27 results for “ai agent”
Related topicsagentic aiai agentsai agent
Apr 16
ReAct's Loop: The Simple Pattern Behind Every Modern AI Agent
The ReAct framework, introduced in a 2022 paper, defined the thought-action-observation loop that underpins nearly all modern AI agents. This article breaks down the pattern, explains why it works, and covers the failure modes that developers still encounter today.
May 22
Your LLM Is a Reward Model: How DPO Cut Alignment Costs
Direct Preference Optimization (DPO) simplifies the complex RLHF pipeline by eliminating the need for a separate reward model. This article explains how DPO works, why it made alignment more accessible, and the trade-offs involved in this more direct approach.
Jan 13
Codex-Max and the $100+ Mistake 24 Hours In
GPT-5.1-Codex-Max can code autonomously for over 24 hours, but this power introduces new risks. This article explores the economics of long-horizon tasks and how to architect systems that prevent costly, deep-rooted errors.
Feb 24
Claude's 1M Tokens & 90% Recall: What It Solves, What It Doesn't
Anthropic's Claude Opus 4.6 offers a 1M token context window, but its real value lies in its claimed 90% recall. This article explores when this massive context replaces RAG and when it's an expensive distraction.
Nov 26
Claude Opus 4.5 at $5/Mtok: When to Upgrade Your Agent's Brain
Anthropic's Claude Opus 4.5, released November 24, 2025, changes the economics of using frontier models. This architectural review explores how its $5/$25 per million token price point forces a redesign of model routing logic, especially for complex agentic workflows.
Dec 09
Opus 4.5 vs Gemini 3 Pro: Which Model for Which Workload
Released six days apart in November 2025, Google's Gemini 3 Pro and Anthropic's Claude Opus 4.5 are good at different things. A comparison on the axes that actually decide a stack: coding and agentic work, context handling, cost per unit of work, and provider concentration risk.
Jun 04
Qualifying Claude Opus 4.8: From Shadowing to Go/No-Go Decision
Anthropic's Claude Opus 4.8 is out, but with no official benchmarks or release notes, upgrading is a gamble. This article provides a complete framework for safely qualifying the new model using shadow traffic, custom evaluation metrics, and a data-driven go/no-go decision.
Jul 29
2.8 Trillion Parameters, Free to Download: Inside Moonshot AI's Kimi K3
Kimi K3 packs 2.8 trillion parameters but activates only 104 billion per token — and Moonshot AI put the full weights on Hugging Face eleven days after announcing it.
Aug 07
GitLost: How a GitHub AI Agent Leaked Private Repos from One Public Issue
Security researchers discovered "GitLost," a vulnerability where a GitHub AI agent was tricked by a malicious prompt in a public issue into leaking data from private repositories, highlighting a fundamental security flaw in agentic AI systems.
Jan 06
The $12 Agent Task: Why Production AI Needs a Harness, Not a Loop
Most AI agent demos are built on a simple loop that is too expensive and unreliable for production. Real-world systems use a "harness" architecture with structured workflows, checkpoints, and budgets to create agents that can survive contact with reality.
Jun 30
Your AI Is Slow: A Latency Budgeting Guide for Sub-Second Agents
Stop thinking about AI latency as a single number. Treat it as a budget to be spent across retrieval, generation, and tool use, and learn how in-process controls can shave 50ms or more from every user interaction.
Apr 02
The 76% Mistake: Why AI Build vs. Buy Is a Layer-by-Layer Choice
The "build vs. buy" decision for LLM applications is no longer a single choice. This article presents a framework for analyzing the AI stack layer by layer—from the model to evaluation—to avoid common pitfalls and focus engineering effort where it creates the most value.
Dec 04
Your Retry Loop Is a Time Bomb: From a $50 API Call to a $5,000 Bill
A simple retry-on-failure wrapper seems harmless, but in AI agent systems, it can lead to silent cost overruns and cascading failures. This article explains why, and how to build resilient, cost-aware error handling using techniques like total budget caps and structured escalation ladders.
