The fenbs blogAll articles
Page 13 of 48
- AI Agents for HR: Onboarding, Policies and QuestionsAI agents for HR do three jobs well: running onboarding checklists, answering policy questions from the handbook, and drafting announcements. What each looks like, which data to keep away from the agent, the US medical-confidentiality rules in plain words, and what a person must approve. · 6 min read
- AI Agents for Lawyers: Research, Drafting and Duty of CareAI agents for lawyers are useful for first drafts, summaries, research leads and matter task lists, and every one of those stays the lawyer’s work product. What each job looks like, what ABA Formal Opinion 512 asks on competence, confidentiality, communication and fees, what fabricated citations have cost lawyers in court, and how to keep client information safe with connected tools. · 7 min read
- AI Agents for Accounting and BookkeepingAI agents for accounting are useful as preparers: suggesting categories, lining up reconciliations, drafting client emails and running the month-end checklist. What each job looks like, the controls that keep an agent a preparer and never an approver, and the US rules on safeguarding taxpayer data in plain words. · 6 min read
- AI Agents for Real Estate Agents: Listings, Leads and Follow-UpAn AI agent can draft listing descriptions, lead follow-ups, showing schedules and transaction checklists for a real estate agent. What it drafts, what you approve, what the Fair Housing Act says about advertising in plain words, what happened to HUD’s 2024 AI guidance, and where the hand-offs should wait. · 7 min read
- AI Agents for Ecommerce: Catalog, Support and StockFive jobs an AI agent can do for an online store: product descriptions, catalog cleanup, support replies, inventory alerts and review responses. What it drafts, what a person approves, the FTC rules on reviews, Made in USA claims and shipping promises in plain words, and where the approvals wait. · 7 min read
- Shopify MCP: Storefront, Dev and Admin Servers ExplainedShopify’s MCP servers changed in 2026. What each one is today: the Dev MCP server for building, Customer Accounts MCP for signed-in shoppers, the UCP catalog, cart, checkout and order servers that replaced Storefront MCP, and how admin work reaches a store without an Admin MCP server. Setup, sign-in and safe use for each. · 8 min read
- BigQuery MCP: Querying Google Cloud Data From an AgentGoogle now offers two ways to put BigQuery behind an AI assistant: a managed remote MCP server and the open-source MCP Toolbox for Databases. What each one exposes, the IAM roles to grant, how to keep it read-only, how to stop a runaway query from costing you, and setup in Claude Code and Gemini CLI. · 7 min read
- Intercom MCP Server: Setup and What It Can DoIntercom runs an official remote MCP server for US and EU workspaces. Its 14 tools, which ones write, how OAuth sign-in works, setup in Claude, Claude Code and ChatGPT, and how to keep customer messages from steering the assistant. · 7 min read
- Agent Memory Handoff: Pass the Work to the Next AssistantTransfer the goal, decisions, completed work, evidence, and next step so another assistant can continue without guessing. Use a short, checked record that stays with the project. · 5 min read
- AI Enablement Team: Responsibilities and a Practical StartAn AI enablement team helps colleagues turn useful experiments into repeatable work. Define its service, ownership boundaries, reusable guidance, and feedback loop before expanding it. · 7 min read
- How to Make a Gemini Chat Private: Settings and LimitsUse a temporary chat, review activity and memory separately, and remove public sharing links. These steps reduce reuse and exposure without promising that Google stores nothing. · 5 min read
- Knowledge Management Team Structure: Who Owns What?Give shared knowledge a named owner, a reviewer, and a clear route for contributions. A small set of responsibilities can keep people and AI assistants working from the same approved record. · 5 min read
- NotebookLM Custom Instructions: Reuse a Clear Research BriefUse the notebook’s Custom chat setting to define the job, audience and evidence rules. Then test the instructions against sources you can check yourself. · 5 min read
- Notion AI Subscription Cost: Check What Your Plan IncludesNotion AI costs depend on your workspace subscription, included AI usage, and any separately purchased extras. Here is how to check the current offer and decide what you actually need. · 5 min read
- Prompt Library Web Apps for Reusable AI WorkCompare browser-based prompt libraries by organization, testing, sharing, and reuse. Pick the app that fits how your team actually runs prompts. · 8 min read
- RAG vs. Long Context: Which Approach Fits Your Work?RAG finds selected sources before an AI answers; long context gives it a larger set to consider together. Choose by the questions, the evidence, and the checks your work needs. · 7 min read
