2026-11-21 A successful search request can still lose useful data inside a local adapter. Trace a missing query field, distinguish absent data from empty results and malformed records, and test the transformation with a sanitized fixture.
中文 2026-11-21 AppSci-Bench evaluates whether a model can elicit requirements, build an agent, create an internal evaluation, and generalize to held-out business tasks.
中文 2026-11-20 OpenAI's Performance Factory shows how to connect performance alerts, agent diagnosis, reviewed fixes, deployment waits, and recovery verification into one stateful workflow.
中文 2026-11-19 Agent-workdays are capacity signals. A practical measurement system separates throughput, validity, intervention, and durable research outcomes.
中文 2026-11-19 Meta's technique propagation case shows why an ML agent needs a stage-by-stage verification contract before a proven improvement can cross a model fleet.
中文 2026-11-18 Separate AI usage, task mix, self-reported productivity, reproducible artifacts, and scientific discovery before making a stronger claim.
中文 2026-11-18 Separate vendor claims, synthetic benchmarks, clinical safety reviews, and FDA CDS applicability before authorizing healthcare computer-use actions.
中文 2026-11-17 Design voice-agent audio decisions as typed, calibrated outputs with abstention and failure routing instead of treating every problem as transcription.
中文 2026-11-17 A practical validation ladder for LLM social simulations: replicate runs, match empirical windows, and separate activity, network, topic, toxicity, and style errors.
中文 2026-11-16 Trace MCP timeouts from discovery to tool execution. Identify deadline owners, test fixes safely, and avoid duplicate side effects after a timeout.
中文 2026-11-16 Validate LLM JSON beyond its schema with source checks, explicit failure routes and 12 runnable Python fixtures.
中文 2026-11-15 A practical action-risk matrix for legacy-portal agents, with executor-side approval, bounded autonomy, and verified business outcomes.
中文 2026-11-14 Agentic payments need intent, payment, and fulfillment receipts to keep approval, money movement, merchant acceptance, and delivery verifiable.
中文 2026-11-13 Treat stale TODOs as cached agent instructions. Join them to issue state, measure detours, and block unapproved scope expansion before commit.
中文 2026-11-12 A corrected Maya-site claim shows why research agents must separate candidate detection from coordinate-level novelty verification.
中文 2026-11-11 A practical gate for separating reproducible capability, grader loopholes, variant robustness, and unresolved contamination in AI leaderboards.
中文 2026-11-10 Audit state, event recovery, permissions, credentials, sandboxes, and data boundaries before moving a long-running agent between managed APIs.
中文 2026-11-09 Audit Claude Code permission files as sensitive data across command approval, local settings, Git history, build artifacts, and credential rotation.
中文 2026-11-07 Reliable AI agent memory needs write-time admission, source binding, correction propagation, and downstream evaluation, not recall alone.
中文 2026-11-06 Test whether safety framing changes agent performance with matched tasks, paired metrics, refusal transparency, and monitor-recall gates.
中文 2026-11-04 LifeVal tests personalized everyday tasks with artifacts and state diffs. Use a risk and verification ladder before an agent acts in the real world.
中文 2026-11-03 Pando keeps Git clients but narrows consistency to each ref. Use this architecture and migration checklist before scaling source control for agents.
中文 2026-10-20 The thin orchestrator pattern keeps AI agent routing deterministic while specialists reason. Use this contract for state, retries, and recovery.
中文 2026-10-19 HybridAgent reports higher quality and faster inference. A causal audit separates latent thinking from training, harness, and benchmark effects.
中文 2026-10-19 Harness² improves frozen models by editing the agent harness. Production use needs hidden evaluations, leakage controls, cost budgets, and rollback.
中文 2026-10-18 Use Claude Code review findings to build reproducible tests, lint rules, evals, and risk gates that prevent the same failures from returning.
中文 2026-10-17 Mercor estimates 90% of US work hours lack benchmark coverage. Separate task coverage, model capability, and job exposure before forecasting.
中文 2026-10-16 Meta Forum shows that AI-native engineering depends on specifications, observable context, verifiers, and human production gates.