Jun 11
Stop Fine-Tuning Your LLM: Why Retrieval Is Cheaper, Faster & More Reliable
Developers often try to add knowledge to a large language model by fine-tuning it, a costly and often ineffective approach. A better strategy follows a clear hierarchy: start with the prompt, then use Retrieval-Augmented Generation (RAG) for facts, and only use fine-tuning to change the model's core behavior.
Apr 30
The Trillion-Parameter Tipping Point: Open-Weight AI Is Now a C-Suite Decision
As of April 2026, open-weight models with over a trillion parameters rival proprietary APIs in capability, forcing a new strategic decision for enterprises. This article breaks down the choice across four critical axes: data control, capability, cost, and the upgrade cycle.
Aug 10
LLM System Design Architecture: The End-to-End Guide to Building Production AI
A comprehensive guide to end-to-end LLM system design architecture, covering the core patterns like RAG, vector databases, and agentic workflows required to ship production-grade AI applications.
Mar 26
Role Confusion: The 61% Attack Success Rate Exposing LLM's Core Flaw
Prompt injection isn't just "tricking an AI"; it's a classic authorization vulnerability. New research shows a 61% attack success rate by exploiting 'role confusion,' where models grant authority based on writing style, not source, demanding a shift to architectural security patterns.
Dec 23
The GenAI Divide: Why 95% of Enterprise AI Projects Delivered No Value
In 2025, enterprises spent $37 billion on generative AI, yet a stark "GenAI Divide" emerged. A landmark MIT study found 95% of projects failed to deliver any business value, revealing that success depends not on model quality, but on organizational strategy.
May 14
Your LLM Is a Bug: A 5-Point Test for When Not to Use AI
Large Language Models are powerful but often the wrong tool for the job. This article provides a five-point decision test to identify when a simpler, more reliable solution like a database query, a regular expression, or a classical classifier is the better choice.
Aug 25
The Model Upgrade Is Not a Drop-In: A Production Migration Guide
A major new foundation model release is not a simple drop-in upgrade. This guide provides a framework for migrating your production pipeline safely, covering re-qualification, prompt auditing, and sequenced rollouts to avoid breaking changes.
Jul 06
HITL That Works: Make Review Faster Than the Original Task
Most human-in-the-loop (HITL) review queues fail because they are slower and more tedious than the manual work they replace. This article details four principles for designing HITL systems that humans will actually use, focusing on making review faster than creation, surfacing model uncertainty, and turning the review process into a continuous evaluation set.
Nov 12
Prompts Are Production Code: Version Them Like It
A 2025 industry survey found that 30-40% of AI development time is spent on prompt engineering. This article explains why treating prompts as versioned, deployable artifacts is the key to reclaiming that time, improving reliability, and shipping faster.
Oct 17
Chunking Is a Retrieval Decision: Up to 9% Recall Is on the Line
Most RAG systems fail due to a single, overlooked setting: chunk size. This article explains how rethinking chunking as a dynamic retrieval parameter, not a static preprocessing step, can boost recall by up to 9% and why the optimal size depends entirely on your users' questions.
Aug 25
Debate and Decompose: When a Second Agent Is Worth the Cost
Multi-agent systems add cost and complexity, but are justified for specific use cases like independent verification (debate) or breaking down tasks that exceed a single model's context window (decomposition). This article provides a framework for deciding when to add a second agent.
Feb 03
How to Log Your LLM: Unlock 99% Automation With Better Data
Traditional logging can't tell you if your LLM is getting worse. This article details the essential fields to log—from costs to verdicts—to debug silent failures, track quality, and automate up to 99% of your AI-driven workflows.
Jan 19
The 100x Cost Gap: When to Replace Your LLM With a Simple Rule
A single powerful LLM is often the most expensive and least reliable way to answer simple questions. This article explains how to build a routing layer that uses deterministic rules for simple queries, saving money and improving reliability, while reserving expensive models for the complex tasks they were built for.
Jul 28
Kimi K3: a 2.8-trillion-parameter open model that bets on linear attention
Moonshot's new open-weight flagship isn't chasing a leaderboard — it's a bet that a cheap million-token context changes what we build. What actually changed under the hood, how to run it, and where it sits against the open-weight frontier.