AI & Agents
Hands-on notes on Codex, Cursor, OpenCLI, MCP, Slock/Raft, Multica, multi-agent collaboration, enterprise SOP agents, model behavior, usage economics, and real AI tooling failures. · 11 articles
Tool access and integration
Give agents reliable access to browsers, local tools, and proxy-backed coding clients.
- OpenCLI for AI Agents: Turning Browser Workflows into Reusable CLI CommandsA practical OpenCLI guide for deciding when a browser workflow should become a reusable CLI, using logged-in Chrome safely, managing profiles and sessions, and keeping risky write actions behind human confirmation.
- How to Connect a Local stdio MCP Server to ChatGPT Web with OpenAI Secure MCP TunnelA hands-on setup and troubleshooting guide for bridging a local stdio MCP server into ChatGPT Web with OpenAI Secure MCP Tunnel, using Google's Analytics MCP on macOS as the real example.
- Fix Codex App instructions_required 400 Behind CLIProxyAPIA real fix for Codex App 400 instructions_required behind CLIProxyAPI (CPA): why filtering failed, how payload.default restored requests, and how to debug the proxy boundary.
Coordination and orchestration
Compare chat-based dispatch, issue-driven workflows, collaboration patterns, and review handoffs.
- Multica Guide: Issues, Agent Workflows, and Multi-Agent CollaborationA hands-on guide to Multica: starting with verifiable Issues, organizing Agent roles, handling handoffs, running multi-machine dispatch, managing governance, and avoiding drift.
- Does Multica Reduce Work or Add More? Lessons from a Real Squad WorkflowA hands-on review of Multica Squad in a real multi-agent workflow: where it reduces coordination overhead, where leader handoffs can stall, where the in-product code review view is still not a continuous interface, and how Squad differs from a strict workflow engine.
- Slock Is Now Raft: Installation, Agent Setup, and Common ProblemsA hands-on guide to Raft, formerly Slock: installation and machine connection, agent runtimes, legacy @slock-ai/* compatibility, @botiverse/* packages, PATH issues, multi-model setup, token cost, and practical multi-agent workflows.
- Multi-Agent Collaboration Patterns: Chat, Issue-Driven, SOP, and Hybrid WorkflowsA practical comparison of multi-agent collaboration patterns from Slock/Raft, Multica, and enterprise SOP agents: when chat works, when issue-driven workflows help, and when strict state and permissions matter more than more agents.
Architecture and usage economics
Look at SOP constraints, long-context behavior, account capacity, and the operating cost of AI workflows.
- Enterprise AI Agents Don’t Always Need Multi-Agent ArchitecturesA practical enterprise agent architecture based on a real AI reception system: use SOPs, state machines, tool allowlists, rule-based arbitration, and traces before adding more agents.
- I Used 1.05 Billion Tokens in Cursor: What a 93.31% Cache Hit Rate Taught MeReal Cursor Usage Events from 80 days: 1.05 billion tokens, a 93.31% cache hit rate, multi-million-token contexts, model usage, and the workflow changes that followed.
- ChatGPT Plus vs Team/Business: A Real-World Usage Capacity TestA historical May 2026 usage-snapshot test comparing one ChatGPT Plus account with two Team/Business workspaces across 5-hour and weekly windows, including the limits of what this sample can prove.
Model behavior and design experiments
Use controlled experiments to compare model behavior, design priors, and practical frontend output.