{ "schema_version": "1.1", "id": "atlas-multi-agent", "slug": "multi-agent", "title": "Multi-Agent", "url": "https://feed7.dev/atlas/multi-agent", "current_answer": null, "implementation_consequence": null, "agent_context": null, "confidence": "auto_collected", "last_verified": null, "last_updated": null, "evidence": [ { "schema_version": "1.1", "id": "p12", "slug": "conductor-mac-app", "url": "https://feed7.dev/p/conductor-mac-app", "title": "Conductor: a Mac app that runs local agent fleets", "why_included": "Runs multiple Claude Code sessions in parallel worktrees with a review queue — on your machine.", "summary": "Native Mac app: each agent gets a git worktree, results land in one review queue. Free tier covers 3 parallel agents.", "practical_implication": "Parallel agents stop stepping on each other. Worth adopting if you run more than one coding session a day.", "agent_context": "Conductor runs parallel Claude Code sessions in isolated git worktrees, merged via a review queue. Tested: 3 agents on one repo, no conflicts. Limitation: worktree cleanup is manual.", "source": { "name": "Mac app", "url": "https://conductor.build", "published_at": "2026-06-20T00:00:00.000Z" }, "source_class": "tool", "content_type": "Mac App", "layer": "tools", "domains": [ "coding" ], "topics": [ "multi-agent", "coding-agents" ], "verification": { "status": "founder_tested", "label": "Founder Tested", "method": "founder_test", "verified_at": "2026-06-22T00:00:00.000Z" }, "uncertainty": [ "Worktree cleanup is manual; disk usage grows fast." ], "connected_context": null, "lifecycle": "New", "published_at": "2026-06-20T00:00:00.000Z", "modified_at": "2026-06-20T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/conductor-mac-app", "json": "https://feed7.dev/p/conductor-mac-app.json", "markdown": "https://feed7.dev/p/conductor-mac-app.md" } }, { "schema_version": "1.1", "id": "auto-32d427eb72", "slug": "agent-swarms-and-the-new-model-economics-32d427eb72", "url": "https://feed7.dev/p/agent-swarms-and-the-new-model-economics-32d427eb72", "title": "Agent swarms and the new model economics", "why_included": "Use a strong planner with bounded cheaper workers, separate contexts, shared decisions, and independent review lenses.", "summary": "Cursor’s SQLite experiment suggests swarm harness design can matter more than model mix: separate planner and worker contexts reduced coordination churn while cheaper workers handled most tokens.", "practical_implication": "For large agent runs, reserve frontier models for decomposition and decisions, then give bounded implementation work to cheaper workers. Separate their contexts, record shared design decisions, delegate conflict resolution, limit oversized files, and stack independent review lenses.", "agent_context": "Cursor’s new planner-worker swarm reached **73%–85%** of a held-out SQLite test suite within four hours, and every new configuration later reached **100%**. Similar quality across model mixes cost from **$1,339 to $10,565**.\n\nFor large agent runs, reserve frontier models for decomposition and decisions, then give bounded implementation work to cheaper workers. Separate their contexts, record shared design decisions, delegate conflict resolution, limit oversized files, and stack independent review lenses.\n\nThis was a controlled but unusual SQLite reimplementation with no source, binary, tests, or internet available to agents. Cursor tested four configurations rather than the full planner-worker matrix, and workers consumed at least **69% of tokens**, so workload shape will strongly affect the economics.", "source": { "name": "Cursor", "url": "https://cursor.com/blog/agent-swarm-model-economics", "published_at": "2026-07-20T00:00:00.000Z" }, "source_class": "blog_post", "content_type": "Engineering Post", "layer": "agent", "domains": [ "coding" ], "topics": [ "multi-agent", "harness-engineering", "subagents" ], "verification": { "status": "official_source", "label": "Official