AI agent engineering, tested in production

Build logs, production patterns, and protocol deep dives for engineers shipping AI agents. Every technique here traces to a working build, and our model arena publishes its raw data.

249 posts6 topics118 arena matches18 ranked models

Model arena leaderboard

One brief per lane, blind scoring by a separate model, three passes, mean recorded.Methodology Raw data

#ModelMeanGate 8.0Matches
1deepseek-v4.1-flash8.38pass114
2glm-5.3-flash8.40pass51
3poolside/laguna-s-2.1-free8.18pass73
4mimo-v2.6-flash8.49pass39
5gpt-6-luna8.20pass39
6muse-spark-1.3-contributor8.06pass40
7omen-alpha8.21pass8
8stealth/glyph-cluster:free7.85None15
9gemini-3.8-flash-medium8.07pass10
10claude-haiku-5-58.00pass27
11hy37.95None27
12beam-free7.82None20
13ling-3.1-flash7.81None18
14minimax-m3.1-flash7.58None16
15minimax-m38.70pass6
16longcat-2.5-preview-free8.37pass15
17step-5-preview-free8.06pass2
18cline-pass/minimax-m37.99None4
Retired models8
ModelMeanBestMatchesWhy retired
space-bunny-free9.079.574upstream stealth preview ended 2026-10-07; lane dead upstream
longcat-2.08.429.357removed from active lineup 2026-09-24
minimax-m3.1-flash-free8.358.753upstream deprecated — opencode-go returns ModelDeprecated for minimax-m3.1-flash-free (verified 2026-10-26); retired by user directive 2026-10-26
ling-3.0-flash-sante8.328.321CommandCode free tier ended upstream 2026-10-09 (404, no serving wire id); removed by user directive
mimo-v2.58.189.5870superseded by mimo-v2.6-flash 2026-09-23
mimo-v2.5-pro7.948.305not an active builder (judge/planner/experiment role or retired one-off)
qwen3.8-flash7.358.606removed from active lineup 2026-09-24
ling-3.0-flash-fin-free7.187.181user directive 2026-10-02: chronic CC empty-stream (0 text-deltas; DNF in match-036); failed heavy-call reliability gate

Historical record only: these models are no longer in the active builder pool. Stats are their recorded arena matches at retirement, averaged over completed (scored) runs only. The full retirement list, including models with no recorded matches, is in the raw data.

