The fenbs blogAll articles
Page 19 of 48
- Is Your Team Ready for AI Agents? A Readiness ChecklistSeven questions to answer before you connect the first AI assistant to your team’s work: where the work lives, who owns it, what the assistant may touch, which jobs are safe, who reviews, what gets recorded, and how you undo a mistake. · 7 min read
- AI Agent Security Best Practices for Small TeamsCoding agents in a terminal, browser agents clicking through websites, and assistants built into the tools you already pay for all need the same few rules. Here they are, with the settings that enforce them. · 8 min read
- OWASP Guidance on AI Agents, Explained for Non-SpecialistsOWASP now publishes a Top 10 for agentic applications alongside its Top 10 for LLM applications. Here is what each list is, the entries that matter to a small team, and one practical control for each. · 7 min read
- AI Agent Governance for Teams Without a Security DepartmentA one-page AI agent policy that a team of five to fifty can adopt in an afternoon and still be following a year later: who connects what, which tools, which data, who owns each agent, and when access ends. · 7 min read
- Giving an AI Agent a Task It Can Finish: Scoping and AcceptanceSeven parts turn a vague request into a card an agent can finish and you can check: the outcome, where, the constraints, acceptance criteria, how it will be checked, what to do when blocked, and a size limit. With a template and three rewrites. · 7 min read
- How to Add Authentication to an MCP ServerThe pieces an MCP server has to implement before an assistant can sign in to it: protected resource metadata, a proper 401, token validation with an audience check, and scopes. Plus when to reuse an authorization server and when a stdio server needs none of it. · 7 min read
- OAuth for MCP Servers: How Sign-In Works for AI AssistantsWhat actually happens between adding a remote MCP server and your assistant being able to use it: the discovery, the browser sign-in, the consent screen, the token, and what you should check on the way. · 8 min read
- Scanning MCP Servers for Security ProblemsWhat an MCP security scanner checks, two open tools that do it and how to run them, where a scan belongs when you add a server, and the problems no scan can see. · 6 min read
- Do You Need an MCP Gateway? Security Trade-offs ExplainedAn MCP gateway puts one controlled door between your assistants and your servers. What it gives you, what it costs, when a small team is better off with per-client settings, and the token rule a gateway must not break. · 7 min read
- MCP Security in VS Code: Trusting Servers SafelyVS Code gives you four places to control an MCP server: where it is configured, the trust decision before it starts, the approval before each tool call, and the policies an organization can set. What each one does, and the settings that quietly switch them off. · 7 min read
- Human-in-the-Loop AI: 12 Everyday ExamplesTwelve places where AI does the work and a person makes the call: agent pull requests, support replies, invoices, refunds, publishing, deletions, hiring screens and more. For each, what the AI does, where the person steps in, and what happens if nobody does. · 7 min read
- Human in the Loop vs Fully Agentic AI: How to ChooseWhether an agent should wait for a person or just get on with it is not a question about the agent. It is a question about the work: how easily it is undone, how far a mistake spreads, how often it happens and how quickly it can be checked. · 7 min read
- AI in the Loop vs Human in the LoopThe two phrases sound like mirror images, and in a sense they are. The difference is who owns the decision: in one, the AI runs the process and a person is consulted; in the other, a person runs it and the AI is consulted. · 6 min read
- Verifying AI-Generated Work: A Review Process That ScalesReading everything an agent produces does not scale, and reading nothing is not review. A process that holds up has three parts: evidence the agent must attach, a check that fits the kind of output, and a verdict written down where the next person will see it. · 7 min read
- Will AI Agents Replace Project Managers?AI agents already do a real share of a project manager’s routine: status, triage, first drafts and chasing. They do not set priorities, weigh trade-offs, manage people or carry the blame. Here is where the line sits today, and what that means for the job. · 7 min read