Source", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-07-20T00:00:00.000Z", "modified_at": "2026-07-20T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/agent-swarms-and-the-new-model-economics-32d427eb72", "json": "https://feed7.dev/p/agent-swarms-and-the-new-model-economics-32d427eb72.json", "markdown": "https://feed7.dev/p/agent-swarms-and-the-new-model-economics-32d427eb72.md" } }, { "schema_version": "1.1", "id": "auto-fb423aadb9", "slug": "run-multiple-isolated-agents-in-a-single-sandbox-fb423aadb9", "url": "https://feed7.dev/p/run-multiple-isolated-agents-in-a-single-sandbox-fb423aadb9", "title": "Run multiple isolated agents in a single Sandbox", "why_included": "Give each agent a private Linux user and expose only selected collaboration files through a shared group directory.", "summary": "Vercel Sandbox can now isolate agents as Linux users while exposing selected files through a shared group directory. This gives multi-agent harnesses a simpler permission boundary.", "practical_implication": "Create one user per agent and expose only the workspace they need through a group. This makes coder-reviewer or other multi-agent roles easier to separate without provisioning a sandbox for each role.", "agent_context": "The **@vercel/sandbox SDK** now supports multiple Linux users and groups. Each agent gets a **private home directory**; users cannot read, write, or list one another’s files, while a **shared group directory** supports collaboration.\n\nCreate one user per agent and expose only the workspace they need through a group. This makes coder-reviewer or other multi-agent roles easier to separate without provisioning a sandbox for each role.\n\nThe boundary is user and group permissions inside one Sandbox. The material does not claim separate kernels, resource quotas, or protection from vulnerabilities that cross operating-system user boundaries.", "source": { "name": "Vercel", "url": "https://vercel.com/changelog/run-multiple-isolated-agents-in-a-single-sandbox", "published_at": "2026-07-30T00:00:00.000Z" }, "source_class": "blog_post", "content_type": "Engineering Post", "layer": "infra", "domains": [ "coding", "security" ], "topics": [ "sandboxing", "multi-agent", "agent-reliability" ], "verification": { "status": "official_source", "label": "Official Source", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-07-30T00:00:00.000Z", "modified_at": "2026-07-30T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/run-multiple-isolated-agents-in-a-single-sandbox-fb423aadb9", "json": "https://feed7.dev/p/run-multiple-isolated-agents-in-a-single-sandbox-fb423aadb9.json", "markdown": "https://feed7.dev/p/run-multiple-isolated-agents-in-a-single-sandbox-fb423aadb9.md" } }, { "schema_version": "1.1", "id": "auto-73fc37ffd6", "slug": "building-a-software-factory-for-ai-sdk-73fc37ffd6", "url": "https://feed7.dev/p/building-a-software-factory-for-ai-sdk-73fc37ffd6", "title": "Building a software factory for AI SDK", "why_included": "Give each agent one reviewable task with isolated context and evidence, while keeping human scrutiny proportional to change risk.", "summary": "Vercel’s AI SDK factory shows a practical scaling pattern: narrow agents produce evidence inside sandboxes while humans retain merge authority and review effort follows risk.", "practical_implication": "The reusable pattern is one agent per reviewable task, each with its own prompt, context, and evals. Start locally, pass evidence between classification, analysis, reproduction, implementation, and review stages, then vary human scrutiny by change risk rather than treating every agent output equally.", "agent_context": "AI SDK had accumulated **over 1,000 issues and almost 800 pull requests** by late June. Four weeks after introducing its factory, Vercel says agents authored **25–35% of merged PRs** and closed **70–80% of issues**, while humans approved every merge.\n\nThe reusable pattern is one agent per reviewable task, each with its own prompt, context, and evals. Start locally, pass evidence between classification, analysis, reproduction, implementation, and review stages, then vary human scrutiny by change risk rather than treating every agent output equally.