- Why Does Claude Argue With Me? How to Reset the TaskSeparate a useful objection from a misunderstood requirement or repeated debate. Give Claude a clear review boundary and a record of what you already decided. · 5 min read
- Zettelkasten in Google Docs: A Simple Linked-Note SetupUse Google Docs to keep small research notes, connect ideas, and find them again. Keep current project decisions separate from the material that informed them. · 5 min read
- Notion MCP vs the Notion API for AI AssistantsNotion’s hosted MCP server and its REST API reach the same workspace by different doors. Whose access each one uses, how each signs in, what each can reach, how both are rate limited, and which an assistant or a script should use. · 8 min read
- MCP Examples From Real WorkflowsNine Model Context Protocol workflows built on real, documented servers: GitHub, Sentry, Playwright, a filesystem, a database, Linear, Notion and a task board. Each with the host, the server, the prompt, and what a person checks afterward. · 7 min read
- MCP With OpenAI Models: What Works TodayOpenAI supports the Model Context Protocol in four places: the Responses API’s MCP tool, the Agents SDK, ChatGPT and Codex. Who acts as the client in each, how sign-in and approvals work, what ChatGPT does with read-only hints, and which to pick. · 8 min read
- MCP With Local Models via OllamaOllama runs the model and understands tool calls, but it does not connect to MCP servers itself. How people pair local models with MCP through a host, which hosts do it, where small models struggle with tools, and what stays private. · 7 min read
- Gemini CLI Plan Mode and Task TrackingGemini CLI has a plan mode: a read-only approval mode in which it researches and writes a plan before it touches your files. How to enter it, what it may do, where the plan goes, and how to keep the work on a board once the plan is approved. · 7 min read
- Using Claude Code and GitHub Copilot on the Same ProjectPlenty of teams run Claude Code in the terminal and Copilot in the editor on the same repository. It works well once three things are settled: one set of instructions, a clear split of who does what, and a rule that keeps two agents out of the same files. · 7 min read
- GitHub Copilot Agents vs Skills: What Each One IsA custom agent changes who is working: a role with its own instructions, tools and model. A skill changes what the current agent knows how to do: a procedure it loads when the task calls for it. Here is how to tell which one you need. · 7 min read
- How to Build Memory for an AI AgentFive decisions make up an agent’s memory: what it keeps, where it keeps it, who writes it, how it gets back into context, and how it is corrected. A build guide, one decision at a time, with the failure each choice invites. · 8 min read
- Claude Projects vs Skills vs ArtifactsA project is where the background lives, a skill is how a job gets done, and an artifact is the thing you hand to someone. What each one is, when to reach for which, and how the three work together on one piece of work. · 7 min read
- Claude Projects vs Custom GPTsA Claude Project is a workspace you and your colleagues work inside. A custom GPT is a configured assistant you build once and hand to other people to use. How they compare on knowledge, instructions, tools and sharing, and which suits which job. · 7 min read
- Trello MCP or CLI: How Should an AI Assistant Reach Your Board?Three doors into Trello for an AI assistant: Atlassian’s MCP server, a command-line tool, or the REST API with a key and token. Who holds the credential in each, what it can reach, how it is limited, and which to pick for which job. · 7 min read
- Todoist MCP vs CLI for AI AssistantsDoist publishes both an MCP server and a command-line tool for Todoist, and the API sits underneath them. How each one signs in, what scope it asks for, where read-only is possible, and which suits which assistant. · 7 min read
- Building an MCP Client in PythonA working Model Context Protocol client with the official Python SDK: connect to a server over stdio and Streamable HTTP, list its tools, call one, read the result properly, and put a model in the loop. · 8 min read
- MCP Specification Changes: What Changed and Why It MattersThe Model Context Protocol has five dated revisions, from 2024-11-05 to 2026-07-28. What each one added, removed and deprecated, how old and new clients and servers still talk, and what to check if you own either side. · 8 min read
- Context Engineering in Cursor: What Its Agent Sees, and How to Change ItCursor fills its agent’s context window from rules, @-mentions, its own search, skills, subagents and MCP servers. Here is what each one costs, what it keeps out, and where each kind of information belongs. · 8 min read
