{ "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" } }, { "schema_version": "1.1", "id": "auto-2e74d0228d", "slug": "addyosmani-agent-skills-2e74d0228d", "url": "https://feed7.dev/p/addyosmani-agent-skills-2e74d0228d", "title": "addyosmani/agent-skills", "why_included": "Turn specs, TDD, review, and shipping into evidence-gated workflows, but adopt only the skills that fit your repository.", "summary": "This pack turns common engineering practices into portable coding-agent workflows for specs, TDD, review and shipping. Its useful idea is to require evidence at each gate, not merely better prompts.", "practical_implication": "Treat the pack as a menu of enforceable workflows: begin with TDD, debugging or code review, then add broader lifecycle automation after checking how each skill fits your repository. Codex can load it as a native plugin from CLI v0.122+.", "agent_context": "The repository contains **24 Markdown skills**, four specialist personas and **8 lifecycle commands** spanning definition, planning, implementation, verification, review and release. The open skills CLI supports installation across **70+ agents**.\n\nTreat the pack as a menu of enforceable workflows: begin with TDD, debugging or code review, then add broader lifecycle automation after checking how each skill fits your repository. Codex can load it as a native plugin from **CLI v0.122+**.\n\nInstalling an individual skill may omit shared files under the repository-level references directory, leaving supplementary checklists unavailable. The workflows also encode strong process opinions, so teams should review their gates and defaults instead of adopting all 24 blindly.", "source": { "name": "GitHub", "url": "https://github.com/addyosmani/agent-skills", "published_at": "2026-08-09T00:00:00.000Z" }, "source_class": "tool", "content_type": "GitHub Repo", "layer": "agent", "domains": [ "coding" ], "topics": [ "skills", "coding-agents", "harness-engineering" ], "verification": { "status": "needs_review", "label": "Needs Review", "method": "unverified", "verified_at": null }, "uncertainty": [ "Automatically selected from source material; feed7 has not independently tested the claim." ], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-09T00:00:00.000Z", "modified_at": "2026-08-09T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/addyosmani-agent-skills-2e74d0228d", "json": "https://feed7.dev/p/addyosmani-agent-skills-2e74d0228d.json", "markdown": "https://feed7.dev/p/addyosmani-agent-skills-2e74d0228d.md" } }, { "schema_version": "1.1", "id": "auto-f5a1122448", "slug": "the-bitter-lesson-of-tool-calling-f5a1122448", "url": "https://feed7.dev/p/the-bitter-lesson-of-tool-calling-f5a1122448", "title": "The Bitter Lesson of Tool Calling", "why_included": "Test typed Python tool stubs: they matched or beat JSON calls in 11 of 14 models on BFCL v4.", "summary": "Across BFCL v4, models usually handled tools as typed Python calls at least as well as native JSON, suggesting code-based orchestration is worth testing for capable coding agents.", "practical_implication": "If your agents already write reliable code, test a typed-stub execution layer that lets one turn chain or parallelize calls. It also matched or exceeded JSON in 13 of 14 models under parallel fan-out and stayed stable in the reported context-rot condition.", "agent_context": "The study compares programmatic tool calling with native JSON calls across **14 models** on BFCL v4. Python-stub calls matched or beat JSON in **11 of 14 models**, while the GPT-5.6 family improved by **10.6%**.\n\nIf your agents already write reliable code, test a typed-stub execution layer that lets one turn chain or parallelize calls. It also matched or exceeded JSON in **13 of 14 models** under parallel fan-out and stayed stable in the reported context-rot condition.\n\nPerformance tracked model capability, so programmatic calls are not automatically better for every model. The evidence comes from one established function-calling benchmark; production safety, debugging, sandboxing, and task-level cost still need separate evaluation.", "source": { "name": "arXiv", "url": "https://arxiv.org/abs/2608.06370v1", "published_at": "2026-08-06T00:00:00.000Z" }, "source_class": "blog_post", "content_type": "Paper", "layer": "benchmark", "domains": [ "coding" ], "topics": [ "tool-use", "agent-evals", "harness-engineering" ], "verification": { "status": "needs_review", "label": "Needs Review", "method": "unverified", "verified_at": null }, "uncertainty": [ "Automatically selected from source material; feed7 has not independently tested the