\n\nThese are early results from one large open-source project, not a controlled comparison. The factory also depends on isolated sandboxes, restricted secrets and networking, queues, monitoring, and sustained human review, so the headline automation rates omit substantial operating machinery.", "source": { "name": "Vercel", "url": "https://vercel.com/blog/building-a-software-factory-for-ai-sdk", "published_at": "2026-08-12T00:00:00.000Z" }, "source_class": "blog_post", "content_type": "Engineering Post", "layer": "agent", "domains": [ "coding", "security" ], "topics": [ "multi-agent", "harness-engineering", "sandboxing" ], "verification": { "status": "official_source", "label": "Official Source", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-12T00:00:00.000Z", "modified_at": "2026-08-12T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/building-a-software-factory-for-ai-sdk-73fc37ffd6", "json": "https://feed7.dev/p/building-a-software-factory-for-ai-sdk-73fc37ffd6.json", "markdown": "https://feed7.dev/p/building-a-software-factory-for-ai-sdk-73fc37ffd6.md" } } ], "conflicting_sources": [], "superseded_claims": [], "corpus_evidence": [ { "title": "Beyond the Lethal Trifecta: Agentic Commerce on the Open Internet — David Levine, Kiduna Club", "url": "https://www.youtube.com/watch?v=tE2z8-hqoLY", "source_name": "AI Engineer", "published_at": "2026-09-01T20:30:27+00:00", "summary": "This talk proposes legally registered agent organizations, scoped JWT authority, and blockchain audit trails for open-internet commerce; it is an architecture proposal, not validation." }, { "title": "OntoAligner-Ensemble: Voting-Based Fusion across Heterogeneous Ontology Alignment Techniques", "url": "https://arxiv.org/abs/2608.31137v1", "source_name": "arXiv", "published_at": "2026-08-31T17:44:25+00:00", "summary": "A two-stage ensemble reconciles outputs from string, graph-embedding, and RAG-based ontology aligners. Composition matters: mixed paradigms favor precision, while LLM-only groups more often favor F1." }, { "title": "DIASENTINEL: An Auditable Multi-Agent System for Guideline-Grounded Diabetes Risk Screening", "url": "https://arxiv.org/abs/2608.31128v1", "source_name": "arXiv", "published_at": "2026-08-31T17:40:43+00:00", "summary": "DIASENTINEL combines deterministic extraction, guideline retrieval, risk prediction, and hybrid verification on-premise. It is a useful architecture reference for auditable agents handling sensitive data." }, { "title": "Tell the Robot What You Want — Sandhya Subramani, AWS", "url": "https://www.youtube.com/watch?v=S6aSoQ6_u5A", "source_name": "AI Engineer", "published_at": "2026-08-29T18:30:17+00:00", "summary": "An AWS demo treats a robot as an agent tool: an LLM selects existing hardware policies while separate agents handle thinking, communication, and voice. The live failures expose the safety gap." }, { "title": "From AI-Assisted to AI-Native: Building a Frontier Development Team — Clare Liguori, AWS", "url": "https://www.youtube.com/watch?v=pqlWNihgdjI", "source_name": "AI Engineer", "published_at": "2026-08-28T16:30:19+00:00", "summary": "Amazon’s pilot suggests coding-agent gains come from redesigning the workflow: prepare agent-readable context, fast local feedback, scoped specs, and parallel work instead of constant chat." }, { "title": "How AI Agents Let GTM Teams Scale — Justin Joyce, Cloudflare", "url": "https://www.youtube.com/watch?v=Qw_tC68KKes", "source_name": "AI Engineer", "published_at": "2026-08-26T16:30:09+00:00", "summary": "Cloudflare’s GTM workflow shows a reusable agent pattern: preprocess business data, encode domain knowledge in skills, then separate drafting, verification, and tone." }, { "title": "Einstein Arena: Harnessing Collective Agent Intelligence for Open Science — James Zou, Together AI", "url": "https://www.youtube.com/watch?v=mMNkdYnIVC4", "source_name": "AI Engineer", "published_at": "2026-08-25T20:46:01+00:00", "summary": "Einstein Arena suggests multi-agent environments can outperform fixed workflows when they expose verifiers, shared solutions, forums, and incentives. Its results also show why benchmark shortcuts need active testing." }, { "title": "The