The fenbs blogAll articles
Page 9 of 48
- Claude Code Usage Limits: How to Check and Stretch ThemClaude Code on a subscription draws on the same five-hour and weekly limits as the rest of Claude. How to read /usage and the limit messages, what Claude Code does when a limit hits mid-task, and the model and compaction choices that make the allowance last. · 7 min read
- Claude Code Sandboxing: What It Blocks and How to Turn It OnClaude Code’s sandbox puts an operating-system boundary around every shell command Claude runs: where it can write, and which hosts it can reach. What it blocks by default, how to switch it on, the settings that matter, and what it leaves uncovered. · 7 min read
- Claude Code Checkpoints and Rewind: Undoing an Agent’s ChangesClaude Code saves a checkpoint before every prompt, and /rewind takes your code, the conversation or both back to it. Every option in the rewind menu, exactly what a checkpoint restores and what it never will, and a way to experiment that leans on both checkpoints and git. · 6 min read
- Claude Code With Git Worktrees: Parallel Sessions Without CollisionsStart Claude Code with --worktree and each session gets its own checkout and branch, so two sessions never edit the same files. How the flag works, what a worktree shares with your main checkout, how isolation is enforced, and how to clean up afterward. · 7 min read
- Git Branching Strategy for Small Teams (and AI Agents)Trunk-based development, GitHub flow and Git flow compared for a team of two to ten, and what changes when AI coding agents open branches too: short-lived branches, one worktree per session, and a person at the merge. · 7 min read
- Claude Code GitHub Actions: Setup and Safe UsesThe official Claude Code action runs Claude inside your own GitHub workflows, answering @claude mentions or running a prompt on any event. How to set it up, what the three layers of permissions are, what it costs in words, and which uses are safe to leave running. · 8 min read
- Claude Code Headless Mode: Running It in Scripts and CIHeadless mode is Claude Code run with -p: one prompt in, one result out, no interface. What it loads, the three output formats, how to settle permissions before the run, and five examples you can put in a script or a pipeline. · 7 min read
- Godot MCP: Game Development With an AI AgentGodot has no official MCP server. What the community servers do, how the two best-known ones differ, how to set each up with Claude Code or Cursor, and the checks to run before letting an agent loose on your project. · 7 min read
- Higgsfield MCP: Generating Video From an AI AssistantHiggsfield runs its own remote MCP server, so Claude, ChatGPT and other assistants can generate images and video with your Higgsfield account. What it does, how to connect it, how credits work, and the content rules that come with it. · 7 min read
- How to Implement Kanban in a WeekKanban does not need a transformation program. Map the work you already do, put it on a board, cap what is in progress, write the rules down and start watching flow. A five-day plan for a small team, and what to leave until later. · 8 min read
- Cumulative Flow Diagram: Reading Bottlenecks at a GlanceA cumulative flow diagram stacks the number of items in each workflow state, day by day. What the bands mean, how to see work in progress growing, the shapes that point at a bottleneck, and how to draw one by hand from a daily count. · 7 min read
- Sprint Velocity: What It Measures and Why Not to Chase ItSprint velocity is the amount of estimated work a team finished in a sprint. How to calculate it, what the major tools count, the ways it gets misused, and why counting finished items is often the better number. · 6 min read
- Stand-Up Meeting Questions That Are Not a Status ReportThe three classic questions turn a stand-up into a round of personal updates. Better questions ask about the work, the goal and who needs help. Five question sets to rotate, a written set for async stand-ups, and the questions to stop asking. · 6 min read
- Manus AI Agent: What It Does and Its LimitsManus is a general AI agent that works on its own virtual computer: it plans a task, browses, writes code and files, and hands back a finished result. How it works, where it can reach, how it compares with what used to be ChatGPT agent, and the limits to plan around. · 8 min read
- GitHub Copilot Models: Which Ones You Can Pick and WhyGitHub Copilot lets you choose the model behind Chat, agent mode, the CLI, the Copilot app and the cloud agent, or leave it to Auto. Which models are on the list, where the picker is in each tool, how AI credits replaced premium requests, and how to choose a model for agent work. · 7 min read
- Is GitHub Copilot Down? Checking Status and Working Around ItWhen Copilot stops answering, GitHub’s status page tells you whether it is an incident, and the error text usually tells you when it is not. How to read githubstatus.com, which errors are limits or network problems rather than outages, and how to keep working, or stop cleanly, until it is back. · 6 min read
- Microsoft Learn MCP Server: Official Docs in Your Coding AgentThe Microsoft Learn MCP Server gives a coding agent search and fetch access to Microsoft’s official documentation and code samples, free and without a sign-in. What its three tools answer, how to add it to VS Code and Claude Code, how to keep its answers small, and where it stops. · 6 min read
