{ "schema_version": "1.1", "id": "atlas-harness-engineering", "slug": "harness-engineering", "title": "Harness Engineering", "url": "https://feed7.dev/atlas/harness-engineering", "current_answer": null, "implementation_consequence": null, "agent_context": null, "confidence": "auto_collected", "last_verified": null, "last_updated": null, "evidence": [ { "schema_version": "1.1", "id": "auto-db6b4984d3", "slug": "harness-engineering-is-not-enough-why-software-factories-db6b4984d3", "url": "https://feed7.dev/p/harness-engineering-is-not-enough-why-software-factories-db6b4984d3", "title": "Harness Engineering is not Enough: Why Software Factories Fail — Dex Horthy, HumanLayer", "why_included": "Agree on design before implementation and build vertical slices so generated changes remain readable and testable.", "summary": "Coding-agent loops can raise throughput without preserving maintainability. Keep human ownership of code, and use upfront alignment to make review affordable instead of trying to automate it away.", "practical_implication": "Keep humans responsible for the resulting code. Use model-assisted planning, agree on design before implementation, and build in vertical slices so every generated change remains practical to read and test.", "agent_context": "The talk argues that coding models are rewarded mainly when **code runs and tests pass**, not when architecture remains easy to change. Review agents and extra loops can raise the floor, but cannot supply a missing maintainability signal.\n\nKeep humans responsible for the resulting code. Use **model-assisted planning**, agree on design before implementation, and build in vertical slices so every generated change remains practical to read and test.\n\nThere is no established benchmark here that proves how well current models preserve codebase quality. Longer-task evaluations such as **SWE Marathon**, DeepSuite, and FrontierCode may help, but model-based quality judges have their own ceiling.", "source": { "name": "AI Engineer", "url": "https://www.youtube.com/watch?v=Ib5GBkD555M", "published_at": "2026-07-23T00:00:00.000Z" }, "source_class": "video", "content_type": "Video", "layer": "agent", "domains": [ "coding" ], "topics": [ "harness-engineering", "agent-reliability", "agent-evals" ], "verification": { "status": "source_linked", "label": "Source Linked", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-07-23T00:00:00.000Z", "modified_at": "2026-07-23T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/harness-engineering-is-not-enough-why-software-factories-db6b4984d3", "json": "https://feed7.dev/p/harness-engineering-is-not-enough-why-software-factories-db6b4984d3.json", "markdown": "https://feed7.dev/p/harness-engineering-is-not-enough-why-software-factories-db6b4984d3.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-5077dc1992", "slug": "how-we-set-up-our-cloud-agent-environment-5077dc1992", "url": "https://feed7.dev/p/how-we-set-up-our-cloud-agent-environment-5077dc1992", "title": "How we set up our cloud agent environment", "why_included": "Treat the dev environment as agent infrastructure with Linux parity, one discoverable CLI, end-to-end tests, and automated repair.", "summary": "Cursor’s cloud-agent adoption grew after it treated the dev environment as agent infrastructure: Linux parity, one discoverable CLI, end-to-end testing, and automated repair.", "practical_implication": "Treat the development environment as part of the agent harness. Match cloud and local dependencies, replace fragile command sequences with one discoverable interface, expose verification workflows, and let supervisors own long-running processes.", "agent_context": "Cursor moved its monorepo onto agent-ready Linux VMs, centralized startup and utilities in an internal CLI, and added end-to-end computer-use testing. Cloud agents went from **roughly 1 in 10 merged PRs in December** to **more than half today**.\n\nTreat the development environment as part of the agent harness. Match cloud and local dependencies, replace fragile command sequences with one discoverable interface, expose verification workflows, and let supervisors own long-running processes.