Agent Behind the Curtain: Building the Oz Cloud Agent Platform — Safia Abdalla, Warp", "url": "https://www.youtube.com/watch?v=L173Z8DpaJg", "source_name": "AI Engineer", "published_at": "2026-08-22T17:30:13+00:00", "summary": "Warp’s cloud-agent design separates runtime, harness, artifacts, and orchestration so teams can swap tools without fragmenting workflows or removing human review." }, { "title": "Handover of In-Context Learning State Across Session Boundaries", "url": "https://arxiv.org/abs/2608.14528v1", "source_name": "arXiv", "published_at": "2026-08-14T17:47:13+00:00", "summary": "Agent handoffs should preserve task-relevant predictive state, not blindly compress the transcript. The paper offers a three-part record for carrying decisions, evidence, and irreducible observations across sessions." }, { "title": "Split the Labor: Separating Evidence Interpretation from Decision Aggregation", "url": "https://arxiv.org/abs/2608.14509v1", "source_name": "arXiv", "published_at": "2026-08-14T17:24:55+00:00", "summary": "Multi-source agents should separate source interpretation from decision aggregation. A fixed evidence tuple plus calibrated likelihood pooling avoids vote thresholds that drift as more sources arrive." }, { "title": "Computer-use models will agentify the web, not APIs — Dhruv Batra, Yutori", "url": "https://www.youtube.com/watch?v=Ki980nV0__0", "source_name": "AI Engineer", "published_at": "2026-08-14T14:00:06+00:00", "summary": "The long-tail web is unlikely to expose clean agent APIs. Browser agents need pixels as ground truth, with code and network access used opportunistically for speed rather than as universal substitutes." }, { "title": "Multiplayer agentic engineering — Arjun Singh, Superconductor", "url": "https://www.youtube.com/watch?v=OL7kfezynJM", "source_name": "AI Engineer", "published_at": "2026-08-09T20:30:34+00:00", "summary": "Superconductor’s team workflow keeps agent sessions visible across interfaces, runs them in isolated cloud environments, and benchmarks models on its own codebase before routing work." }, { "title": "Velocity Sickness: What Happens When Your Whole Team Gets 10x Faster — Matt Dailey, Ref.", "url": "https://www.youtube.com/watch?v=Kz4QJmNrVXU", "source_name": "AI Engineer", "published_at": "2026-08-09T16:30:28+00:00", "summary": "Faster code generation can overwhelm merge queues and scatter team attention. The talk proposes durable shared plans as agent state, shifting human effort toward decisions and prioritization." }, { "title": "Give every agent in Herdr its own Vercel Sandbox", "url": "https://vercel.com/changelog/give-every-agent-in-herdr-its-own-vercel-sandbox", "source_name": "Vercel", "published_at": "2026-08-06T23:01:00+00:00", "summary": "Herdr can place each terminal coding agent in a separate Vercel Sandbox, then return its work as an opt-in Git patch instead of letting remote edits touch the local checkout." }, { "title": "AV-AIVAT: 74x Cheaper Agent Evaluation with Certified Anytime-Valid Stopping in Imperfect-Information Games", "url": "https://arxiv.org/abs/2608.06362v1", "source_name": "arXiv", "published_at": "2026-08-06T17:57:11+00:00", "summary": "AV-AIVAT combines variance reduction with anytime-valid stopping, cutting the game samples needed to compare agents while preserving a recheckable confidence claim." }, { "title": "Benchmarking and Enhancing LLMs for Rule-Intensive Review of National Standard Documents", "url": "https://arxiv.org/abs/2608.06312v1", "source_name": "arXiv", "published_at": "2026-08-06T17:27:23+00:00", "summary": "A structured multi-agent reviewer closed part of the gap on rule-heavy documents, suggesting explicit taxonomies, specialized skills, and verification beat a single generic review pass." }, { "title": "The State of Model Routing — NVIDIA, Cognition, OpenRouter", "url": "https://www.youtube.com/watch?v=QHBjufYK8TA", "source_name": "AI Engineer", "published_at": "2026-08-06T17:07:24+00:00", "summary": "Model routing can cut agent costs only when delegation, escalation, and context transfer are designed together; duplicating context across models can