- Microsoft Work IQ MCP Servers ExplainedMicrosoft ships two generations of Work IQ MCP: one unified server with generic tools, and a set of per-app servers for Mail, Calendar, Teams, Word and SharePoint that Copilot Studio now calls legacy. What each does, how to connect a coding agent, what is preview, and why Microsoft’s own licensing pages disagree. · 7 min read
- Feature Prioritization Frameworks: RICE, MoSCoW, KanoRICE, MoSCoW, Kano and WSJF each answer a different question about what to build next. What each one measures, when it fits, a side-by-side comparison, templates to copy, and how to record the framework your team picked so it survives the next argument. · 7 min read
- RICE Prioritization: Formula, Example and PitfallsRICE scoring ranks features by reach, impact, confidence and effort. The formula as Intercom published it, a worked example with five features, a spreadsheet you can build in two minutes, and the pitfalls that make a RICE score look more certain than it is. · 6 min read
- Customer Feedback Management: From Inbox to RoadmapCustomer feedback management is the loop that turns scattered comments into decisions: capture from every channel, tag, merge duplicates while counting demand, decide, and tell people what happened. A small-team version you can run in an hour a week. · 7 min read
- Product Roadmap Template: Now, Next, LaterA Now, Next, Later product roadmap template in three formats: a spreadsheet for Excel or Google Sheets, a one-slide layout for PowerPoint, and a filled-in example. Plus what changes between the internal roadmap and the one customers see. · 7 min read
- Agentic Commerce: When AI Agents Buy ThingsAgentic commerce is shopping where an AI agent finds, chooses and sometimes pays on a person’s behalf. What the Agentic Commerce Protocol, Google’s AP2 and UCP actually specify, what changed in 2026, which claims are ahead of reality, and the risks to settle before an agent spends money. · 8 min read
- AI Agents for Beginners: What They Are and a First Safe SetupAn AI agent is an AI model that can use tools, one step after another, to finish a job you gave it. What that means in plain words, what to hand one first and what not to, and a first setup with one tool and one board that you can check and switch off. · 7 min read
- What Is an Agentic OS? The Idea and the Products“Agentic OS” means three different things depending on who says it: an operating system that hosts AI agents as users, an enterprise platform with “OS” in its name, or a research design for managing agents like processes. What each vendor actually ships, what is still preview, and which claims are marketing. · 7 min read
- AI Agent Architecture: The Parts Every Agent HasEvery AI agent, from a sixty-line script to a coding assistant, is built from the same six parts: a model, instructions, tools, memory, a loop and guardrails. What each part does, the decision you make for it, how it fails, the common multi-agent patterns, and the two things that sit outside the agent. · 8 min read
- ServiceNow MCP Server: Options and SetupServiceNow ships its own MCP server, MCP Server Console, and it can also call other MCP servers. What it needs, how to connect Claude or Claude Code to it with OAuth, where community servers fit, what the MCP Registry and Docker list, and why the ServiceNow user the assistant signs in as matters more than any setting. · 7 min read
- SharePoint MCP: Documents From an AgentMicrosoft offers a SharePoint MCP server in preview, a newer Work IQ route that Copilot Studio now prefers, and Claude reaches SharePoint through its Microsoft 365 connector. What each can read and change, where Microsoft’s own pages disagree, how to connect Claude Code, and how Selected permissions keep a custom agent to one site. · 7 min read
- Async Communication: Rules for Teams That Rarely MeetAsync communication works when a team agrees what goes in writing, how fast each channel gets an answer, and where work lives. What to move out of meetings, a response-time ladder, how to write a message that settles things in one round, and why the board beats the thread. · 7 min read
- Managing a Remote Team: Practices That Scale Down to FiveMost remote team management advice is written for companies with hundreds of people. Seven practices that still work for a team of five: write it down first, a small overlap window, decisions in writing, one board, results over presence, regular one-on-ones and a written onboarding. · 7 min read
- One-on-One Meeting Template for ManagersA one-on-one belongs to the person who reports to you. A 30-minute agenda that puts their items first, questions for when they bring nothing, a first one-on-one with a new report, where private notes go, and how follow-ups become tasks without anything sensitive landing on the team board. · 7 min read
- Client Project Tracker: What Clients Should SeeA client project tracker is not your internal tracker with the client added. What clients should and should not see, the columns written in their words, what a client should be able to do, how often to update it, and how to share it without sharing everything. · 7 min read
