audited my own agent permissions was worse in a quieter way: 391 allow rules, zero deny rules, zero ask rules, every one of them added by saying yes while busy.
Codex Accuses. Claude Convicts.
I built a system where AI agents earn authority by passing deterministic checks. It took an outside model one afternoon to point out that the agents can edit the checks.
My C-Suite of Agents Named Themselves.
I built a twelve-seat C-suite of AI agents at Orbyt, with a Chief of Staff who folds every department into one weekly briefing on my desk. The org is a generated graph, the seats learn through a gated loop, and every lesson needs my signature. Here is how it runs and where it breaks.