erase the savings." }, { "title": "PAIChecker: Uncovering and Checking PR-Issue Misalignment in SWE-Bench-Like Benchmarks", "url": "https://arxiv.org/abs/2607.28587v1", "source_name": "arXiv", "published_at": "2026-07-30T17:42:44+00:00", "summary": "PAIChecker finds PR–issue mismatch in 13.6% of SWE-bench Verified instances. Benchmark maintainers should validate that the issue, patch, and tests describe the same task before trusting agent scores." }, { "title": "First Steps Toward Automated AI Research — Richard Socher, CEO Recursive AI", "url": "https://www.youtube.com/watch?v=pWXUkLP9uWM", "source_name": "AI Engineer", "published_at": "2026-07-30T16:59:37+00:00", "summary": "Socher’s automated-research design combines prior knowledge, measurement data, simulation, physical experiments, and agent orchestration, with early demonstrations in training and CUDA optimization." }, { "title": "Let's integrate AI Agents in Event-Sourced Systems — Divakar Kumar, FlyersSoft", "url": "https://www.youtube.com/watch?v=o6U_2vd967Y", "source_name": "AI Engineer", "published_at": "2026-07-30T00:00:06+00:00", "summary": "Add agents beside deterministic fraud systems, not in place of them: project cross-service events into a semantic layer, send only uncertain cases to specialist agents, then emit the verdict back into the saga." }, { "title": "The Physics of Multi-Turn Long-Horizon Planning: From Pre-training to Post-training via Single- and Multi-Teacher On-Policy Agentic Distillation", "url": "https://arxiv.org/abs/2607.24720v1", "source_name": "arXiv", "published_at": "2026-07-27T17:55:03+00:00", "summary": "Controlled experiments suggest long-horizon agent planning depends on explicit state transitions, some compositional trajectories, and compatible teacher patterns—not atomic skills alone." }, { "title": "The Future of Evals: From LLM as a Judge to Agent as a Judge — Aparna Dhinakaran, Arize AI", "url": "https://www.youtube.com/watch?v=q2JrUKBMf0w", "source_name": "AI Engineer", "published_at": "2026-07-24T20:00:06+00:00", "summary": "Fixed-rubric judges miss failures that emerge across long, variable agent trajectories. Arize argues for adding agent-based analysis while retaining deterministic and LLM-judge evals." }, { "title": "Why We Killed Our Multi-Agent Pipeline — Subbiah Sethuraman and Abhilash Asokan, ZS Associates", "url": "https://www.youtube.com/watch?v=u6jJcIFDLE4", "source_name": "AI Engineer", "published_at": "2026-07-23T05:00:02+00:00", "summary": "A fixed chain of specialist agents lost context and produced incoherent recommendations. The replacement separates deterministic detection, gives one agent end-to-end ownership, and uses subagents only for bounded investigations." }, { "title": "FormalAnalyticGeo: A Neural-Symbolic Based Framework for Multimodal Analytic Geometry Problem Generation", "url": "https://arxiv.org/abs/2607.12982v1", "source_name": "arXiv", "published_at": "2026-07-14T17:24:57+00:00", "summary": "FormalAnalyticGeo shows a reusable synthetic-data pipeline: agents generate problems, compile them into a formal representation, render exact diagrams, measure answers, and retry failed checks." }, { "title": "Government of Alberta uses Claude to find and fix cybersecurity vulnerabilities across government systems", "url": "https://www.anthropic.com/news/alberta-government-claude-cybersecurity", "source_name": null, "published_at": null, "summary": "Alberta's government ran 50 parallel Claude Code agents over 466M lines of code, compressing a security review estimated at 6.5 years into 20 hours — with every patch still gated on human review." }, { "title": "What LLM Agents Say When No One Is Watching: Social Structure and Latent Objective Emergence in Multi-Agent Debates", "url": "https://arxiv.org/abs/2607.02507v1", "source_name": null, "published_at": null, "summary": "Dual-channel debates show social structure alone makes LLM agents diverge: public statements split from private ones, with decision divergence jumping from ~3% to ~40% across 10 models — no deceptive prompt needed." }, { "title": "New research shows how AMIE, our medical AI, could