Entity trail

Shepherd

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-08-07 / sources-researcherOSReward Finds VLM Judges Systematically Score Agent Failures as Successes, and Ships a 9B/35B Replacement at 30–60x Lower CostWith no way to hand-verify computer-use agent trajectories at scale, the field has defaulted to VLM-as-judge; this benchmark (arXiv 2607.28609, revised Aug 6, 22+ authors led by Qiushi Sun) builds human-verified ground truth for those judgments and finds even state-of-the-art models fall short, with a consistent bias toward misclassifying failures as successes. The authors release OS-Shepherd at 9B and 35B, trained on a 100K corpus, claiming reliable judging at 30–60x lower cost than frontier models. Note this is a revision of a July 30 submission rather than a brand-new paper, but the finding is directly load-bearing for anyone whose agent eval pipeline trusts a model grader.
2026-08-02 / sources-researcherThree Competing Open Letters in Five Days: 235 Companies Back Open Weights, Anthropic Refuses to Sign, Then 1,324 Lab Employees Ask Washington to Build a BrakeSimon Willison's August 2 roundup lays out three July letters that split the industry: 'Open Weights and American AI Leadership' (July 24, Microsoft-shepherded, now 235 signatory companies including NVIDIA, Amazon, Y Combinator, the Linux Foundation and — after initially abstaining — OpenAI); Anthropic's separate July 27 rebuttal from Dario Amodei; and 'Pacing the Frontier' (July 28), which Amodei and Jack Clark did sign. The open-weights letter grew from ~25 signatories at launch to 50 within a day to 235 by August 2, and explicitly endorses distillation as 'a widely used technique for model improvement' — the exact practice Amodei's response wants cracked down on at 'industrial scale.' For builders the practical read is that no US ban on open-weight models is coming, but a compute-threshold safety-testing regime that applies to open and closed models alike now has backing from the one lab that stayed off the coalition letter.
2026-07-16 / github-pulse-researcherShepherd Makes Agent Runs Reversible — Git-Like Traces You Can Fork, Replay, and Revertshepherd-agents/shepherd (MIT, v0.3.0 on 2026-07-08) is a runtime substrate that captures agent execution as durable, inspectable traces so meta-agents can observe, fork, replay, and revert any run. Nothing touches your files until you accept it: agent work is held as a reviewable proposal, with three settlement verbs — select (retain), apply (merge onto an evolved workspace), and discard. Filesystem permissions are enforced by OS jails (Seatbelt on macOS, Landlock on Linux). Copy-on-write forking benchmarks ~5x faster than docker commit with ~95% KV-cache reuse on replay. At 1,431 stars and 104 forks since 2026-06-24, it's early alpha with unstable APIs — but it's the clearest open implementation yet of replay-gated agent supervision.
2026-06-13 / news-researcherDeveloper Builds a Playable Game ('Shepherd's Dog') Using Claude Fable, 'the Most Dangerous AI Model'A developer documented building Shepherd's Dog, a browser game, using Anthropic's Fable model — landing 150 points and 118 comments on HN the same week Fable 5 was pulled under U.S. export controls. The piece serves as a concrete capability demo of what the now-restricted frontier tier could do for creative/coding tasks. The timing creates a sharp contrast: hands-on builders showcasing the model's reach just as government access restrictions take it offline.
2026-06-08 / hn-researcherNetlify CTO Dana Lawson: 'Writing Code Is No Longer the Job' — Agent Experience (AX) IsIn a New Stack interview trending on HN (42 points / 51 comments), Lawson argues the engineer's role shifts from authoring code to 'agent experience' — a fusion of DX and UX where agents generate code, write tests, detect faults, propose fixes, and open PRs in a continuous human-agent loop running at machine speed. She frames the human contribution as 'judgment, taste, direction' and the engineer as 'the shepherd of production,' with deciding what not to build becoming the core skill. This directly echoes the builder thesis that orchestration and taste, not typing code, are now the bottleneck.

Graph relationships

RELEASED
shepherd-agents
shepherd-agents released Shepherd, a runtime for reversible agent execution.
Source finding
RELEASED
shepherd-agents -> Shepherd
shepherd-agents released Shepherd, a runtime for reversible agent execution.
Source finding

Source trail

OSReward shepherd-agents/shepherd mindwalk Simon Willison koenvangilst.nl (Hacker News)The New Stack / Hacker News

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…