claim." ], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-06T00:00:00.000Z", "modified_at": "2026-08-06T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/the-bitter-lesson-of-tool-calling-f5a1122448", "json": "https://feed7.dev/p/the-bitter-lesson-of-tool-calling-f5a1122448.json", "markdown": "https://feed7.dev/p/the-bitter-lesson-of-tool-calling-f5a1122448.md" } }, { "schema_version": "1.1", "id": "auto-b3425f8b3f", "slug": "how-cursor-router-chooses-the-right-model-for-the-task-b3425f8b3f", "url": "https://feed7.dev/p/how-cursor-router-chooses-the-right-model-for-the-task-b3425f8b3f", "title": "How Cursor Router chooses the right model for the task", "why_included": "Route models using real task outcomes, corrections, cache costs, confidence thresholds, and an explicit per-turn budget.", "summary": "Cursor Router learns task complexity and model fit from production behavior, showing why agent routing should include correction signals, cache costs, and per-task performance.", "practical_implication": "Builders implementing model routing should learn from real task outcomes, including corrections and task progression, rather than rely only on benchmarks. Include token use, cache misses, model-switching costs, confidence thresholds, and an explicit per-turn budget.", "agent_context": "Cursor routes each turn in two stages: Compass estimates complexity, then a production-derived taxonomy selects among eligible frontier models. **Auto Intelligence costs 68% less than Fable**, while **Auto Balance costs 41% less than Opus 4.8** and reports higher satisfaction.\n\nBuilders implementing model routing should learn from real task outcomes, including corrections and task progression, rather than rely only on benchmarks. Include token use, cache misses, model-switching costs, confidence thresholds, and an explicit per-turn budget.\n\nCursor's satisfaction measure is inferred from subsequent user behavior, so it is a proxy rather than a direct quality score. The reported results come from Cursor traffic and may not transfer to another workload, user population, or pricing mix.", "source": { "name": "Cursor", "url": "https://cursor.com/blog/how-cursor-router-works", "published_at": "2026-08-06T00:00:00.000Z" }, "source_class": "blog_post", "content_type": "Engineering Post", "layer": "agent", "domains": [ "coding", "data" ], "topics": [ "model-selection", "harness-engineering", "agent-reliability" ], "verification": { "status": "official_source", "label": "Official Source", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-06T00:00:00.000Z", "modified_at": "2026-08-06T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/how-cursor-router-chooses-the-right-model-for-the-task-b3425f8b3f", "json": "https://feed7.dev/p/how-cursor-router-chooses-the-right-model-for-the-task-b3425f8b3f.json", "markdown": "https://feed7.dev/p/how-cursor-router-chooses-the-right-model-for-the-task-b3425f8b3f.md" } }, { "schema_version": "1.1", "id": "auto-31ceb9467c", "slug": "quotebench-how-matched-scores-can-hide-command-path-fail-31ceb9467c", "url": "https://feed7.dev/p/quotebench-how-matched-scores-can-hide-command-path-fail-31ceb9467c", "title": "QuoteBench: How Matched Scores Can Hide Command-Path Failures", "why_included": "Test commands through the exact production transport and verify final state, since one parser cut task completion by 55.4–73.2 points.", "summary": "QuoteBench shows that shell-command scores can conceal failures introduced by serialization and reparsing. Agent evals should identify the execution path, not attribute every result to the model.", "practical_implication": "When the boundary was disclosed, six configurations recovered 30.4–60.7 points. Builders should test generated commands through the exact production transport and validate final state, especially where wrappers interpolate or reparse shell text.", "agent_context": "QuoteBench tests **56 one-shot tasks** from 14 incident-derived families across eight configurations. Adding one unescaped parser cut replayed-command success by **55.4–73.2 percentage points**.\n\nWhen the boundary was disclosed, six configurations recovered **30.4–60.7 points**. Builders should test generated commands through the exact production transport and validate final state, especially where wrappers interpolate or reparse shell text.\n\nAdaptation was uneven: two configurations recovered nothing or declined slightly. GPT-5.6-sol’s **−3.6-point matched gap** concealed −64.3 points of transport damage plus 60.7 points of compensation, so aggregate scores can misstate both model and harness quality.", "source": { "name": "arXiv", "url": "https://arxiv.org/abs/2608.13547v1", "published_at": "2026-08-13T00:00:00.000Z" }, "source_class": "blog_post", "content_type": "Paper", "layer": "benchmark", "domains": [ "coding", "security" ], "topics": [ "agent-evals", "benchmark-integrity", "harness-engineering" ], "verification": { "status": "needs_review", "label": "Needs Review", "method": "unverified", "verified_at": null }, "uncertainty": [ "Automatically selected from source material; feed7 has not independently tested the claim." ], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-13T00:00:00.000Z", "modified_at": "2026-08-13T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/quotebench-how-matched-scores-can-hide-command-path-fail-31ceb9467c", "json": "https://feed7.dev/p/quotebench-how-matched-scores-can-hide-command-path-fail-31ceb9467c.json", "markdown": "https://feed7.dev/p/quotebench-how-matched-scores-can-hide-command-path-fail-31ceb9467c.