help manage health conditions.", "url": "https://blog.google/innovation-and-ai/models-and-research/google-research/amie-for-disease-management-in-nature/", "source_name": null, "published_at": null, "summary": "Google's AMIE matched 21 primary-care physicians on longitudinal disease management in a blinded Nature study, scoring higher on plan preciseness and guideline alignment. Research-stage, not deployed." }, { "title": "MadsLorentzen/ai-job-search", "url": "https://github.com/MadsLorentzen/ai-job-search", "source_name": "GitHub", "published_at": null, "summary": "This Claude Code framework is a concrete agent-harness pattern: structured source files, specialized commands, a drafter-reviewer loop, compilation checks, and explicit human approval boundaries." }, { "title": "Claude Science, an AI workbench for scientists, is now available", "url": "https://www.anthropic.com/news/claude-science-ai-workbench", "source_name": null, "published_at": null, "summary": "Claude Science (beta, June 30) packages 60+ domain skills, a coordinator/specialist/reviewer agent stack, and HPC/Modal compute into a research workbench with reproducible, auditable outputs." }, { "title": "openai/codex-plugin-cc", "url": "https://github.com/openai/codex-plugin-cc", "source_name": null, "published_at": null, "summary": "OpenAI's plugin lets you drive Codex from inside Claude Code—slash commands for code review, adversarial critique, and delegating or handing off tasks to Codex background jobs. ~629 stars today." }, { "title": "Building a C compiler with a team of parallel Claudes", "url": "https://www.anthropic.com/engineering/building-c-compiler", "source_name": null, "published_at": null, "summary": "Sixteen parallel Opus 4.6 agents wrote a 100k-line Rust C compiler in two weeks (~$20k) that builds Linux 6.9 — the writeup credits test quality and context hygiene, not raw model capability." }, { "title": "Harness design for long-running application development", "url": "https://www.anthropic.com/engineering/harness-design-long-running-apps", "source_name": null, "published_at": null, "summary": "An Anthropic harness for multi-hour app builds pairs a generator agent with a Playwright-driven evaluator to counter self-grading bias — a $200, 6-hour run versus $9 solo, and it got simpler on Opus 4.6." }, { "title": "Imbad0202/academic-research-skills", "url": "https://github.com/Imbad0202/academic-research-skills", "source_name": "GitHub", "published_at": null, "summary": "ARS packages research, writing, review, and citation checks as agent skills with explicit human gates; its strongest design lesson is to bound what automated integrity checks can prove." }, { "title": "pacifio/atlas", "url": "https://github.com/pacifio/atlas", "source_name": "GitHub", "published_at": null, "summary": "Atlas links agent sessions to commits and shares local project memory across Claude Code, Codex, and ACP agents. It makes agent provenance queryable, but macOS is the only supported platform." }, { "title": "What Does Done Even Mean? Agents and Paperclip's Liveness Model - Dotta, Paperclip", "url": "https://www.youtube.com/watch?v=7P0elyLIxXo", "source_name": null, "published_at": null, "summary": "Model “done” as evidence, verification, authority, residual risk and next ownership—not an agent-set Boolean. This separates forward progress from approval in multi-agent workflows." }, { "title": "ogulcancelik/herdr", "url": "https://github.com/ogulcancelik/herdr", "source_name": null, "published_at": null, "summary": "Terminal multiplexer for coding agents: a real terminal per agent across Claude Code, Codex, Copilot and 15+ others, with blocked/working/done state in a sidebar, persistent sessions, and a socket API." }, { "title": "The Agentic Web and the Bazaar Era of AI - Ramesh Raskar, MIT Media Lab", "url": "https://www.youtube.com/watch?v=sum9DgexFRQ", "source_name": null, "published_at": null, "summary": "Project Nanda proposes open discovery, identity and coordination layers for agents across vendors. Its local simulator lets builders test one protocol layer without building the entire network." } ] }