中文 2026-10-14 Anthropic Scion redesigns source control for always-online agents. Learn its architecture, tradeoffs, and evidence gates before moving beyond Git.
中文 2026-10-13 AI self-replication needs a five-layer evidence contract covering capability, resources, persistence, propagation, and containment.
中文 2026-10-11 Evaluate AI agent context compression with task outcomes, decision equivalence, constraint retention, and cost per successful task.
中文 2026-10-10 ThinkingBox repeats 507 stateful agent tasks 20 times. This audit separates capability, consistency, observed 20/20, and production reliability.
中文 2026-10-08 GPT-6 Intelligent UI turns model output into a live interface. Production trust requires explicit state, capability, recovery, and outcome contracts.
中文 2026-10-07 Network Time Security authenticates NTP, but an AI agent audit trail also needs synchronization health, explicit time semantics, and evidence binding.
中文 2026-10-04 A workload-first playbook for Kimi K2.6 inference: measure request shape, find the current bottleneck, then tune decode, cache, and routing.
中文 2026-10-04 WhatWorkedBench shows why finding the best configuration cannot prove that a research agent understands the experimental system.
中文 2026-10-02 Failure Map shows how to turn boundary-case bugs into executable coding-agent tasks without confusing public checks with hidden evaluation.
中文 2026-10-02 Claude-shaped science works when research problems, executable checks, and expert judgment fit the model's real capabilities.
中文 2026-10-01 OpenAI's Australia incident shows how AI agent governance must link task authority, runtime evidence, disclosure, remediation, and closure.
中文 2026-09-29 Claude Code session retention is 30 days by default for CLI transcripts, with Desktop exceptions. Build a four-layer archive and retrieval contract.
中文 2026-09-28 OpenAI's DNS egress incident shows why AI agent sandbox containment must cover resolvers, transitive services, telemetry, and automatic shutdown.
中文 2026-09-24 Audit coding agent data flows before login with canary repositories, auth-state diffs, filesystem traces, egress capture, and deletion evidence.
中文 2026-09-22 AI agent memory becomes trustworthy when every actionable claim keeps its source, validity, dependencies, and invalidation conditions.
中文 2026-09-21 Patch half-life turns rewrite debates into a measurable decision: track recurring failures, repair durability, migration cost, and rollback readiness.
中文 2026-09-16 Benchmark Gemini 3.8 Live with two clocks: response latency and verified task completion across turns, tools, and interruptions.
中文 2026-09-15 LLM relays can read prompts and rewrite tool calls. Use this evidence-based audit contract before production data or agents cross a proxy.
中文 2026-09-11 A practical contract for tracing every AI analytics claim from business question to source, query, result, chart, refresh, and approved action.
中文 2026-09-11 Audit state, authority, recovery, observability, and cost before migrating production agents to OpenAI's managed Codex harness.
中文 2026-09-10 Web Bot Auth verifies a signing key. Robot access needs declared purpose, directory classification, local policy, and auditable edge enforcement.
中文 2026-09-09 KV-cache manipulation lets pretrained LLMs observe, reason, and respond concurrently. The mechanism works; the production runtime is still emerging.
中文 2026-09-09 Quantum calibration AI agents need every action tied to a measurable signal, a pass condition, and an escalation path. MIT's case shows the contract.
中文 2026-09-09 The DseWiki incident shows why read-only web access must be defined by side effects, not HTTP method names. Here is what the public evidence proves.
中文 2026-09-05 Astra scored 62.7% or 99.9% on ARC-AGI-3 under different harnesses. Here is what changed, what the numbers mean, and what remains unproven.
中文 2026-09-05 Claude formalized Fermat's Last Theorem in Lean in 11 days. Here is what the 13-million-line proof and its verification stack establish.
中文 2026-09-05 Persistent AI agents need separate contracts for wake timing and action authority. This checklist makes long-running agents safer.
中文 2026-09-03 Build an organizational second brain with provenance, versioned knowledge, expert review, replay tests, and safe retirement instead of static RAG.