\n\nSkills alone only helped at the margins because the underlying commands remained convoluted. Cursor also built **Cursor Cloud MCP** and **Cloud Doctor** to diagnose and repair environment failures, but the adoption figures are internal and do not isolate which change caused the increase.", "source": { "name": "Cursor", "url": "https://cursor.com/blog/cloud-agent-environment", "published_at": "2026-07-30T00:00:00.000Z" }, "source_class": "blog_post", "content_type": "Engineering Post", "layer": "agent", "domains": [ "coding" ], "topics": [ "cloud-agents", "harness-engineering", "skills" ], "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/how-we-set-up-our-cloud-agent-environment-5077dc1992", "json": "https://feed7.dev/p/how-we-set-up-our-cloud-agent-environment-5077dc1992.json", "markdown": "https://feed7.dev/p/how-we-set-up-our-cloud-agent-environment-5077dc1992.md" } } ], "conflicting_sources": [], "superseded_claims": [], "corpus_evidence": [ { "title": "Learning on the Job: The Future of Post-Training — Raymond Feng, Applied Compute", "url": "https://www.youtube.com/watch?v=k35LeKZEhiE", "source_name": "AI Engineer", "published_at": "2026-07-31T22:30:06+00:00", "summary": "Training on a production harness could adapt an agent to real workflows, but non-replayable interactions break common RL assumptions and make feedback harder to turn into updates." }, { "title": "Data and Environment Curation for Post-Training LLMs — Mahesh Sathiamoorthy, Bespoke Labs", "url": "https://www.youtube.com/watch?v=ewtOo0scUh0", "source_name": "AI Engineer", "published_at": "2026-07-31T22:00:06+00:00", "summary": "Post-training gains depend heavily on task selection, rollout quality, and environment design. For many enterprise agents, curated SFT may deliver most of the value before costly RL." }, { "title": "Scaling to Long Horizons — Ross Taylor & Chengxi Taylor, General Reasoning", "url": "https://www.youtube.com/watch?v=2bvtay8wGYI", "source_name": "AI Engineer", "published_at": "2026-07-31T21:30:06+00:00", "summary": "Long-running agents strain context, reward design, and GPU scheduling at once. Compaction and value models can help, but each introduces another optimization trade-off." }, { "title": "Emulated: The Data for Fully Autonomous Software Engineers and Companies — Joseph Wang", "url": "https://www.youtube.com/watch?v=zkX03APVj0M", "source_name": "AI Engineer", "published_at": "2026-07-31T21:00:03+00:00", "summary": "Repo-only tasks omit deployments, incidents, traffic, and distributed failures. Training infrastructure agents may require multi-node environments with real resources, despite much higher setup cost." }, { "title": "Reinforcement Learning without Verifiable Rewards — Will Brown, Prime Intellect", "url": "https://www.youtube.com/watch?v=AQv3qRCG6Gw", "source_name": "AI Engineer", "published_at": "2026-07-31T15:00:06+00:00", "summary": "For agent work without deterministic answers, mine production traces into tasks, build controllable simulators, and use scaled judges to derive rubrics while watching for reward hacking." }, { "title": "Agents at Scale: Inside MiniMax's Model and the Infrastructure Behind It — Olive Song", "url": "https://www.youtube.com/watch?v=AVMr9PMINyo", "source_name": "AI Engineer", "published_at": "2026-07-31T03:00:14+00:00", "summary": "MiniMax and Together describe why agent workloads change inference priorities: large codebase context, repeated tool turns, multimodality, and long-horizon RL all demand purpose-built environments and kernels." }, { "title": "fighting slop with slop — Vaibhav Gupta, Boundary", "url": "https://www.youtube.com/watch?v=AMiyLItEtLA", "source_name": "AI Engineer", "published_at": "2026-07-31T01:11:30+00:00", "summary": "Boundary replaces blanket code review with stable architecture rules, agent-generated tests, transcript inspection, execution traces, and human triage—a concrete harness pattern, not a license to skip verification." }, { "title": "Your Finance Agent's Bottleneck Is You — Ramana Siddanth Emani, Auditoria AI", "url": "https://www.youtube.com/watch?v=z0sh8HyTrDo", "source_name": "AI Engineer", "published_at": "2026-07-30T03:00:06+00:00", "summary": "Production agent velocity depends less on model swaps than on automating the developer loop: isolate parallel work, encode workflows as skills, connect tools, and keep humans as verifiers." }, { "title": "Build for the Memo, Not the Demo — Shawn Chan, China Resources Holdings", "url": "https://www.youtube.com/watch?v=tJFjeMBKbIY", "source_name": "AI Engineer", "published_at": "2026-07-30T02:00:06+00:00", "summary": "Finance agents need claim-level provenance, explicit uncertainty, consistency checks, surfaced contradictions, and logged approval. Fluent output without those controls will not survive diligence." }, { "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": "We Vetted 2000 AI Skills Before They Reached Developers — Lucas Palma, Nubank", "url": "https://www.youtube.com/watch?v=iKQ78wyJEXU", "source_name": "AI Engineer", "published_at": "2026-07-29T22:00:06+00:00", "summary": "Treat agent skills as supply-chain dependencies. Nubank scans them locally and in CI with deterministic rules plus LLM review, then gates marketplace distribution and feeds findings into vulnerability management." }, { "title": "How Kepler Built Verifiable AI for Financial