#Agent
137 related articles
From Vibe Coding to AI-Automated Research: How Agents Are Reshaping the Research Workflow
From Vibe Coding to Coding Agent to AI-automated research, we analyze how Agents intervene in literature review, data analysis, and experiment execution, and which judgment tasks remain irreplaceable for researchers.
Needle: An Agent-Native Outbound Engine Powered by Claude/ChatGPT/Grok Prompts
Needle is an agent-native outbound engine that lets you find and reach ideal customers on LinkedIn, Instagram, and email via Claude, ChatGPT, or Grok prompts. This article analyzes its positioning, agent-native design logic, and key details missing from the launch.
Deconstructing Coze 3.0 Agent Core Capabilities: From Beginner to AI Talent in Three Months
Deconstructing Coze 3.0 agent core capabilities: a three-step learning path from prompt engineering and RAG knowledge bases to multi-agent practice, plus a rational look at three-month employment promises.
Running More Agents Doesn't Mean More Output: Verification Is What Really Determines Efficiency
Opening more Agent windows doesn't mean more output? A Bilibili video points out that Agents only amplify the volume of changes—what really determines output quality is verification. Drawing on the AlphaCode case, this article analyzes why no matter how many candidates there are, you still need a sieve, and how to make verification small and accurate when the tests themselves leak.
62 Lines of Bash Builds an AI Agent: How the Minimal Harness Works
Build an AI Agent from scratch with a 62-line Bash script: from a single LLM call to a chat loop to a coding Agent with bash tool calling. Plus a live Tetris test.
Optimist Deep Research: A State-Centric Iterative Literature Research Agent
Optimist Deep Research is an open-source literature research agent that reframes literature search as iterative evidence acquisition centered on a mutable research state, with sub-question trees, four-layer deduplication, citation snowballing, and an eight-dimensional critic, producing traceable citations in the final report.
OpenFlow: A Unified Data Plane for Agent Orchestration?
OpenFlow proposes "ultimate agent orchestration" on Hacker News, seeking to make multi-agent scheduling an independent data plane. This article analyzes its decoupling approach, observability and protocol compatibility value, and practical concerns over ecosystem position and latency.
Stop Auditing the Agent, Audit the Counters: An AI Coding Workflow Retrospective
AI coding Agent workflow retrospective: three layers of retry sharing one log makes diffs untraceable, reviewer opinions can't replace test exit codes, and writer/reviewer need directory isolation. Includes a six-step replayable loop and bash gate scripts to avoid five wasted hours.
Does AI Get Dumber the Longer You Chat? How Claude Code and Cursor Manage Context
Why does AI get dumber in long conversations? This article analyzes why models misuse outdated information and introduces Claude Code and Cursor's approaches: Agent status bars, search verification, Sub-Agent isolation, and task-based compaction.
After Removing the Install Step, We Realized It Held Up the Entire Design
CloudPal removed the install step and discovered this small packaging decision underpinned the entire architecture. A deep dive into the constraints it exposed.
Paid Users Exceed 20%: Why E-commerce Merchants Are Flocking to Qwen Office
Two months after launch, Qwen Office E-commerce Edition sees paid merchant users exceed 20%. AI Agents are taking over data aggregation, price monitoring, and material generation.
Gemini 4 Argon Released: One-Million Output Tokens and Self-Evolving Agents
Gemini 4 Argon launches with industry-leading 1M output tokens, agent-assisted training that spins up an RSI flywheel, 100K-engineer internal testing, and a full writing-style overhaul.
Agent Memory Without Contamination: A Guide to Multi-Tier Memory Architecture Design
A deep dive into Agent memory contamination and multi-tier memory architecture: three-tier storage, write gates, read ranking strategies, poisoning test suites, GDPR-compliant deletion, and compression drift prevention—helping developers build trustworthy AI Agent memory systems.
MCP Tool Call Security: The Agent Tool Surface Is the Attack Surface
MCP lets AI Agents call tools, but most teams overlook who decides the call. This article analyzes injection risks, trust boundary determination, and three missing defense lines.
Understand Machine Learning and GPT Principles to Write Better Prompts
From Transformer to GPT to Agents: understand the statistical nature of LLMs and the causes of hallucination to write sharper prompts, cut token costs, and prevent Agent overreach and prompt injection.
Definition Is the Core: When Both People and Models Change, What Assets Still Run
Starting from Tencent's OK platform's 68% automation rate, this asks what counts as an organizational asset in AI development: definition is the core. Only rules that still constrain the next run after swapping people, models, and vendors are worth keeping—everything else is a carrier.
Stripe Projects Lets Agents Provision WorkOS Accounts Autonomously
At WorkOS init(), Stripe's Rami Banna demoed a coding agent autonomously provisioning a WorkOS account and credentials via Stripe Projects and the Provisioning API.
Drex v2.0 Released: A Decision Model That Makes Judgments Instead of Writing Prose
Nace.AI releases Drex v2.0, a decision model that outputs probabilistic judgments instead of prose text, supporting yes/no, multiple-choice, and scoring questions. Output is free, and it shares an API contract with TypeSafe Jev — decision models are becoming a distinct category.
Claude Architect Certification in Practice: A Complete Guide to the CCAR-F Exam
A developer's full account of Anthropic's CCAR-F exam: eligibility gates, two-hour proctoring, question style, and study strategies covering Claude Code, Agent SDK, and MCP.
Routing LLMs by Prompt Length Backfired: A 12-Word Request That Broke a Production Agent
A router that split LLM traffic by prompt length sent a 12-word request with 6 tool definitions to the cheapest model, causing an infinite agent loop. Here's why length measures cost, not capability.