中文 2026-09-02 EvoUndo shows why capability gains are insufficient for self-modifying Agent Harnesses. Require witnessed, counterfactual recovery before merge.
中文 2026-08-30 MCP agent identity is becoming a protocol priority. This guide separates stable authorization from draft workload identity, delegation, and DPoP work.
中文 2026-08-26 OpenAI's Admin Plugin turns workspace administration into a permission-aware agent loop.
中文 2026-08-26 OpenAI Jalapeño Benchmark Audit: Full-Stack Inference
中文 2026-08-25 A study of 8,135 agent trials finds skills work mainly as procedural anchors. Use this evidence-based checklist to design better SKILL.md files.
中文 2026-08-24 UiPath Maestro Flow lets coding agents create workflows, then captures value in durable execution, observability, governance, and metered runtime use.
中文 2026-08-23 A code and policy audit of PhoneBuddySDK shows why mobile AI agents need an in-process runtime, plus the evidence still required to ship.
中文 2026-08-19 Route LLM tasks by output contract, failure cost, deadline, and trajectory signals, then optimize cost per completed task.
中文 2026-08-19 Measure Claude Code routines from trigger to unreverted merge, separating run health, task success, CI, review, and workflow losses.
中文 2026-08-18 Anthropic's conflict tests show why multi-agent coordination needs mandate invariants, escalation rules, and independent acceptance checks.
中文 2026-08-12 Go helps AI coding agents by turning compiler, formatting, test, module, and security checks into a consistent machine-readable verification surface.
中文 2026-08-12 A 2026 study decoded 315,320 client-side reasoning blocks. The durable lesson is that encryption needs identity, context, expiry, and replay controls.
中文 2026-08-11 Muse Glimmer fits a 30B agent model into 24 GB. The real deployment contract spans memory, latency, tools, state, permissions, and recovery.
中文 2026-08-10 Muse Code has moved beyond beta. Its event-log claim still needs failure tests for workspace recovery, upgrades, and remote side effects.
中文 2026-08-09 Cloudflare Kitesurf is an agent-first browser on V8 isolates. Learn its architecture, limits, benchmarks, and when Chromium remains the better choice.
中文 2026-08-08 OpenAI has formally designated GPT-6 Astra as Critical for cybersecurity. Here is the evidence, safeguard case, and residual risk.
中文 2026-08-06 A capability matrix for agent security hooks that separates visibility, evidence quality, activation, failure behavior, and enforcement.
中文 2026-08-06 A five-layer verification stack for scientific AI agents, from reference parity and parameter tolerances to simulation truth and stewardship.
中文 2026-08-04 Test whether AI agents can use your product with dynamic task evals that measure correct state changes, recovery, cost, and regression stability.
中文 2026-08-04 AI slop CVEs can trigger harmful patches. Require provenance, source, reachability, reproduction, and fix evidence before auto-remediation.
中文 2026-08-02 Flint adds a semantic intermediate representation between AI intent and chart code, making visualization generation editable, portable, and testable.
中文 2026-08-02 Design realtime transcription as an independent protocol contract with explicit turn identity, deltas, completion, latency, revisions, and fallbacks.
中文 2026-08-02 HANDBOOK.md shows why policy in context is not control. Build AI agent compliance with retrieval, preflight decisions, commit gates, and verification.
中文 2026-08-02 Closed-loop voice agent analytics connects every conversation to recommendations, purchases, refunds, handoffs, and incremental contribution profit.
中文 2026-07-30 The Hugging Face incident shows how persistence compounds small weaknesses. Long-horizon AI safety needs trajectory-level control.
中文 2026-07-30 Agent observability needs evidence about outcomes, actions, cost, identity, and approvals. Chain-of-thought is a signal, not the production record.
中文 2026-07-29 AI agent benchmark scores mix models, harnesses, tasks, tests, and environments. Use this framework to build a reproducible coding-agent eval.
中文 2026-07-28 AI coding agent backups need versioned, isolated copies and tested restores. Build a recovery contract that proves the workspace can work again.
中文 2026-07-26 A controlled method for making GPT-5.6 prompts leaner without deleting the constraints that keep production agents reliable.