- Context Engineering for Multi-Agent Systems: What Each Agent Should SeeSplitting work across agents only helps if each one sees the right things and nothing else. How to write the brief, isolate each agent’s context, pass back summaries instead of transcripts, keep shared state outside the agents, and stop instructions contradicting each other. · 7 min read
- Multi-Agent Setups With GitHub Copilot: What You Can Run TodayGitHub Copilot can now run several agents at once in four different ways: subagents inside one chat, custom agents that hand off to each other, parallel sessions in VS Code, and a fleet of cloud agents on GitHub, including Claude and Codex. What each one is, and how to keep them from colliding. · 7 min read
- Choosing an Agent Memory System: Criteria That MatterThere is no best agent memory system in general, only the one that fits who reads the memory, who writes it and who has to fix it when it is wrong. Seven criteria, and how six common approaches score against them. · 8 min read
- Agent Memory Security: What an Assistant Should Not RememberWhatever an AI assistant remembers, every later session reads before it acts. Four ways memory goes wrong, the things an assistant should never keep, and five controls that make memory safe to share. · 7 min read
- Agents vs Tasks in Multi-Agent FrameworksIn CrewAI an agent is who does the work and a task is what the work is. Other frameworks draw the same line with different names, or barely draw it at all. Here is the split, framework by framework, and how it maps onto a task board. · 7 min read
- Context Switching Between Humans and AI AgentsEvery time work passes between you and an agent, or from one agent session to the next, something is lost. What goes missing at each kind of switch, what to write down so it does not, and where that state should live. · 8 min read
- Orchestrating Coding Agents Without Losing TrackCopilot’s cloud agent, Claude Code and Codex can all work on the same repository at once. Who hands out the work, how each agent is kept on its own branch, how a task is claimed across tools, and the order the results should be reviewed and merged. · 7 min read
- ChatGPT Connectors and Apps: What They Can ReachChatGPT’s connectors became apps, and apps now arrive inside plugins. What they are called today, what they can read and change, how they work inside ChatGPT Projects, how to add your own MCP server, and what an admin controls. · 7 min read
- Can Claude Projects Reference Each Other?Not in the chat app, as Anthropic documents it today: each project’s knowledge, chat search and memory stay inside that project. What the help center actually says, the one documented bridge in Cowork, and four ways to share material between projects without copying it by hand every week. · 7 min read
- Claude Projects in Chat vs CoworkAnthropic says projects work the same way in chat and in Cowork, and the same project can start either. What changes is what the work does: a chat answers, a Cowork task goes away and comes back with files. What each surface adds to a project, where memory is shared, and where it is not. · 7 min read
- Claude Connectors Explained: What They Can ReachA connector lets Claude read from and act in another service, as you and never as more than you. The three kinds — directory, custom and desktop extensions — where each one works, what the approval settings do, and the controls Team and Enterprise owners have. · 7 min read
- Claude Project Memory: What Carries Over Between ChatsEvery Claude project keeps its own memory, apart from your other projects and your everyday chats. What Claude saves there, what a new chat in the project actually starts with, how to see and edit it, and how to keep a conversation out of it. · 7 min read
- Claude Projects vs NotebookLM: Which Fits Your Research?NotebookLM, now called Gemini Notebook, is built to answer from your sources and turn them into study material. A Claude Project is a workspace for ongoing work that starts from your files. How they differ on sources, citations, outputs and sharing, and which suits which kind of research. · 7 min read
- Claude Projects vs Gemini GemsA Gem is a reusable set of instructions and files you call up in Gemini whenever you need it. A Claude Project is a workspace where the chats themselves live. How they compare on knowledge, instructions, sharing and connected apps, and where Gemini’s own notebooks fit in. · 7 min read
- How to Organize Claude Projects for Ongoing WorkOnce you have more than a handful of Claude Projects, the question is no longer how to run one but how to keep them all findable, current and separate. A structure for many projects: what gets its own project, how to name them, what goes where, when to archive, and where status lives. · 7 min read