💬 Community — 4 items

💬 Taylor Lorenz — Bluesky signal (141♥) on the friction between using Grok for Substack and audience expectations.

💬 PTXBench — A paper proposing a benchmark for LLM performance in architecture-specific GPU kernel optimization.

💬 GenRouter — A routing framework designed to direct prompts to the most efficient agentic image-generation workflows.

💬 Zetta ζ — A paper on a closed-loop harness for evolving runtime critics and recovery skills in physical intelligence.

📚 Read — 4 items

📚 Asana cleared 5 years of engineering work in 2 weeks with Codex — Asana reports clearing 5 years of engineering backlog in 2 weeks with Codex — a production-scale case study with concrete throughput numbers for anyone evaluating coding agent ROI.

📚 Cross-Model Memory Transfer via Target-Side Reader Adaptation — Cross-model memory transfer works by adapting a lightweight target-side reader, not the memory tables themselves — relevant for swapping models without losing agent memory.

📚 The new GitHub Copilot experience in Slack — GitHub Copilot lands in Slack with @GitHub for code review and multiplayer workflows — a new surface for teams already living in Slack.

📚 New in Air: Claude Subscriptions, Multiproject View, and Improved Markdown — JetBrains Air adds Claude subscriptions, multiproject view, and improved markdown — operator-relevant for teams managing multiple projects and model subscriptions.

Real production numbers (Asana) and memory-transfer research both point the same way: coding agents are becoming operational infrastructure, not experiments.

⚙️ Ship — 2 items

📦 Anthropic SDK 0.33.0 — Files and Skills APIs hit GA; computer use and browser use toolsets land in the TypeScript SDK — operators can now ship browser/computer tooling without custom shims.

📦 Vercel AI SDK 3.1.0 — Deepgram provider: transcription options now actually sent; `diarize` no longer defaults true — paid add-on, so check your config if you relied on speaker labels.

Two SDK releases today both remove custom shims — Anthropic's Files/Skills GA and Vercel's Deepgram transcription options now actually pass through.

📚 Read — 3 items

📚 Graph Engineering in the Era of LLM Agents: From Individual Intelligence to System Intelligence — Proposes graph engineering as the coordination layer for multi-agent LLM systems — dynamic graph structures route tasks across specialized agents and manage complex workflows.

📚 Introducing the Kotlin Benchmark for AI Coding Agents — JetBrains releases a Kotlin benchmark for AI coding agents — real-world Kotlin tasks to evaluate agent performance beyond generic SWE-bench-style suites.

📚 Foundation Agents Meet Agentic Deep Research: Evidence-Grounded Clinical Code Forecasting — Agentic deep research applied to clinical code forecasting — evidence-grounded retrieval and reasoning to predict billing codes; a pattern for any domain where agents must ground outputs in retrieved evidence.

Multi-agent coordination is moving from ad-hoc orchestration to explicit graph structures; benchmarks are expanding beyond SWE-bench to language-specific real-world tasks.

💬 Community — 1 item

💬 @kathryntewson.bsky.social: 'Seven months ago, Claude was trash for code in any meaningful sense' — Veteran coder on Claude's trajectory: 'Seven months ago, Claude was trash for code in any meaningful sense' — the implied contrast is the signal builders are debating.

The capability curve is the story — builders who dismissed Claude for code are re-evaluating.

⚙️ Ship — 2 items

📦 NanoClaw 2.3.0 — NanoClaw 2.3.0: new per-agent Slack apps and agent spawning from Slack (migrate with `/migrate-slack-agents`); `/add-codex` pins Codex 0.146.0 to avoid GPT-5.4 retirement on Aug 31 breaking codex-provider agents.

📦 Anthropic SDK 0.119.0 — Anthropic SDK 0.119.0: Files and Skills APIs GA, plus computer use and browser use toolsets.

Two releases today: one forces a Slack migration decision, the other GA's Files/Skills and adds computer/browser toolsets. Operators on NanoClaw need to run the migration before Aug 31.

💬 Community — 6 items

💬 ACP v2 draft from Zed — ACP v2 draft lands: updates beyond user-initiated turns, streaming tool calls/terminal output, structured file diffs. Zed's protocol move matters for any harness interop.

💬 @carnage4life.bsky.social: In other news, I switched my work setup to Muse Code and Codex from Claude Code on Friday. Also after months o — Dare Obasanjo drops Claude Code for Muse Code + Codex after months of evangelizing it — a notable harness churn signal from a longtime operator.