- AI Agent Task Decomposition: Breaking Work Into Tasks an Agent Can FinishA goal is not something an agent can pick up; a short row of small, outcome-shaped cards is. How to slice a goal, record what depends on what, order the result in Next Up, and let an agent propose the split while a person approves it. · 7 min read
- Sprint Planning With AI: A Workflow That Keeps People in ChargeAn AI assistant can do most of the preparation for sprint planning: tidy the backlog, find duplicates, draft a sprint goal, and lay out what fits. The choices in the room stay with people. A workflow for before, during and after the meeting. · 7 min read
- Jira MCP With Claude Code: Setup, What It Can Do and Its LimitsHow to connect Claude Code to Jira through Atlassian’s own remote MCP server: the one command, the browser sign-in, the first prompts worth trying, how permissions work, and the limits to know before you rely on it. · 7 min read
- What Can the Jira MCP Server Do? Tools, Permissions and GapsA plain guide to what Atlassian’s MCP server lets an AI assistant do in Jira: the tools in each permission group, what it can read and change, whose rights it uses, what admins control, and what it does not cover. · 6 min read
- Linear MCP With Claude Code: Install and Use ItConnect Claude Code to Linear through Linear’s hosted MCP server: the install command, the OAuth sign-in, a read-only option, token access for headless machines, and the prompts that make it useful day to day. · 6 min read
- Notion MCP With Claude Code: Install, Connect and UseHow to connect Claude Code to your Notion workspace with Notion’s hosted MCP server: the command, the OAuth sign-in, what the tools can read and change, which features depend on your plan, and the rate limits. · 6 min read
- Trello MCP With Claude Code: Setup and Everyday UseHow to connect Claude Code to Trello through the official Trello MCP server from Atlassian: the one command, the browser sign-in, what it can and cannot do yet, the prompts worth using day to day, and how to remove it. · 6 min read
- Asana Claude Connector: MCP Setup, Tools and PermissionsHow to connect Claude to Asana through Asana’s official MCP server, in the Claude app and in Claude Code: the two setup routes, the tools it exposes, whose permissions it uses, and the limits worth knowing first. · 6 min read
- monday.com MCP With Claude: Setup GuideHow to connect Claude to monday.com through monday’s hosted MCP server: the connector in the Claude app, the one-line setup in Claude Code, what the tools can change, whose permissions they use, and what an admin controls. · 6 min read
- ClickUp MCP With Claude Code: Setup and LimitsHow to connect Claude Code to ClickUp through ClickUp’s own hosted MCP server: the one command, the browser sign-in, first prompts, permissions, the daily call limits, and how to remove it. · 7 min read
- Todoist MCP With Claude: Your To-do List in the ChatHow to connect Todoist to Claude and to Claude Code through Doist’s official MCP server: the connector, the plugin, the one command, first prompts, what it can change, and the limits worth knowing. · 7 min read
- Claude and the GitHub MCP Server: Issues, PRs and ProjectsHow to connect Claude Code to GitHub’s official MCP server: the remote address, why Claude Code uses a personal access token, how to choose toolsets, switch on Projects, go read-only, and the limits to know. · 8 min read
- What Can the Notion MCP Server Do (and Not Do)?A plain guide to what Notion’s hosted MCP server lets an AI assistant do in your workspace: the jobs its tools cover, whose access it uses, what admins control, the limits that shape how it edits, and the things it does not do at all. · 7 min read
- How to Use Notion MCP With ChatGPTChatGPT reaches Notion in two ways: the Notion app in ChatGPT’s settings, and Notion’s own MCP server added as a custom app in developer mode. What each does, the steps for both, which plans get which, and the quirks worth knowing. · 6 min read
- What Does the Linear MCP Server Do?What Linear’s hosted MCP server lets an AI assistant find, create and change, going by what Linear itself documents: the objects it reaches, whose permissions it uses, the read-only and app-identity options, and what Linear does not spell out. · 6 min read
- How to Use the Jira MCP Server in CursorConnect Cursor to Jira through the Atlassian Rovo MCP Server: the one-click plugin or an mcp.json entry, the OAuth sign-in and what to do when an admin blocks it, which tools to let run without asking, and token access for machines without a browser. · 6 min read
