ai-agents
267 posts tagged ai-agents.
- Amplitude's Wave and the agent loop that starts before anyone asks for a feature
- Contrastive Discovery: Why Science Agents Should Weigh Rival Explanations, Not Just Reject One
- ReTrace turns repeatable agents into code paths
- Tether’s BotPsy points to smaller phone agents, but the proof is still thin
- What Your Agent Evaluation Harness Can Actually See
- Agentic-OPD says your best RL checkpoint is not the whole model
- Dense rewards made clinical agents ask better questions
- HiVe-Mem and the Case for Two-Dimensional Agent Memory
- Near AI’s breach shows where agent privacy really breaks
- A Generalist Model Beat Specialist Robot Policies Without a Single Demo
- Radiology agents are only as good as their tool catalog
- Agent memory is architecture, not a chat history
- Cloudflare’s agent hints look like SEO’s next control layer
- Local AI on 16GB VRAM is a scope problem, not just a model problem
- Local GUI agents need disposable desktops, not prettier demos
- OpenAI's LifeVal and the gap between knowing and doing
- The Label Wording Problem Nobody Tests in Agent Routers
- AI exploit generation is getting cheap enough to change patch work
- Richer Context Does Not Automatically Fix LLM Social Agents
- What Anthropic's SDK Error Codes Tell You About Running Claude in Production
- Cognitive symbionts shift the agent question back to the human
- The OCC puts AI agent autonomy on the bank examiner checklist
- Websites need an agent policy before agents become customers
- DiabAgent tests medical agents where mistakes compound
- Google Search Is Becoming an AI Agent Manager
- Strategist treats agent reasoning as code that can evolve
- WebHarbor Puts 153 Fake Real Websites in a Box for Agents to Train On
- Claude-written Linux exploit puts patch lag in the spotlight
- LangChain’s Python 3.10 cutoff is a maintenance warning for agent stacks
- Claude, game reverse engineering, and the product boundary problem
- Factorio Peaceful Mode Is a Useful Agent Test, Not a Finish Line
- Malware using LLMs for victim triage is the practical AI security problem
- Memdoor's Claim That Five Small Models Beat One Big One on Hard Coding
- LangChain’s tiny core releases are really about state discipline
- The Economics of Proactive AI: When a Single 1B Decoder Beats a Two-Model Pipeline
- Verifying Computer-Use Agents by Letting the Verifier Go Look
- ChatGPT browsing access is now part of site visibility
- OpenAI and ServiceNow move agents closer to enterprise work queues
- What OpenAI's Misalignment Detection Post Actually Describes
- Hidden AI Opinions Can Steer Expert Analysts
- What LLM Group Memory Keeps, and Why It Still Gets the Answer Wrong
- The .agent domain fight is about AI distribution, not vanity
- Inbox agents need labels before models
- When an LLM's 'True' and 'False' Don't Sum to One
- Claude Duet tests the proactive assistant, not a new chatbot
- OpenAI’s Dots demo turns bug triage into an agent workflow
- Agents fail when true tool outputs omit the one fact that matters
- METTA splits agent memory into moments and episodes
- ASTEROID teaches agents to explore before they act
- Quant-Arena Tests Whether LLMs Can Actually Build Trading Strategies
- The Teacher That Scores Worse Than It Plays: Fixing Agentic Distillation
- When AI Agents Start Finding Bugs Faster Than Maintainers Can Patch Them
- LatComm makes long context reusable across models
- WeAct Turns LoRA Adapters Into Something an Agent Can Pick Mid-Task
- WildAWM trains computer-use agents in synthetic worlds
- An AI Agent That Reproduces Cancer Studies and Checks Where They Wobble
- What OpenAI's 'ultrafast' service tier for Agents actually signals
- StepFun’s agent phone is a bet against the app grid
- FDE-Gym Puts Agents in the Messy Last Mile of Customer Delivery
- MLExpert and the Case for Training Agents on Whole Experiments, Not Single Turns
- CoPE trains agents to forecast their own future actions
- The Case for Making Research Agents Forget Their History
- When Renaming a Tool Breaks Your Agent: What CIPR Fixes
- ChunkAct moves robot agents from constant planning to reusable routines
- Latent Thinking, Text Acting: What HybridAgent Changes About Agent Cost
- Self-Evolving Agent Harnesses Need Change Control More Than Hype
- Tev's Dense Progress Rewards and the Sparse-Reward Problem in Long-Horizon Agents
- On-device agents are a systems bet, not just a model release
- When the Vulnerability Scanner Becomes an Attack Planner
- Agent red teams need maps, not just scary prompts
- Tool retrieval should know when the tool is broken
- Marketing AI has budget before it has proof
- Oracle’s Codex story is about killing the report backlog
- Accuracy is not enough when agents meet conflicting evidence
- OnTrack watches agent trajectories before the bill is paid
- ViSkill and the case for keeping agent memory visual
- Asana’s browser-agent test points to the real AI cost story
- Manus, Meta, and the new deal risk around AI agents
- Probes That Catch an Agent Lying Before It Says a Word
- What Sophos actually automated to cut threat response from 38 minutes to 89 seconds
- SciExam Tests AI Agents on Climate Modeling Without Giving Them the Answer
- When a hallucination enters the context, can the model talk itself out of it?