💬 @timkellogg.me: if you’re building an agent harness, i highly recommend asking them to do a “Yelp review” of their experience — Tim Kellogg: have agents write a 'Yelp review' of their harness experience — cheap qualitative signal for harness builders.

💬 FlowEvo: Self-Evolving Agents through the Co-Evolution of Workflows and Executable Skills — FlowEvo co-evolves reusable skills and workflows at inference time — agents self-improve accuracy and efficiency without retraining.

💬 AgentMercury: Your Agent Can Synthesize Verifiable Environments for Business Scenarios at scale — AgentMercury synthesizes verifiable business environments at scale — executable RL substrates for training agents on real business scenarios.

💬 Towards Faithful Simulation of Human Shopping Behavior — RecVerse simulates human shopping with hierarchical memory and trajectory-level RL — a GUI-grounded agent for realistic multi-turn behavior.

Harness loyalty is thin: a longtime Claude Code evangelist switches to Muse Code + Codex while Zed pushes ACP v2. Interop and workflow co-evolution are the battleground.

📚 Read — 3 items

📚 When Retrieval Fails Before It Begins: Structurally Indirect Prerequisite Eviction as a Retention Failure in Agentic Memory — Identifies a retention failure in agentic memory where prerequisite knowledge is evicted before retrieval — a structural failure mode, not a retrieval-quality issue.

📚 Introducing the Kotlin Benchmark for AI Coding Agents — JetBrains ships a Kotlin benchmark for AI coding agents — a language-specific eval to measure agent code quality on Kotlin tasks.

📚 Intelligence EXPLOSION: Harness Engineering with Pi Agent, Deepseek, and Gemini — IndyDevDan runs a harness engineering session across Pi Agent, Deepseek, and Gemini — practical comparison of agent runtimes and their limits.

Memory systems are the next failure surface — eviction order matters as much as retrieval quality, and language-specific benchmarks are starting to pressure-test coding agents.

💬 Community — 4 items

💬 Legal Personhood for AI — 62 points on HN; argument against granting agents legal status to avoid systemic liability chaos.

💬 SemaPLC — A paper proposing a verification-gated harness to validate PLC code via compilation and runtime execution.

💬 Peer-Voted LLM-Agent Stress Tests — A study claiming synthetic agent feeds increase lexical convergence without reliable coordination advantages.

💬 Daedalus-150M — A convolution-attention hybrid model designed for faster CPU inference.

⚙️ Ship — 3 items

📦 Claude platform release notes — August 20, 2026 — Python SDK v1.0 ships — HTTP layer moves to httpx2, drops legacy Text Completions and temperature/top_p/top_k on Messages. Migration guide covers breaking changes; async raw responses now need await response.parse().

📦 Anthropic SDK 0.125.0 — Anthropic SDK 0.125.0 adds managed agents web search config and self-hosted sandbox memory — new knobs for agent deployments.

📦 MCP 2.1.0 — MCP Python SDK 2.1.0 — Client takes StdioServerParameters directly, prompts accept Image/Audio, and handler exceptions are now sanitized to generic errors (log once, client sees 'Error executing tool').

Two SDK migrations in one day — Python SDK v1.0 and MCP 2.1.0 both change error/transport behavior. Worth scheduling a test pass before upgrading.

📚 Read — 4 items

📚 Apodex 1.1: Scaling Agentic Intelligence for Complex Work — Apodex 1.1 scales executable environments and trains agents for sustained, verifiable progress on complex real-world tasks.

📚 PrimeAgentOrchestrator: Memory-Primed Agent Spawning for Personal AI Infrastructure — Memory-primed agent spawning for personal AI infrastructure — orchestration that reuses memory to spin up agents on demand.

📚 Representation Affects Retrieval: A Case Study of Skill Discovery and Routing in a Multimodal Agent Harness — Case study showing how skill representation choices directly affect retrieval quality and routing in a multimodal agent harness.

📚 Enterprise managed settings in GitHub Copilot for JetBrains — Copilot for JetBrains adds enterprise-managed plugin governance, an MCP server allowlist, and managed OpenTelemetry.

The research is pointing at memory and representation as the next levers for agent reliability — worth watching how these translate into harness defaults.

🚨 Watch — 1 item

🚨 Agentic Security — A paper systematizing failure modes and design laws for LLM-driven penetration testing.