中文 2026-07-25 Turn the Claude Cookbook into an Agent Regression Suite
中文 2026-07-25 Cursor's agent swarm reached 100% on held-out SQL logic tests.
中文 2026-07-25 Drone-Bench and AI-controlled F-16 tests show rapid capability progress.
中文 2026-07-25 GEPA optimize_anything makes the evaluator the stable interface for prompts, code, agents, and configs. Here is how to design one safely.
中文 2026-07-23 OpenAI Presence turns policies, approved actions, simulations, evaluations, escalation, and controlled updates into one operating loop.
中文 2026-07-23 A practical workflow for AI agents in mathematical proofs: durable state, hostile audits, blind reconstruction, and evidence-gated knowledge.
中文 2026-07-23 AI agent sandbox persistence varies by lifecycle action. Use this five-layer model to verify files, memory, volumes, connections, and side effects.
中文 2026-07-22 A four-layer compatibility model for moving SKILL.md workflows across Codex, Claude Code, Gemini CLI, and GitHub Copilot without confusing shared syntax with equivalent behavior.
中文 2026-07-22 AI coding agents may execute dependencies before review.
中文 2026-07-20 "When Should an AI Agent Get Its Own DSL? A Minimal Decision Framework"
中文 2026-07-19 MCP Elicitation turns missing input into recoverable control flow. Learn form vs URL mode, state handling, security boundaries, and retries.
中文 2026-07-19 J-space handles concepts inside one inference. Reliable agents still need context, checkpoints, long-term memory, and external traces.
中文 2026-07-19 Browser agent detection now needs three layers: behavioral risk signals, signed identity, and explicit authorization for high-risk actions.
中文 2026-06-06 "OpenAI's B2B Signals report reveals frontier firms use 3.5x more AI per worker than typical firms.
中文 2026-05-19 Enterprise AI adoption has a pattern that gets repeated across industries.
中文 2026-05-19 When Dennis Hannusch needed an internal podcast recording app at NVIDIA, his first thought was not to build one.
中文 2026-05-17 "Codex Windows Sandbox: Inside OpenAI's Secure Coding Environment Architecture"
中文 2026-05-14 "OpenAI and PwC CFO Alliance: How AI Agents Are Rewiring Enterprise Finance from Budgeting to Forecasting"
中文 2026-05-03 "Anthropic ran a week-long experiment where Claude autonomously traded items across 4 parallel markets.
中文 2026-04-26 "GPT-5.5 is OpenAI's first fully retrained foundation model since GPT-4.5.
中文 2026-04-21 "How DeepMind's Lyria 3 breaks the AI music quality ceiling through latent diffusion architecture, SynthID watermarking, and licensed training data.
中文 2026-04-20 Discover why prompt engineering is the wrong abstraction level.
中文 2026-04-20 Claude Opus 4.7 migration checks for adaptive thinking, visible summaries, output limits, API versions and Claude Code permissions.
中文 2026-04-20 Discover why 78% of enterprises use AI but only 21% reach production scale.
中文 2026-04-16 On April 13, 2026, Cloudflare and OpenAI launched Agent Cloud with GPT-5.4 at the edge.
中文 2026-04-13 "Claude Sonnet 4.6 delivers 79.6% on SWE-bench Verified at $3/$15 per million tokens — within 1.2 points of Opus 4.6 at 60% of the cost.
中文 2026-04-06 Understand Anthropic emotion vectors, the limits of steering experiments, and practical verification checks for coding agents.
中文 2026-04-04 "Harness engineering is the defining discipline of AI coding in 2026.
2026-04-02 Klarna replaced 1200 SaaS tools with AI, saved $60 million, then quietly reversed course.
中文 2026-04-02 Compare MCP and CLI by tool discovery, execution, permissions and recovery. Choose the interface that fits your agent rather than a market winner.
中文 2026-04-01 Harness Engineering is the core AI engineering paradigm of 2026.
2026-04-01 On March 27, 2026, Anthropic leaked approximately 3,000 unpublished assets due to a CMS configuration error.
2026-03-28
2026-03-28