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" } }, { "schema_version": "1.1", "id": "auto-a44214149b", "slug": "use-acp-compatible-harnesses-with-the-ai-sdk-harness-lay-a44214149b", "url": "https://feed7.dev/p/use-acp-compatible-harnesses-with-the-ai-sdk-harness-lay-a44214149b", "title": "Use ACP-compatible harnesses with the AI SDK harness layer", "why_included": "Use ACP for portable harness integration, but test permissions, events, and runtime-specific controls before replacing a direct adapter.", "summary": "A new ACP meta-adapter lets HarnessAgent run compatible runtimes that lack dedicated integrations. Keep direct Claude Code and Codex adapters where tighter behavior matters.", "practical_implication": "Use the meta-adapter to connect a compatible harness that has no native AI SDK integration. For Claude Code and Codex, Vercel recommends their dedicated adapters because they can expose runtime behavior more precisely.", "agent_context": "The new **@ai-sdk/harness-acp** package adapts the Agent Client Protocol rather than one runtime. Builders pass an ACP-compatible package to **createACP**, map the harness basics, and use the result with HarnessAgent.\n\nUse the meta-adapter to connect a compatible harness that has no native AI SDK integration. For **Claude Code and Codex**, Vercel recommends their dedicated adapters because they can expose runtime behavior more precisely.\n\nACP is not universal, and its abstraction may limit or alter access to a harness’s internals. Treat protocol compatibility as a portability option, then test permissions, events, and runtime-specific behavior before replacing a direct adapter.", "source": { "name": "Vercel", "url": "https://vercel.com/changelog/use-acp-compatible-harnesses-with-the-ai-sdk-harness-layer", "published_at": "2026-08-13T00:00:00.000Z" }, "source_class": "blog_post", "content_type": "Engineering Post", "layer": "tools", "domains": [ "coding" ], "topics": [ "coding-agents", "agent-sdks", "harness-engineering" ], "verification": { "status": "official_source", "label": "Official Source", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-13T00:00:00.000Z", "modified_at": "2026-08-13T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/use-acp-compatible-harnesses-with-the-ai-sdk-harness-lay-a44214149b", "json": "https://feed7.dev/p/use-acp-compatible-harnesses-with-the-ai-sdk-harness-lay-a44214149b.json", "markdown": "https://feed7.dev/p/use-acp-compatible-harnesses-with-the-ai-sdk-harness-lay-a44214149b.md" } }, { "schema_version": "1.1", "id": "auto-137531afb5", "slug": "apache-maka-137531afb5", "url": "https://feed7.dev/p/apache-maka-137531afb5", "title": "apache/maka", "why_included": "Use its append-only run log and shared runtime to trim context without losing evidence, inspect failures, and recover crashes.", "summary": "Maka gives local coding-agent runs a durable event log, sandbox boundary, recovery path, and shared runtime across desktop, CLI, and evals. It is early, source-only software with changing interfaces.", "practical_implication": "Builders can treat prompts as projections of durable run history: trim old tool output from context without deleting evidence, inspect failures, recover crashes, and compare reproducible experiment cells. Its graph mode also uses isolated Git worktrees for parallel implementation.", "agent_context": "Apache Maka records model messages, tool calls, results, permission decisions, and termination events in an **append-only local log**. Desktop, CLI, TUI, and evaluations execute through one **Runtime Host**, with sandbox approval for file writes and shell commands.\n\nBuilders can treat prompts as projections of durable run history: trim old tool output from context without deleting evidence, inspect failures, recover crashes, and compare reproducible experiment cells. Its graph mode also uses isolated Git worktrees for parallel implementation.