Services — Vinoo Ganesh", "url": "https://www.youtube.com/watch?v=Tt2kX2sgQio", "source_name": "AI Engineer", "published_at": "2026-07-29T21:00:06+00:00", "summary": "Kepler’s pattern keeps LLMs away from numeric mutation: agents choose sources and steps, while deterministic systems persist, calculate, and reject unverifiable values." }, { "title": "Why Off-the-Shelf AI Doesn't Understand Money — Udi Menkes, Intuit", "url": "https://www.youtube.com/watch?v=Owb8g3yDyzo", "source_name": "AI Engineer", "published_at": "2026-07-29T20:00:06+00:00", "summary": "Intuit argues that domain context alone does not create experience. Its approach learns from verified state-action-outcome histories, using frontier models only to propose candidates." }, { "title": "SimulationMaxxing: How we ship agents 20× faster — Aman Gupta (Nubank) + Shreya Rajpal (Snowglobe)", "url": "https://www.youtube.com/watch?v=KMR_RBoCa4M", "source_name": "AI Engineer", "published_at": "2026-07-29T19:00:06+00:00", "summary": "Nubank uses simulated multi-turn traces to evaluate agent changes before production, shortening release cycles while checking simulation results against real data and human review." }, { "title": "Skills are new features: Building Skill-Centric Harness — Yogendra Miraje, FactSet", "url": "https://www.youtube.com/watch?v=7jjudsEhBtM", "source_name": "AI Engineer", "published_at": "2026-07-29T18:00:06+00:00", "summary": "FactSet treats skills as versioned product features and the harness as their runtime. Routing descriptions, model-specific evals, access controls, and governance matter as libraries grow." }, { "title": "MindForge: Teaching Small Language Models Whole-Life-Cycle Software Engineering via Source-Free Program Synthesis", "url": "https://arxiv.org/abs/2607.27146v1", "source_name": "arXiv", "published_at": "2026-07-29T17:23:02+00:00", "summary": "MindForge trains a 27B coding model on source-free program synthesis, raising ProgramBench pass rate from 37.98% to 49.51% and improving across seven unseen benchmarks." }, { "title": "How Forward Deployed Engineering is done at Factory — Eno Reyes", "url": "https://www.youtube.com/watch?v=wpOA-UXynoM", "source_name": "AI Engineer", "published_at": "2026-07-28T22:00:06+00:00", "summary": "Factory’s deployed-engineering model treats agent adoption as workflow design: instrument the path from signal to deploy, build validators, and tie autonomy to measurable business outcomes." }, { "title": "AI tools for Forward Deployed Engineering — Vasuman Moza, Varick Agents", "url": "https://www.youtube.com/watch?v=l0FLhNqBOic", "source_name": "AI Engineer", "published_at": "2026-07-28T21:00:06+00:00", "summary": "Varick treats enterprise agents as process-reengineering systems: capture how work really happens, encode that context, then automate only the steps whose risk permits it." }, { "title": "How Forward Deployed Engineering is done at Ramp — Leo Mehr", "url": "https://www.youtube.com/watch?v=ITMXwI6QL6A", "source_name": "AI Engineer", "published_at": "2026-07-28T19:00:06+00:00", "summary": "Ramp pairs strict request scoping with agents across the delivery pipeline. Its intake agent reportedly saves about 20% of scoping time, while humans retain final judgment." }, { "title": "How Forward Deployed Engineering is done at Decagon — Sunny Rekhi", "url": "https://www.youtube.com/watch?v=7wu2hsRfvV0", "source_name": "AI Engineer", "published_at": "2026-07-28T17:00:35+00:00", "summary": "Decagon splits deployment between configuring each customer’s agent and turning repeated enterprise requests into product features. The scarce skill is resisting brittle one-offs." }, { "title": "How Forward Deployed Engineering is done at Kepler — Vinoo Ganesh", "url": "https://www.youtube.com/watch?v=1OMHGsUZiqA", "source_name": "AI Engineer", "published_at": "2026-07-28T16:00:00+00:00", "summary": "Kepler frames forward deployment as product discovery: observe real work, ship the smallest useful fix, then turn repeated pain and customer vocabulary into durable product leverage." }, { "title": "AI Agents for Performance: Ship Faster, Pay Less — Rajat Shah, Netflix", "url": "https://www.youtube.com/watch?v=CgsWxRUY5Eo", "source_name": "AI Engineer", "published_at": "2026-07-28T00:59:04+00:00", "summary": "Netflix’s performance-agent playbook feeds production profiles and exact deployed code to agents, then relies on canaries and engineers to validate proposed optimizations." }, { "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": "State of Data — Sean Cai, Independent / State of Data", "url": "https://www.youtube.com/watch?v=ZyIoTOAbRfs", "source_name": "AI Engineer", "published_at": "2026-07-26T17:00:06+00:00", "summary": "Real workflow traces may teach agents more than manufactured