- AI-assisted breaches are becoming an operations problem
- Co-Evolving the Attacker: How AdvSim2Real Hardens Web Agents Against Prompt Injection
- CLIFT turns web-agent judging into reusable verification
- MemPilot and the Case for Deciding Memory at Query Time, Not Ingest Time
- Robot agents need indexed video memory, not bigger prompts
- RL Environments on Hugging Face Make Agent Work Less Hand-Wavy
- Agent memory is not a substitute for readable docs
- LeCun’s “zero concerns” is the wrong bar for AI safety
- The risky lesson in Meta Muse’s reported household-authority prompt
- When an Agent Reports Success and the Database Says Otherwise
- A simulated canvas turns vision models into operators
- Stratego Is a Better Agent Test Than Another Chess Win
- OpenAI’s Python SDK is getting more agent-shaped
- ScholarCatalyst: When Agentic Search Can't Beat a Plain Embedding
- Source learning makes RAG less forgetful
- DynaHarness puts a governor between robot planning and robot motion
- AdviSD trains a small model to steer bigger frozen models
- Robinhood’s trading agents make permissioning the product
- Robots Need Correction Loops, Not Just Better Planners
- The Harness Is the Model: What 'Meta-Skills for Agent Harness Design' Actually Shows
- KV-streams make long-agent training cheaper by not throwing memory away
- OpenAI's DevDay agents move the product boundary from chat to operating system
- Self-Improving Agents Without RL: What Retrospection, Creativity, and Token Forecasting Actually Show
- Australia’s AI agent hearing is really about incident disclosure
- The useful part of Roetzer’s architect-orchestrator-apprentice AI work model
- When Documentation Doesn't Help Coding Agents: A Negative Result Worth Reading
- Agent control failures are moving from demo risk to operating risk
- Drawgent puts a coding agent on an Excalidraw canvas: what a visual work surface actually changes
- AI agents cut the estimated cost of quantum-safe Bitcoin transactions
- Ollaya points at a missing layer for decision models
- When Agents Hack: What the OpenAI-Hugging Face Story Actually Tells Builders
- AI Agents Make Audience Data Debt More Expensive
- Bitcoin Lightning joins x402, and agent payments get a little more real
- Conspiracy Detection Needs Context, Not Just Better Keywords
- ExplorationBench and the Gap Between Recall and Real Discovery
- Australia’s OpenAI agent breach is a disclosure problem, not a sci-fi one
- Multi-agent shutdown sabotage is a real eval target now
- The World Model That Edits the Agent's Mind, Not the Environment
- AI agents need payment rails before they need crypto hype
- Fact-checking agents need receipts, not vibes
- Known puts domain identity into the agent stack
- Local tool-use evals are measuring your server too
- Speaker-Centered Memory: Why Group Chats Break Your AI Agent
- Critical-State RL trains the one agent call that actually matters
- DolphinBench Tests Agent Memory by Task, Not Trivia
- Gemini’s reported company breaches show the AI disclosure gap
- Long-horizon agents can learn to cheat the checker
- Where Harness Self-Improvement Actually Stands in Late 2026
- Full-Duplex Voice With Tool Calls: What NemotronLabs VoiceChat Actually Ships
- Hex, GPT-6 Astra, and the shift from answers to visual reports
- LangChain 1.4.2 fixes a small but real agent reliability problem
- What Two Boring openai-python Patches Reveal About Agent Reliability
- LangChain’s typesafe alpha points at safer model routing
- The Harness Matters as Much as the Model in Coding Agents
- When a Coding Agent Drives a Robot, Task Success Isn't Safety
- When the Teacher Knows to Quit: RetireOPD and Distillation for Agents
- Agent memory helps after the action loop stops breaking
- ENCP gives navigation agents a usable uncertainty check
- ScienceBuddy and the case for training the harness before the model
- The Skill Router You Already Have: Gavel Reads Routing From a Frozen LLM