\n\nThis is an **Apache Incubator** project with no approved Apache release. The desktop build currently supports **Apple Silicon macOS**; formats and commands may change, Windows is an unsigned preview, and secrets remain plaintext to the local OS account.", "source": { "name": "GitHub", "url": "https://github.com/apache/maka", "published_at": "2026-08-23T00:00:00.000Z" }, "source_class": "tool", "content_type": "GitHub Repo", "layer": "agent", "domains": [ "coding" ], "topics": [ "harness-engineering", "agent-reliability", "sandboxing" ], "verification": { "status": "needs_review", "label": "Needs Review", "method": "unverified", "verified_at": null }, "uncertainty": [ "Automatically selected from source material; feed7 has not independently tested the claim." ], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-23T00:00:00.000Z", "modified_at": "2026-08-23T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/apache-maka-137531afb5", "json": "https://feed7.dev/p/apache-maka-137531afb5.json", "markdown": "https://feed7.dev/p/apache-maka-137531afb5.md" } }, { "schema_version": "1.1", "id": "auto-b53948be9f", "slug": "give-the-agent-a-budget-not-a-token-sachin-malhotra-anth-b53948be9f", "url": "https://feed7.dev/p/give-the-agent-a-budget-not-a-token-sachin-malhotra-anth-b53948be9f", "title": "Give the Agent a Budget, Not a Token — Sachin Malhotra, Anthropic", "why_included": "Bound write volume, rate, reversibility, and visibility, with human-only overrides outside the agent session.", "summary": "Production agents need bounded write authority, infrastructure-stamped identity, and human-only overrides; a broad token and tool list cannot control blast radius.", "practical_implication": "Classify write verbs by how failures surface, rate-limit every write, and keep overrides outside agent sessions. Record aggregate behavior with trip wires, size limits using an undo test, and let a trusted proxy stamp agent and session identity.", "agent_context": "A cleanup agent’s empty pipeline stage removed its filter, matching and deleting about **200 workloads** used by **20 engineers** in **90 seconds**. The proposed alternative treats authority as budgets across action volume, rate, reversibility, and visibility.\n\nClassify write verbs by how failures surface, rate-limit every write, and keep overrides outside agent sessions. Record aggregate behavior with trip wires, size limits using an undo test, and let a trusted proxy stamp agent and session identity.\n\nThese are production safety primitives, not a guarantee against bad actions. Limits still need tuning by resource and namespace, while quiet or irreversible operations may require a separate human-held credential.", "source": { "name": "AI Engineer", "url": "https://www.youtube.com/watch?v=rbjWzZK2LU0", "published_at": "2026-08-22T00:00:00.000Z" }, "source_class": "video", "content_type": "Video", "layer": "agent", "domains": [ "coding", "security" ], "topics": [ "harness-engineering", "tool-use", "sandboxing" ], "verification": { "status": "source_linked", "label": "Source Linked", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-22T00:00:00.000Z", "modified_at": "2026-08-22T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/give-the-agent-a-budget-not-a-token-sachin-malhotra-anth-b53948be9f", "json": "https://feed7.dev/p/give-the-agent-a-budget-not-a-token-sachin-malhotra-anth-b53948be9f.json", "markdown": "https://feed7.dev/p/give-the-agent-a-budget-not-a-token-sachin-malhotra-anth-b53948be9f.md" } }, { "schema_version": "1.1", "id": "auto-105166eadd", "slug": "ai-agents-are-just-distributed-systems-now-salman-munaf--105166eadd", "url": "https://feed7.dev/p/ai-agents-are-just-distributed-systems-now-salman-munaf--105166eadd", "title": "AI Agents Are Just Distributed Systems Now — Salman Munaf, TikTok", "why_included": "Make mutating tools idempotent, persist each step, bound retries and spend, and treat timeouts as unknown outcomes.", "summary": "Once agents mutate external state, timeouts mean unknown outcomes. Builders need idempotent tools, bounded retries, scoped credentials, durable traces, and explicit recovery paths.", "practical_implication": "Design every agent tool like a distributed-system boundary. Persist each step, define the source of truth, treat memory as an invalidatable cache, and bind approvals to the exact action, actor, timestamp, parameters, and expiration.", "agent_context": "A tool timeout does not prove failure: the remote side may already have committed the action. The talk recommends request IDs, **idempotency keys**, status lookups, circuit breakers, compensation operations, and limits on turns, spend, and parallel calls.\n\nDesign every agent tool like a distributed-system boundary. Persist each step, define the source of truth, treat memory as an invalidatable cache, and bind approvals to the exact action, actor, timestamp, parameters, and expiration.