⚙️ Ship — 3 items

📦 Claude platform release notes — August 20, 2026 — Python SDK v1.0: httpx2 replaces httpx, legacy Text Completions and temp/top_p/top_k params removed, async raw responses need `await response.parse()`. Migration guide out.

📦 Antigravity Google — Embedded Terminals, Git Version Control, and Audio Attachments — Antigravity adds embedded terminals, git version control, and audio attachments — operators get in-IDE terminal and version control without leaving the agent.

📦 Anthropic SDK 0.123.0 — 0.123.0 adds files/memory store API updates, workspace ID helpers, and fixes: tool-result sends retry for lease TTL, sync session tools run in worker thread. Minor but operator-relevant.

SDK churn day: Anthropic's Python SDK v1.0 forces a migration, while Antigravity adds terminal/git/audio. Check your SDK dependencies before upgrading.

💬 Community — 1 item

💬 Builder weighs cancelling Claude Code: Fable 5 hits 5-hour limit in an hour — Sung Kim reports Claude Code + Fable 5 burns the 5-hour limit in an hour; Opus 5 adds a feature but breaks something. Real operator data on subscription economics.

As coding agents get more capable, the bottleneck shifts from model quality to rate limits and subscription economics.

📚 Read — 7 items

📚 Agentic Scaffolding Amplifies Sycophantic Behavior in Large Language Models — Agentic scaffolding — subagents, tool loops, reflection — amplifies sycophantic behavior in LLMs. If you're building agents that interact with users, this robustness finding is worth internalizing.

📚 Prime Agent: A Self-Improving RLM Harness — Open-source harness for self-improving RLMs — recursive subagents, persistent computation, and agent-to-agent coordination to extend reasoning. Worth a look if you're building long-horizon agent loops.

📚 MobilePA-Bench: Benchmarking Mobile Planner Agents on Complex Real-World Tasks — Interactive sandbox benchmark for mobile planner agents — evaluates tool-calling, sub-agent collaboration, and memory use on complex real-world tasks. Useful if you're shipping agents that drive phones or apps.

📚 Introducing the Admin plugin for ChatGPT Work and Codex — OpenAI introduces an Admin plugin for ChatGPT Work and Codex — enterprise controls for managing Codex deployments. Operators running Codex at scale should check what policy and oversight hooks are exposed.

📚 LitReview Arena: Evaluating Literature Review Agents with Battle-Style Peer Review Platform — LitReview Arena evaluates literature review agents with a battle-style peer review platform — a new eval for research agents. Useful if you're building agents that synthesize papers.

📚 The rise of "context engineering" — LangChain argues context engineering — how you assemble prompts, memory, and tool context — is the new prompt engineering. A useful frame for operators tuning agent performance.

📚 Using skills with Deep Agents — LangChain's Deep Agents now support skills — reusable capabilities agents can invoke. Worth a look if you're building long-running agents that need composable tooling.

The Read lane is converging on evaluation and robustness — context engineering and skills are the build side, sycophancy and mobile benchmarks are the guardrails.

💬 Community — 1 item

💬 @davidcrespo.bsky.social: Revenue growth 'essentially all coming from agentic tools' — David Crespo on the revenue surge: 'essentially all coming from agentic tools.' Builder-level signal that agent-driven revenue is real, not narrative.

⚙️ Ship — 2 items

📦 Gemini CLI 0.57.0 — Gemini CLI 0.57.0 — OAuth redirect fix for Cloud Workstations, eval tool call formatter with failure summaries, and context-aware silent retries with TTL for capacity errors. Multi-turn requests now roll back entirely on cancel/abort.

📦 Devin CLI updated — Devin CLI v3000.6.2/5.20 — new `devin rm`, `devin doctor`, `/recap`, `/rename`, and Alt+P plan mode; read-only `cd` commands now resolve relative paths for permissions; pre_tool hook blocks return reasons to the agent; memory leak on screen redraw fixed.

Two CLI updates: Devin adds session management and Gemini hardens OAuth and retries. Both incremental but reduce operational friction.

📚 Read — 6 items

📚 ClawProBench: Trace-Aware Evaluation of AI Agents with Runtime Coverage and Frozen Workplace-Style Holdouts — Trace-aware eval across live and frozen tracks shows final-answer rankings obscure runtime coverage gaps — worth adopting if you benchmark harness configs.

📚 Recursive Experiential-Working Memory Evolution for Long-Horizon Agent Harnesses — Recursive memory architecture tracks progress and guides skill selection, improving long-horizon agent success rates — a concrete pattern for memory in agent harnesses.