- A vulnerability is not fixed because an AI bot saw it
- Siri as a model router, not a single assistant
- What Fyxer's AI inbox assistant gets right about trust
- When Agents Lie to Pass the Eval
- Devin testing its own work with GPT-6 Astra: what's real, what's reported
- Perplexity Handing GPT-6 Astra Production Access: What OpenAI's Claim Actually Means
- The RubyGems agent report is a supply-chain warning
- An 86% Bitcoin quantum benchmark cut is not an 86% attack
- Recursive Self-Improvement Has a Roadmap Now, and Most of It Isn't Built
- CEO clone chatbots expose the limits of personality wrappers
- Browser agents are becoming practical comment analysts
- Can AI Agents Do Their Own Interpretability Research? A New Benchmark Says Not Yet
- OpenAI’s reported Millennium math claim needs proof, not applause
- Procedural Graphs give agents a memory of what to do, not just what happened
- OpenAI's Python SDK Gets More Honest About Agent Failures
- How a 9B model learned to chain Korean government APIs by actually calling them
- When You Upgrade the Model, Does the Agent's Memory Come With It?
- Agent Harnesses Are Becoming a Token Efficiency Fight
- AI can help with PCB design, but it cannot own signoff
- GPT-6 Astra's Big Claim: Agentic Skill Meets Alignment
- What LangChain Core 1.6.2 Fixes for Agent Builders
- Compiling a Prompt Into a Small Neural Function You Can Version
- ESPO's fix for prompt bloat: diagnose errors, then stop appending rules
- The OpenAI agent hack report is a disclosure problem first
- Repo-distilled skills are the missing middle layer for research agents
- Web agents need useful predictions, not prettier page snapshots
- AI agent interviews are a workflow fix, not a content strategy
- Proactive writing agents need timing more than autocomplete
- SAGE Uses a Big Model as a Coach, Not a Crutch
- Verbal reinforcement learning is a feedback routing problem
- Gemini’s agentic video framing shifts the hard part from seeing to checking
- The Evaluate-First Turn in AI Research Agents
- Small dialogue agents need repair loops, not just bigger training
- What Happens When You Let AI Agents Build Their Own Society
- Agent payments are a product problem before they are a currency problem
- Treat the smart TV like an untrusted computer
- WikiSkill and the Case for Giving Agents a Memory That Compounds
- Anthropic’s Python SDK is tightening the agent plumbing
- Correct SQL answers can still hide broken agent work
- SwarmWorld makes a case for agents that coordinate through artifacts
- Claude Code as a domain renewal research assistant
- Recuris and the Case for Memory That Rewrites Itself
- The Retrieval-Integration Gap: When Your AI Analyst Reads the Risk and Ignores It Anyway
- Prime Agent treats the harness as part of the model
- Gemini and Apex point prediction markets toward brokerage plumbing
- Verification becomes the operator loop for AI-built silicon
- A 36-node DGX Spark homelab points at agent infrastructure, not just bigger inference
- Munder Difflin and the real work behind AI clone offices
- Treat the Texas AI hacking story as a disclosure systems test
- What LangChain's perplexity 1.4.1 patch says about agent plumbing
- Agent memory works better when skills are small and written in text
- Binance Agent OS puts trading agents behind user-controlled gates
- Task Model Induction turns messy screen traces into reusable agent skills
- What AI4AI-Bench Says About Recursive Self-Improvement Right Now
- When a Zero-Shot LLM Ties a Random Forest on Travel Behavior
- SPADE Makes the Training Environment a Thing the Model Learns to Build
- The Self-Improving Agent Demo That Falls Apart When You Shuffle the Tasks
- AutoSR turns symbolic regression into a research-state search
- PIHF turns prompts into a versioned policy layer
- AI coding feels like managing a very literal junior engineer
- What Actually Makes a Claude Code Session Productive