\n\nA stronger model can reduce reasoning mistakes but cannot remove network ambiguity, stale state, or adversarial input. Some irreversible actions cannot be truly undone, so compensation and human approval remain domain-specific safeguards.", "source": { "name": "AI Engineer", "url": "https://www.youtube.com/watch?v=hD9-V56FNRI", "published_at": "2026-08-29T00:00:00.000Z" }, "source_class": "video", "content_type": "Video", "layer": "agent", "domains": [ "coding" ], "topics": [ "harness-engineering", "tool-use", "agent-reliability" ], "verification": { "status": "source_linked", "label": "Source Linked", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-29T00:00:00.000Z", "modified_at": "2026-08-29T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/ai-agents-are-just-distributed-systems-now-salman-munaf--105166eadd", "json": "https://feed7.dev/p/ai-agents-are-just-distributed-systems-now-salman-munaf--105166eadd.json", "markdown": "https://feed7.dev/p/ai-agents-are-just-distributed-systems-now-salman-munaf--105166eadd.md" } }, { "schema_version": "1.1", "id": "auto-7a4a8849cf", "slug": "how-to-get-your-org-to-adopt-coding-agents-without-shipp-7a4a8849cf", "url": "https://feed7.dev/p/how-to-get-your-org-to-adopt-coding-agents-without-shipp-7a4a8849cf", "title": "How to Get Your Org to Adopt Coding Agents (Without Shipping Garbage) — Eyal Blum, Figma", "why_included": "Write executable checks before implementation, then split agent plans into independently validated, reviewable phases.", "summary": "Figma’s cautious agent adopters expose missing safeguards, not resistance. Use their feedback to improve verification, plans, review size, and the labeling of AI-generated work.", "practical_implication": "Make verification executable before implementation: use TDD, encode repeated checks deterministically, and split plans into independently validated, reviewable phases. Put the reason for the change first, and use skeptical maintainers’ feedback as the safety roadmap.", "agent_context": "Figma reports uneven agent adoption and says experienced engineers may be slowest because they see codebase risks first. One detailed plan turned an estimated **six weeks of coding into about one week**, including review.\n\nMake verification executable before implementation: use TDD, encode repeated checks deterministically, and split plans into independently validated, reviewable phases. Put the reason for the change first, and use skeptical maintainers’ feedback as the safety roadmap.\n\nThe workflow is still evolving and depends on planning and review effort that raw generation metrics can hide. AI also increases communication volume, so clearly separate human-written intent from generated descriptions to protect scarce reviewer attention.", "source": { "name": "AI Engineer", "url": "https://www.youtube.com/watch?v=5Bn0xro2ol8", "published_at": "2026-08-28T00:00:00.000Z" }, "source_class": "video", "content_type": "Video", "layer": "agent", "domains": [ "coding" ], "topics": [ "coding-agents", "agent-reliability", "harness-engineering" ], "verification": { "status": "source_linked", "label": "Source Linked", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-28T00:00:00.000Z", "modified_at": "2026-08-28T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/how-to-get-your-org-to-adopt-coding-agents-without-shipp-7a4a8849cf", "json": "https://feed7.dev/p/how-to-get-your-org-to-adopt-coding-agents-without-shipp-7a4a8849cf.json", "markdown": "https://feed7.dev/p/how-to-get-your-org-to-adopt-coding-agents-without-shipp-7a4a8849cf.md" } }, { "schema_version": "1.1", "id": "auto-f8c22d41c8", "slug": "cursor-is-now-available-in-the-ai-sdk-harness-layer-f8c22d41c8", "url": "https://feed7.dev/p/cursor-is-now-available-in-the-ai-sdk-harness-layer-f8c22d41c8", "title": "Cursor is now available in the AI SDK harness layer", "why_included": "The HarnessAgent interface can reduce application changes when testing or switching among eight coding-agent harnesses.", "summary": "Vercel’s AI SDK can now run Cursor behind the same HarnessAgent interface as seven other coding agents, reducing application changes when swapping harnesses.", "practical_implication": "Use HarnessAgent as the application boundary if you expect to test or switch among Cursor, Claude Code, Codex, Cline, Deep Agents, Grok Build, OpenCode, and Pi. That keeps agent selection out of the surrounding application code.", "agent_context": "The official **@ai-sdk/harness-cursor** adapter adds Cursor to Vercel’s AI SDK harness layer. It connects through **Agent Client Protocol (ACP)** and joins seven other supported coding-agent harnesses.\n\nUse **HarnessAgent** as the application boundary if you expect to test or switch among Cursor, Claude Code, Codex, Cline, Deep Agents, Grok Build, OpenCode, and Pi. That keeps agent selection out of the surrounding application code.