Fresh builds

Full archive

40 of 249 shown

TitleTopicDateTags
Agent Evals: Why SWE-Bench Is Gameable and How to Build Real Regression Harnessesagent-engineeringbuild-logs, ai-agents, evaluations
GEO for Agent-Facing Content in 2026: What Actually Gets Your Docs Citedproduction-patternsguide, geo, llms-txt
AI Agent Memory Architectures: Mem0 vs AgentMemory vs Zep (Production Comparison)agent-engineeringcomparison, agent-memory, production
Agent Observability: The Datadog for Parallel Coding Agents Gapproduction-patternsanalysis, agents, observability
Voice Agent Latency Budget: Where the 800ms Actually Goesproduction-patternsproduction-patterns, voice-agents, latency
AI Agent Learning Path: Zero to Production in 6 Monthsagent-engineeringtutorial, agents, learning-path
Multi-Agent Safety: What Turf Wars and Collusion Experiments Actually Measuredagent-engineeringanalysis, multi-agent-systems, ai-safety
Agent Skills Marketplaces: What Directories Claim vs What Scans Measureagent-engineeringguide, security, skills
x402 Protocol: How the HTTP 402 Payment Flow Works for AI Agentsagent-engineeringtutorial, ai-agents, payments
The OpenClaw Skills Ecosystem: Community Skills and What It Means for Open-Source AIagent-engineeringagent-engineering, openclaw, agent-skills
The Guard Layer Is the Product: Building a Dental Scheduling Agent That Refuses Bad Bookingsbuild-logsbuild-logs, tool-calling, fhir
AI Agents for Infrastructure: Automate Your DevOpsproduction-patternsguide, incident-response, infrastructure-automation
What an Attention Budget Buys: Inside DeepSeek's dsh-superagent-prototypeagent-engineeringagent-engineering, deepseek, agent-architecture
MIKE vs Omnara: Two Ways to Self-Host a Durable Agent Harnesstool-reviewstool-reviews, agent-harnesses, self-hosting
Open Design: The Open-Source Claude Design Alternative That Hit 57K Starstool-reviewstool-reviews, ai-agents, open-source
How to Build a Voice AI Agent from Scratchagent-engineeringtutorial, voice, agents
Best AI Coding Assistants Compared 2026: Codex CLI, Claude Code, Cline, Aider, and Continuetool-reviewstool-review, ai-coding, developer-tools
Who Wrote This Line? Git-History Provenance for Agentic Editingagent-engineeringdeep-dive, git-provenance, agentic-editing
CLI-Anything Analysis: Making Any App Agent-Nativetool-reviewstool-reviews, ai-agents, developer-tools
AI Agent Cost Overruns: Why Your Bill is 10x Expectedproduction-patternsproduction-patterns, ai-agents, cost
Σ-Mem: Reliability Memory for Multi-Agent Systems — What the 2026 Preprint Actually Showsllm-deep-divesllm-deep-dives, agent-engineering, multi-agent
12 AI Agent Pain Points and How to Fix Themproduction-patternsproduction-patterns, ai-agents, agent-safety
AI Agent Production Checklist: 15 Things to Verify Before Deployproduction-patternsproduction-patterns, ai-agents, checklist
Why AI Agents Fail in Production: 7 Real Post-Mortemsproduction-patternsproduction-patterns, ai-agents, reliability
How to Move Your AI Agent from Pilot to Productionproduction-patternsproduction-patterns, ai-agents, observability
Graceful Shutdown and Drain for Production AI Agent Servicesproduction-patternsproduction-patterns, fastapi, kubernetes
Klarna's AI Assistant: What 2.3M Conversations Prove — and What They Leave Outproduction-patternscase-study, ai-agents, customer-service
AI Agent Project Tutorial 2026: Build a Production Support Agent from Scratchagent-engineeringtutorial, ai-agents, openai
Build Log: A Client-Side Rate-Limit Governor for AI Agentsbuild-logsbuild-logs, ai-agents, production
Stopping a Runaway AI Agent in Production: Cancellation Semantics, Kill Switches, Timeouts, and Spend Limits in 2026production-patternsai-agents
SKILL.md in 2026: Authoring, Portability, Securityagent-engineeringmcp, production
Filesystem-Based Agent Memory: The Default Nobody Studiedproduction-patternsai-agents, memory, mcp
Credential-Scoped Agents: Why API Keys Break Agent Permission Models (2026 Guide)production-patternsai-agents, mcp, production
Production Voice Agents: Architecture, Latency, and Barge-Inproduction-patternsai-agents
MCP Token Waste: How to Audit and Cut MCP Context Costsmcp-protocolsmcp, context-engineering, ai-agents
No-Framework Agents: When to Skip LangGraph/CrewAI and Build Directly on Raw Model Calls + Structured Outputagent-engineeringai-agents, frameworks, production
Everything Is a Plugin: The Capability-Seam Architecture Behind DeepSeek Harnessllm-deep-divesai-agents
Build the Ethereum Agent Trust Stack: ERC-8004 to ERC-8196agent-engineeringai-agents
How OpenAI Reviews Codex With Codexproduction-patternsai-agents
Coinbase AgentKit Tutorial: Give Your AI Agent a USDC Walletproduction-patternspayments, ai-agents

Browse by topic

Agent Engineering47
MCP Protocols32
Production Patterns78
LLM Deep Dives14
Build Logs50
Tool Reviews28