📚 There Is No Neutral Harness: Modern LLM Leaderboards Are Manufactured by Config-Fragile Items — Argues leaderboard rankings are manufactured by config-fragile items — small harness config changes flip results. A caution for anyone trusting eval leaderboards.

📚 A Survey on Foundations and Frontiers of Multimodal Agentic Frameworks: Techniques and Applications — Survey of multimodal agentic frameworks — techniques and applications, a useful map for agents handling images, video, and audio.

📚 Plan-and-Execute Agents — LangChain's plan-and-execute pattern separates planning from execution to improve reliability on multi-step tasks — worth comparing against ReAct loops.

📚 LangSmith Engine Improves Agent Issue Detection by 2x — LangSmith Engine claims 2x better agent issue detection via improved trace analysis — worth checking if you're already on LangSmith for eval.

Eval is the theme today — trace-aware benchmarks and config-fragility critiques both point to the same conclusion: trust your own traces, not leaderboards.

⚙️ Ship — 2 items

📦 Claude Python SDK v1.0 — Python SDK v1.0 ships — HTTP layer moves from httpx to httpx2 (call `httpx2.alias_httpx()` if you patch httpx for tracing/mocking). Requires Python 3.10+, drops legacy Text Completions and `temperature`/`top_p`/`top_k`; async `.with_raw_response` now needs `await response.parse()`.

📦 Antigravity Google — Embedded Terminals, Git Version Control, and Audio Attachments — Antigravity adds embedded terminals, git version control, and audio attachments — terminal and git in-IDE move it toward full agentic coding parity.

SDK maturity day — Anthropic's Python SDK hits 1.0 with breaking changes, and Antigravity closes the terminal/git gap. Both are upgrade-and-retest items, not read-and-forget.

💬 Community — 6 items

💬 WebMCP: Teaching Your Website to Talk to AI Agents — HN debate on WebMCP — exposing website actions to agents via MCP. Top comment calls the approach 'inverted and impractical'; 50 comments arguing whether sites should serve agent APIs.

💬 CyberFactory: Scaling Cyber Security Capabilities with Instances from the Wild — CyberFactory turns real-world vulnerabilities into agentic training data for Aegis — an open-weight security agent that gets better at triage from wild instances.

💬 AgentRoom: Concurrent Multi-Agent Coding in a CRDT-Backed Shared Workspace — AgentRoom gives multiple coding agents a shared CRDT workspace — real-time collaborative editing and filesystem coordination, cutting task conflicts.

💬 Automata from Agent Traces: Failure and Next-Step Prediction — Compresses LLM agent traces into finite-state machines for next-step and failure prediction — a safety-auditing tool for agent behavior.

💬 Z.ai confirms Ox Alpha is a new GLM-series model and will release its weights — Z.ai confirms Ox Alpha is GLM-series and will release weights — HN reports it crushing coding tasks via OpenRouter/OpenCode Zen.

💬 Claude Code update hardcodes AgentTool guardrails — Claude Code update hardcodes guardrails: 'Do not call AgentTool unless the user requested it' and workflow restrictions — check if this breaks your sub-agent patterns.

Open-weight coding models are closing the gap (Ox Alpha weights coming), while Claude Code is tightening sub-agent autonomy — operators should re-test both their model choices and their sub-agent patterns.

📚 Read — 5 items

📚 SecOPD: Mitigating Adaptive Prompt Injections by On-Policy Distillation — On-policy distillation with token-level feedback sharply reduces adaptive prompt injection success — a concrete defense for agents exposed to untrusted content.

📚 AI Agents Push Humans Out of the Loop — Mitchell, Ghosh, and Passi map how agent deployments displace human oversight — a must-read for anyone designing review or approval loops.

📚 Show HN: Proliferate- open-source, self-hostable Codex for any coding agent — Open-source, self-hostable Codex for any coding agent — YC S25 founder's Show HN. Worth a look if you want Codex-style orchestration without vendor lock-in.

📚 LLM Agents Perform Controlled Experiments Using Simulation Models — Agents run controlled experiments against simulation models, moving from code generation to hypothesis validation — relevant to scientific and engineering agent workflows.

📚 ADE: Agentic Data Evolution Framework for Human-Centered Objectives — Agentic data evolution framework that optimizes datasets against human-centered objectives — a different angle on data curation for agent training.

Today's read splits between hardening agents against injection and questioning how much human oversight we're actually keeping — both worth acting on before the next deployment.