\n\nThe announcement establishes interface compatibility, but does not compare feature parity, behavior, or portability across harnesses. Agent-specific capabilities may still require separate validation.", "source": { "name": "Vercel", "url": "https://vercel.com/changelog/cursor-ai-sdk-harness-adapter", "published_at": "2026-08-27T00:00:00.000Z" }, "source_class": "blog_post", "content_type": "Engineering Post", "layer": "tools", "domains": [ "coding" ], "topics": [ "agent-sdks", "coding-agents", "harness-engineering" ], "verification": { "status": "official_source", "label": "Official Source", "method": "source_feed", "verified_at": null }, "uncertainty": [], "connected_context": null, "lifecycle": "New", "published_at": "2026-08-27T00:00:00.000Z", "modified_at": "2026-08-27T00:00:00.000Z", "supersedes": [], "expires_at": null, "formats": { "html": "https://feed7.dev/p/cursor-is-now-available-in-the-ai-sdk-harness-layer-f8c22d41c8", "json": "https://feed7.dev/p/cursor-is-now-available-in-the-ai-sdk-harness-layer-f8c22d41c8.json", "markdown": "https://feed7.dev/p/cursor-is-now-available-in-the-ai-sdk-harness-layer-f8c22d41c8.md" } } ], "conflicting_sources": [], "superseded_claims": [], "corpus_evidence": [ { "title": "SafeEvolve: Harness-Policy Co-Evolution from Agent Experience for Safety Alignment", "url": "https://arxiv.org/abs/2609.02786v1", "source_name": "arXiv", "published_at": "2026-09-02T16:19:54+00:00", "summary": "SafeEvolve turns completed agent trajectories into reversible harness updates and policy training, linking runtime safeguards with learned multi-step behavior." }, { "title": "Your Agent Just Authorized What?! — Jay Mok & Ben Coumes, Paypal", "url": "https://www.youtube.com/watch?v=vGn6N4-bxBY", "source_name": "AI Engineer", "published_at": "2026-09-01T18:30:06+00:00", "summary": "Agent authorization should scale with stakes and reversibility: logs may cover coding changes, while open-ended payments need scoped, time-bound, independently verifiable mandates." }, { "title": "CordisBench: Can Language Models Reason About Component Lifecycles in Dynamic Agent Harnesses?", "url": "https://arxiv.org/abs/2609.01600v1", "source_name": "arXiv", "published_at": "2026-09-01T17:59:13+00:00", "summary": "CordisBench finds that models lose reliability as harness lifecycle interactions grow, while deterministic execution semantics can replace costly reasoning on controlled cases." }, { "title": "The Rise of Verbal Reinforcement Learning", "url": "https://arxiv.org/abs/2609.01597v1", "source_name": "arXiv", "published_at": "2026-09-01T17:58:18+00:00", "summary": "This survey separates verbal feedback by when it acts and what it changes, helping agent builders distinguish task specification, test-time guidance, and parameter updates." }, { "title": "Multimodal Collaborative Agents for Next-Gen Commerce — Nidhi Kaushik Vyas, Google DeepMind", "url": "https://www.youtube.com/watch?v=AhQpRalYlyg", "source_name": "AI Engineer", "published_at": "2026-09-01T17:00:27+00:00", "summary": "For fuzzy requests, an agent should identify the missing constraint with the most decision value, elicit it in the right modality, then choose a response format suited to the task." }, { "title": "Teaching agents to pay — Anna Spysz, Stripe", "url": "https://www.youtube.com/watch?v=A-zeQiYkmXk", "source_name": "AI Engineer", "published_at": "2026-09-01T16:30:06+00:00", "summary": "Agent checkout needs more than payment tools: enforce user-set limits outside the agent, expose structured merchant data, honor cancellation, and log every decision." }, { "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": "How our agents build on-brand pages with design.md", "url": "https://vercel.com/blog/how-our-agents-build-on-brand-pages-with-design-md", "source_name": "Vercel", "published_at": "2026-08-31T04:00:00+00:00", "summary": "Vercel found that prose alone produced inconsistent agent-made pages, then paired design.md with fixed CSS primitives and repeatable evals to encode brand judgment." }, { "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": "Tribal Dungeons of Global Shipping: AI Agents at Global Scale — Dmitry Buykin, Maersk", "url": "https://www.youtube.com/watch?v=dQ-_i1tZiws", "source_name": "AI Engineer", "published_at": "2026-08-29T17:30:21+00:00", "summary": "Maersk’s production agents depend less on a clever loop than on executable SOPs, bounded tools, replayable traces, and a correction system shared by experts and engineers." }, { "title": "Agents Are Where Microservices Were in 2015 — Roberto Milev & Uday Kanagala, Navan", "url": "https://www.youtube.com/watch?v=32nrHU6zHU8", "source_name": "AI Engineer", "published_at": "2026-08-29T16:30:28+00:00", "summary": "Navan treats production agents as stateful services needing persistence, composable skills, traces, authorization gates, and trajectory-based tests before multi-agent