- Can AI Agents Talk to Each Other? A2A, MCP and HandoffsYes, in three ways: inside one framework through handoffs and messages, across vendors through the A2A protocol, and indirectly through a shared list both can read. How each works, where MCP fits, and the pitfalls that start when agents take each other’s word for things. · 8 min read
- Can AI Agents Send Emails? Yes, and Here Is When They Should NotAI agents can send email through connectors to your mailbox, through email APIs, or by working a browser. Which assistants send today and which stop at a draft, how to put an approval gate in the right place, the messages an agent should never send alone, and where the CAN-SPAM line falls. · 7 min read
- Can AI Agents Make Phone Calls? Voice Agents and the RulesAI voice agents can place and answer real phone calls, and Google Search already calls local businesses for its users. How a voice agent connects to the phone network, what the FCC’s 2024 ruling means for calls with an AI voice, the other US rules that still apply, and which calls to keep for people. · 7 min read
- What AI Agents Really Cost to Run (Without Price Lists)An AI agent’s bill is mostly tokens, and it grows with the number of steps and the size of the context at each step, not with the length of your request. What drives the cost, how caching and batching cut it, the limits that stop a runaway agent, and how to track cost per task. · 8 min read
- Zoho MCP: Connecting Zoho Apps to an AI AssistantZoho MCP is a console where you build your own MCP servers from Zoho and third-party tools, then paste one URL into Claude, ChatGPT, Cursor or VS Code. How setup works, where the scopes really come from, why the URL is a password, and what to approve by hand. · 7 min read
- Databricks MCP: Managed Servers for Unity Catalog and GenieDatabricks hosts MCP servers for Genie, AI Search, Databricks SQL and Unity Catalog functions, governed by Unity Catalog. Which one to expose, what is preview and what was renamed, OAuth scopes, the write default to change, connecting Claude Code, and where Box and other outside servers fit. · 7 min read
- How to Install Gemini CLI on Windows, Mac and LinuxGemini CLI installs with one npm command once Node.js 20 or later is in place. The requirements, the install on each system, the other routes (Homebrew, MacPorts, npx, conda), which sign-in still works after June 18, 2026, checking and updating the install, and the errors people hit. · 7 min read
- Agentic Coding Best Practices: Ten Habits That Keep Agents on TrackTen habits that work with any coding agent, whether Claude Code, Codex, Copilot, Cursor or Gemini CLI: write the task first, plan before large changes, keep tasks small, give the agent a check it can run, review the diff yourself, and keep all the work on one board. · 8 min read
- Vibe Coding Tools Compared: App Builders and Coding AgentsVibe coding tools come in two kinds: browser app builders such as Lovable, Bolt, v0 and Replit Agent, and coding agents such as Claude Code. What each one builds, where your code ends up, which ones you can start free, and which fits a founder, a designer, a product manager or a developer. · 7 min read
- Devin vs GitHub Copilot Cloud AgentTwo agents you hand a task to and meet again at the pull request. Where each one runs, where a task can start, what it is allowed to do on GitHub, how you give it standing instructions and tools, and which kind of team each suits. · 7 min read
- Claude Code vs Cursor vs GitHub Copilot: Picking One for a TeamFor one developer the choice is about taste. For a team it is about what an admin can enforce, what you can see afterward, where code goes, and whether everyone can share one setup. The three side by side on those questions, with links to the head-to-head comparisons. · 8 min read
- LangGraph vs CrewAI: Choosing a Multi-Agent FrameworkLangGraph gives you a graph over shared state and leaves every decision to you. CrewAI gives you a team of role-based agents and an event-driven flow around them. How each models the work, keeps state, and pauses for a person, with code for both. · 7 min read
- Best MCP Servers by Job: A Curated Map, No HypeThere is no single best MCP server, only the right one for a job. A map of well-documented servers grouped by what you need an assistant to do, each linked to a setup guide, plus the checks that separate a good server from a risky one. · 6 min read
- Where to Find MCP Servers: Registries and DirectoriesThe MCP server list you want depends on where you will use it: the official MCP Registry, GitHub’s registry, Docker’s catalog, Anthropic’s Connectors Directory, or the catalogs inside Cursor and Gemini CLI. What each one lists, what its labels do and do not prove, and six checks before you connect anything. · 6 min read
- How to Make an MCP Server in TypeScript, Step by StepA step-by-step build of an MCP server with the official TypeScript SDK: a tool with structured output, a resource and a prompt, compiled to JavaScript, tested in-process with node:test, checked in the MCP Inspector, and packaged so anyone can start it with npx. · 7 min read
- n8n MCP: Using n8n as an MCP Server and Clientn8n speaks MCP in both directions. As a server, it lets Claude, Cursor or Codex find, run and even build your workflows; as a client, its nodes call any MCP server. Which of the four routes to use, how to add n8n to Claude, and what each one exposes. · 7 min read