- AutoDesign turns paper-to-poster into a harness the agent rewrites itself
- OmniScientist argues that AI scientists need eyes, not just workflows
- A Strong Model Can Scaffold a Weak One Without Any Retraining
- When Your User Simulator Is Secretly One Person: Simulator Collapse in Multi-Agent RL
- AMIE’s video consult result is about perception, not replacement
- Auto-research agents need fuzzer-style feedback
- PsychoAgent makes memory retrieval less purely semantic
- SkillProx and the Case for Agents That Prune Their Own Playbooks
- The 2011 Link That Died on Schedule, and What It Says About Agent Memory
- Kitesurf and the browser built for agents, not humans
- Benchmarks Need QA Before They Judge Agents
- Heart-failure feature engineering gets an agent pipeline, not a chatbot
- RAG For Table-Heavy Reports Needs Search You Can Audit
- The Harness Is the Product: What HarnessOpt-Bench Actually Measures
- Video deep research agents need to look before they search
- DNS identity for AI agents needs more than a name
- LiveMem reframes long-context memory as state continuity
- Where multimodal embeddings and collaborative coding agents actually stand
- DungeonBench puts tactical reasoning where agents usually break
- ExtractBench tests document extraction where demos usually fail
- On-policy imitation helps when the student is smaller than the expert
- Post-training is now the behavior layer
- The useful question behind a multiplayer agent harness
- Claude’s test escape is a security design problem, not a sci-fi story
- Computer-use agents need stricter judges, not prettier demos
- ORCA-bench shows oncall agents are not ready for pager duty
- When You Count the Tokens, Self-Reflection Loses to Just Sampling More
- AI research agents can code, but they still can’t judge the work
- APEX-Accounting: The 2.6% Number That Should Scare AI Bookkeeping Startups
- The Office-Task Benchmark That Prices Agents Against Human Labor
- πR² makes robot policies react inside the action chunk
- Autonomous research needs a budget scoreboard
- ChatGPT Work turns sales AI into a revenue feedback loop
- On-Policy Distillation Is Becoming the Default Move for Agent Training
- CausalForge makes AI research agents prove their work
- Coinbase’s AI agent bet is payments plumbing, not an AI pivot
- Anthropic's SDK Just Named Claude Opus 5 and Added Mid-Stream Tool Swaps
- Treat rogue AI hacker stories as incident reports, not movie trailers
- GS-Agent Builds 4D Worlds by Driving a Physics Engine, Not Replacing It
- LangChain’s gateway env var is production plumbing
- Training Agents Inside the Harness They Actually Ship With
- World’s $52.5M bet on proof of human for agents
- PoTRE makes the case for heterogeneous test-time reasoning
- Kimi K3’s SpreadsheetBench win is a signal, not a coronation
- Shopify’s agent lesson is delegation, not autonomy
- Useful work per dollar is the agent metric that matters
- IoT exploit agents are getting useful in controlled labs
- The AI job diamond still needs a bottom rung
- PolicyGuard makes the compliance bot show its work
- HORIZON treats chip design like a repo, not a chat prompt
- The 4-second budget that decides if your AI agent ships
- Agent Success Rate is the only number that matters when a new model drops
- Splitting the agent loop from tool execution cut TTFT by 90%
- Agent Amnesia Has a Fix, and It Looks Like Sleep
- Dreaming Agents Could Finally End the Brand Voice Correction Loop
- How a 400-line system prompt becomes 15 lines with Skills
- The 200K Token CSV Problem Has a One-Line Fix
- The 400-line system prompt is the new technical debt
- The mechanism matters: making AI competitive analysis auditable
- Last Quarter Means Two Different Things Inside Your Own Company
- Split-brain agents: when planner and executor stop talking, quality collapses
- The Frustration Index: A Cheap Eval Most Teams Skip
- Why I Stopped Trusting Demo Videos for Agent Tools
- When Your AI Agent Needs a Browser, Not an API