complexity." }, { "title": "The Half Life of Agent Infrastructure — Ben Kus, Box", "url": "https://www.youtube.com/watch?v=sM1iYgz93HI", "source_name": "AI Engineer", "published_at": "2026-08-29T14:30:22+00:00", "summary": "Agent architectures are expiring quickly. Keep model, search, and orchestration choices replaceable, and evaluate platforms by how well they handle repeated change." }, { "title": "AI-Native Organisations Run on Skills: How to Structure and Scale Them — Imad Touil, QuantumBlack", "url": "https://www.youtube.com/watch?v=M05vON8i0aI", "source_name": "AI Engineer", "published_at": "2026-08-28T18:30:04+00:00", "summary": "Shared agent skills can turn team know-how into portable workflows, but unmanaged catalogs create duplication and policy risk. Treat skills as versioned, evaluated dependencies." }, { "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": "On the Maintenance and Co-evolution of Agent Plugins: An Empirical Study of Claude Code Plugin Marketplaces", "url": "https://arxiv.org/abs/2608.28497v1", "source_name": "arXiv", "published_at": "2026-08-28T16:27:35+00:00", "summary": "A study of 8,351 Claude Code plugins finds instruction files and scripts often require coordinated maintenance, treating skills as evolving software rather than static prompts." }, { "title": "Persona-Execution Separation: An Architecture Pattern for Evolving LLM Agents under Execution Audit", "url": "https://arxiv.org/abs/2608.27427v1", "source_name": "arXiv", "published_at": "2026-08-27T17:50:07+00:00", "summary": "Persona-Execution Separation isolates mutable agent behavior from audited, stateful work through a governed bridge. It is an architecture option for multi-user agents under compliance constraints." }, { "title": "How Anthropic Builds: Lessons from Labs — Mike Krieger, Anthropic", "url": "https://www.youtube.com/watch?v=qqrk7CtkuIw", "source_name": "AI Engineer", "published_at": "2026-08-27T16:00:06+00:00", "summary": "Mike Krieger describes shifting from stepwise prompting to goal-level delegation, backed by verification, observability, feature flags, and frequent decisions to continue or stop experiments." }, { "title": "TraceML: An Empirical Analysis of Human-Agent Planning in Machine Learning Development", "url": "https://arxiv.org/abs/2608.26086v1", "source_name": "arXiv", "published_at": "2026-08-26T17:50:13+00:00", "summary": "TraceML exposes why coding agents stall on long ML work: they repeat narrow optimization loops instead of alternating work types, revisiting discarded ideas, and pivoting like experts." }, { "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": "Knowledge Systems: The New GTM Stack — Jeffrey Wang, Exa", "url": "https://www.youtube.com/watch?v=6pbQgnJ9Voc", "source_name": "AI Engineer", "published_at": "2026-08-26T16:00:26+00:00", "summary": "Exa’s internal agent clone combines writing examples, decision evals, and permission-scoped tools—a concrete pattern for encoding judgment without sharing full access." }, { "title": "Building GTM AI Agents: Lessons from Deploying to 6,000 Users — Sait Izmit, Snowflake", "url": "https://www.youtube.com/watch?v=DrTdD-ttjCY", "source_name": "AI Engineer", "published_at": "2026-08-26T15:00:06+00:00", "summary": "Snowflake’s rollout favors narrow, high-accuracy coverage, staged adoption, and log-driven iteration over connecting every data source before launch." }, { "title": "The Building Blocks of GTM Orchestration — Arman Vaziri, Ramp", "url": "https://www.youtube.com/watch?v=VjEP0xqTUI0", "source_name": "AI Engineer", "published_at": "2026-08-26T14:30:38+00:00", "summary": "Ramp’s GTM agents share a customer-data layer, durable execution, tools, and skills. The reusable pattern is to automate one narrow workflow, then extend the same substrate across teams." }, { "title": "GTM Engineering: The Technical Bits — Everett Berry, Clay", "url": "https://www.youtube.com/watch?v=UhCY231d0FQ", "source_name": "AI Engineer", "published_at": "2026-08-26T13:30:31+00:00", "summary": "Clay frames GTM agents as a data and orchestration problem: resolve entities, evaluate providers, isolate agent-written state, and account for delayed syncs before automating decisions." }, { "title": "Agent Frameworks Considered Harmful — Rémi Louf, .txt", "url": "https://www.youtube.com/watch?v=KHudyx5wW3U", "source_name": "AI Engineer", "published_at": "2026-08-22T16:30:39+00:00", "summary": "Reliable background agents need runtime primitives more than elaborate framework abstractions: durable events, queues, versioned inputs, causal traces, diffs, and replay." }, { "title": "Coding Agents Don't Scale Themselves. Neither Do Your Teams. — Patrick Debois, Tessl", "url": "https://www.youtube.com/watch?v=zCJtYuqwm7E", "source_name": "AI Engineer", "published_at": "2026-08-22T16:00:06+00:00", "summary": "Agent