- The GitHub Copilot Coding Agent: How It Takes a Task From an IssueAssign an issue to Copilot and it works on GitHub, not in your editor: it reads the issue, changes code on its own branch, runs the tests and hands you a pull request. Here is what happens at each step, where its limits are, and how to keep that work on your board. · 7 min read
- GitHub Copilot Agent Mode: Best Practices for Real WorkAgent mode edits your files and runs commands in your editor while you watch. It works best with one scoped task per session, approvals you actually read, a review before every commit, and a record of the work that outlives the chat. · 7 min read
- Does GitHub Copilot Read AGENTS.md? How Instructions Files WorkYes, in most places but not all. Copilot’s cloud agent, the Copilot CLI, chat in VS Code and code review on GitHub read AGENTS.md; chat in several other editors does not. Here is which surface reads which file, and a layout that covers them all. · 7 min read
- Cursor Rules for AI Projects: Keep the Agent on the TaskProject rules in .cursor/rules tell Cursor’s agent how your codebase works. Written well, they also keep it on one task: which task it is, what it may touch, and where it reports what it did. The four rule types, what they do not cover, and a board rule you can copy. · 7 min read
- MCP vs REST API: What Changes When an AI Is the ClientAn MCP server rarely replaces an API. It usually sits on top of one, rewritten for a client that reads the interface at run time and decides for itself what to call. Here is what that changes, and when you need which. · 7 min read
- MCP vs Function Calling: How They RelateFunction calling is how a model asks for a tool to be run. MCP is how an application finds tools and reaches the servers that run them. They are two layers of the same loop, and most assistants use both. · 7 min read
- MCP vs Claude Skills: When to Use WhichAn MCP server lets Claude reach a system. A skill teaches Claude how to do a job. One is access, the other is know-how, and the useful setups usually have both. · 7 min read
- How to Build an MCP Server for Your Own ToolA working MCP server in TypeScript in about fifty lines: three tools, tested in the MCP Inspector, connected to Claude Code. Then the part that decides whether assistants use it well: how to design the tools. · 8 min read
- Context Engineering for AI Agents: A Practical GuideAn agent only knows what is in its context window at the moment it acts. Context engineering is deciding what goes in, what stays out, and where the rest lives so the next session can find it. · 7 min read
- Context Rot: Why Long AI Sessions Get Worse, and What to DoThe longer a conversation with an AI model runs, the less reliably it uses what is in it. That effect has a name, context rot, and a short list of habits that keep it from spoiling a day’s work. · 6 min read
- MCP vs RAG: Different Jobs, Often Used TogetherRAG decides what text the model reads before it answers. MCP gives the model tools it can call while it works. One is about knowledge, the other about reach, and an MCP server is often where the retrieval lives. · 7 min read
- MCP vs A2A: Tools for Agents vs Agents Talking to AgentsMCP connects an agent to the tools and data it uses. A2A connects one agent to another agent it cannot see inside. They sit side by side, and most teams will meet MCP long before they need A2A. · 7 min read
- Context Engineering vs Prompt EngineeringPrompt engineering is about writing the instruction well. Context engineering is about everything else the model sees when it acts. Here is where one ends, where the other begins, and why the second term arrived. · 6 min read
- AI Context Windows Explained for People Who Manage WorkAn AI assistant can only work with what is in its context window at that moment. What a context window is, what fills it, why a bigger one is not the whole answer, and what that means when you hand work to an assistant. · 7 min read
- Agent Memory: Short-Term, Long-Term and SharedAn AI agent starts every session with an empty context window. Agent memory is whatever puts the right knowledge back. The main kinds, what each is good for, and who can see and change what the agent remembers. · 7 min read
- Can Claude Projects Be Shared With Others?Yes, inside an organization on Team and Enterprise plans, with two permission levels. What a shared project hands over, what stays private, and what to use when people and assistants need the same state. · 7 min read
- Claude Projects vs Claude Code: Which to Use for WhatA Claude Project is a chat workspace with knowledge and instructions. Claude Code is an agent that works inside your repository. They answer different questions, and most real work uses both. · 7 min read