tasks, while benchmark scores can shift with the harness. Build pipelines around live work and test across scaffolds." }, { "title": "Loop Engineering from First Principles — Kyle Mistele, HumanLayer", "url": "https://www.youtube.com/watch?v=xIt_mTQp6mY", "source_name": "AI Engineer", "published_at": "2026-07-25T20:41:40+00:00", "summary": "Replace unbounded coding-agent loops with measured, incremental changes: sense code state, choose one bounded task, open a reviewable PR, and pause until a human responds." }, { "title": "From Agent Traces to Agent Simulations — Rustem Feyzkhanov, Snorkel AI", "url": "https://www.youtube.com/watch?v=Ib5t2RLtxvM", "source_name": "AI Engineer", "published_at": "2026-07-25T01:00:02+00:00", "summary": "Reconstructing production traces as replayable environments lets teams compare agent stacks under fixed conditions and gate releases on task outcomes, cost, latency, and retries." }, { "title": "From Signal to PR: Anatomy of a Self-Improving Agent — Jason Lopatecki, Arize", "url": "https://www.youtube.com/watch?v=9HbzAWnKbo4", "source_name": "AI Engineer", "published_at": "2026-07-24T20:15:32+00:00", "summary": "Agents can turn production traces, evals, repository context, and observability skills into investigated issues or draft fixes. Human review remains more important as changes grow." }, { "title": "CausalForge: A Formally Grounded, Self-Improving Agentic Framework for Automated Research in Causal Inference", "url": "https://arxiv.org/abs/2607.22511v1", "source_name": "arXiv", "published_at": "2026-07-24T17:32:35+00:00", "summary": "CausalForge pairs a Lean-verified causal-inference library with an autonomous research pipeline and a semantic statement audit. Formal proof checks derivation, not whether the theorem matches the intended claim." }, { "title": "MineValiCoder: Reliable Code Generation with Test Case Quality Mining and Bipartite Graph-Based Mutual Validation", "url": "https://arxiv.org/abs/2607.22471v1", "source_name": "arXiv", "published_at": "2026-07-24T16:39:06+00:00", "summary": "MineValiCoder filters generated tests, refines several code candidates, then scores code and tests together. The pattern targets a core agent risk: unreliable tests steering implementation." }, { "title": "MedGame: Storytelling Gamification Empowered by Large Language Models for Medical Education", "url": "https://arxiv.org/abs/2607.21570v1", "source_name": "arXiv", "published_at": "2026-07-23T17:50:28+00:00", "summary": "MedGame turns static clinical cases into executable decision stories with separate narrative and orchestration stages, a useful architecture pattern for case-grounded learning agents." }, { "title": "The Unreasonable Effectiveness of Separating the Task from the Model — Maxime Rivest & Isaac Miller", "url": "https://www.youtube.com/watch?v=GgLQ02aO-hs", "source_name": "AI Engineer", "published_at": "2026-07-23T17:45:06+00:00", "summary": "DSPy separates an AI task’s contract from its model, prompt, tools, and harness, letting builders swap or optimize implementations while holding behavior against the same evals." }, { "title": "OpenForgeRL: Train Harness-native Agents in Any Environment", "url": "https://arxiv.org/abs/2607.21557v1", "source_name": "arXiv", "published_at": "2026-07-23T17:38:30+00:00", "summary": "OpenForgeRL trains agents inside their deployment harnesses by proxying model calls into standard RL stacks and isolating rollouts in Kubernetes containers; error recovery remains weak." }, { "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": "Do AI Agents Know When a Task Is Simple? Toward Complexity-Aware Reasoning and Execution", "url": "https://arxiv.org/abs/2607.13034v1", "source_name": "arXiv", "published_at": "2026-07-14T17:59:31+00:00", "summary": "E3 makes agents estimate task scope, try the minimum viable path, and expand only after verification fails. In a controlled edit benchmark, it preserved task completion while sharply reducing work." }, { "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": "In Code They Act, In Proof We Trust — Erik Meijer, Leibniz Labs", "url": "https://www.youtube.com/watch?v=-CnA2lGfymY", "source_name": null, "published_at": null, "summary": "A proposed agent harness separates planning from execution, represents tool calls as inspectable programs, and requires machine-checkable safety proofs before side effects run." }, { "title": "Don't Build Agents You Can't Answer For — Addy Osmani", "url": "https://www.youtube.com/watch?v=n97BCfyFIvw", "source_name": null, "published_at": null, "summary": "Osmani argues that agent throughput makes evidence-backed judgment the engineering bottleneck: builders should require explainable diffs, tests, logs, and explicit ownership before shipping." } ] }