adoption becomes a team-systems problem: improve shared context and harnesses, assign platform ownership, and measure fewer human interventions instead of individual prompt speed." }, { "title": "FinOps for AI Agents: Who Spent All the Tokens? — Tisha Chawla & Susheem Koul, Microsoft", "url": "https://www.youtube.com/watch?v=GJX19pNhmSw", "source_name": "AI Engineer", "published_at": "2026-08-22T14:30:07+00:00", "summary": "Run-level policies cut benchmarked agent spend while completing more runs, suggesting builders should steer loops, context, and tool output before enforcing hard caps." }, { "title": "Asymmetric Capacity Allocation in Self-Refinement Pipelines", "url": "https://arxiv.org/abs/2608.21345v1", "source_name": "arXiv", "published_at": "2026-08-21T17:52:17+00:00", "summary": "Self-refinement pipelines need not use equally capable models: invest capacity in generation and revision, while a small critic may preserve gains at lower compute cost." }, { "title": "Re$^3$Cap: Retrieval-Guided Refinement for Image Captioning Enhancement via Reinforcement Learning", "url": "https://arxiv.org/abs/2608.21305v1", "source_name": "arXiv", "published_at": "2026-08-21T17:07:41+00:00", "summary": "Re³Cap uses multimodal retrieval to find caption omissions and hallucinations before refinement, offering a concrete retrieval-and-review pattern for vision agents." }, { "title": "How I automate my own job at Hugging Face using agents — Niels Rogge, Hugging Face", "url": "https://www.youtube.com/watch?v=FLUoowDJg4I", "source_name": "AI Engineer", "published_at": "2026-08-20T15:30:35+00:00", "summary": "Hugging Face automated research-artifact outreach with a CLI, one skill, and a sandbox. The case shows when an agent can replace custom workflow code, but undisclosed automated outreach raises trust questions." }, { "title": "The Era of Compound Engineering — Kieran Klaassen, Every/Cora", "url": "https://www.youtube.com/watch?v=_ehJyfHg1Vk", "source_name": "AI Engineer", "published_at": "2026-08-20T15:00:06+00:00", "summary": "Compound engineering treats each agent mistake and product judgment as reusable context. The useful discipline is to invest in memory and review so the next feature becomes easier, not merely faster." }, { "title": "IT Admin for the AI Workforce — Sarthak Aggarwal, Decawork", "url": "https://www.youtube.com/watch?v=q-WOjZhOMCA", "source_name": "AI Engineer", "published_at": "2026-08-20T14:30:38+00:00", "summary": "Agent instructions are not security boundaries. Put identity, scoped capabilities, policy gates, audit receipts, and revocation outside the model before granting production access." }, { "title": "Prototyping as Leadership: How a CTO Ships with AI Agents — Hursh Agrawal, The Browser Company", "url": "https://www.youtube.com/watch?v=bdHaOXZOhcM", "source_name": "AI Engineer", "published_at": "2026-08-20T14:00:21+00:00", "summary": "A manager can turn fragmented time into agent steering: prepare context, run bounded work overnight, then personally test and review small PRs before involving the team." }, { "title": "Your Agent Evolved. Your Evals Didn't. — Ameya Bhatawdekar, Braintrust", "url": "https://www.youtube.com/watch?v=nxokqOq1imY", "source_name": "AI Engineer", "published_at": "2026-08-20T13:00:06+00:00", "summary": "When an agent moves from fixed chains to model-led orchestration, memory, tools, and sandboxes, its old answer-quality evals cover only part of the system. Evals must follow the architecture." }, { "title": "How to build an AI-Native Health Company — Dan Feng, Maven Clinic", "url": "https://www.youtube.com/watch?v=WJRdLNhrsLQ", "source_name": "AI Engineer", "published_at": "2026-08-19T17:30:19+00:00", "summary": "Maven’s agent-heavy workflow shortens planning, caps reviewable PR size, and tests nondeterministic features repeatedly. The bottleneck shifts from implementation to decisions and verification." }, { "title": "Trading Desks to Clinical Trials: Parallels in Applied Vertical AI — Ayush Bhardwaj, Allos AI", "url": "https://www.youtube.com/watch?v=Yphdry8ttAQ", "source_name": "AI Engineer", "published_at": "2026-08-19T17:00:31+00:00", "summary": "Vertical agents need narrow jobs, proprietary data, observability, and expert judgment. Generic models and self-grading cannot establish whether domain-specific output is actually useful." }, { "title": "Healthcare’s Agent Bytecode: X12 as the Harness for AI Agents — Vasant Kearney, Onlay", "url": "https://www.youtube.com/watch?v=UyyOoJmuATU", "source_name": "AI Engineer", "published_at": "2026-08-19T16:30:32+00:00", "summary": "Healthcare agents can use X12 transactions as a typed contract across calls, portals, and claims. The standard constrains actions and validation, but payer data can still be stale or wrong." } ] }