The fenbs blogAll articles
Page 17 of 48
- Aider vs Claude Code: Git-First Pair Programmer or Agent LoopAider is an open-source pair programmer for the terminal that commits every edit to Git and works with almost any model. Claude Code is Anthropic’s agent that searches, edits, runs and checks on its own. How they differ on context, commands, undo, models and extensions, when each wins, and where OpenCode and Codex fit. · 7 min read
- Devin vs Claude Code: Autonomous Engineer or Terminal PairDevin is Cognition’s autonomous engineer: you hand it a ticket from Slack, Linear or Jira and it works on its own machine until there is a pull request. Claude Code is Anthropic’s agent that works beside you, and can also run in the cloud. Delegation against collaboration, the review burden, how to scope a delegated task, and Devin vs Codex. · 7 min read
- Google Jules vs Codex: Cloud Coding Agents ComparedJules and Codex cloud both take a task, work on a copy of your repository on someone else’s machine and hand back a diff. They differ in the plan step, where you can start a task, how limits are counted and what each can connect to. Plus short comparisons with Gemini CLI and the Copilot cloud agent. · 8 min read
- Gemini Code Assist vs GitHub CopilotTwo IDE assistants sold to teams: Google’s through Google Cloud, GitHub’s through GitHub. How their editions, supported editors, agent modes, code review, context, admin controls and MCP support compare, and where Claude Code and Codex fit instead. · 7 min read
- Using Codex CLI With Local or Other ModelsCodex CLI can run on a model on your own machine through Ollama or LM Studio, or on another provider such as OpenRouter. The flags and config.toml entries that do it, why the provider must speak the Responses API, what gets worse, and when it is worth the trouble. · 6 min read
- Codex Usage Limits: How to Check Them and Make Them LastCodex on a ChatGPT plan comes with a usage allowance that resets every five hours, may have a weekly cap, and is shared by the CLI, the IDE extension, the desktop app and cloud tasks. How it is counted, the figures OpenAI publishes, where to see what is left, what happens when it runs out, and the habits that make it last. · 6 min read
- Codex Subagents and Multi-Agent RunsCodex can split a job across subagents inside one session, run several cloud chats side by side, and be driven in parallel from a script. What each option does, how to set it up, and which one fits the work in front of you. · 8 min read
- ChatGPT Projects vs Custom GPTs: Which to BuildA ChatGPT project is a workspace you keep coming back to. A custom GPT was an assistant you packaged for other people, and OpenAI is retiring GPTs in favor of plugins. How the two compare on files, instructions, memory, sharing and apps, and how to move a GPT before it stops running. · 8 min read
- ChatGPT Apps and Connectors Not Showing or Not Working: FixesConnectors are now apps inside plugins, which is the first reason people cannot find them. The other causes, in the order to check them: your plan, your workspace admin, developer mode, an expired sign-in, the app not being on in this chat, region and platform, and the model. · 8 min read
- ChatGPT Custom Instructions That Actually WorkCustom instructions work when they are a few short rules you could check an answer against, not a biography. Where they live now, how to write them, examples for work, coding and writing, how they sit alongside memory, and what they cannot do. · 6 min read
- Moving From ChatGPT to Claude: Memory, Projects and GPTsMemory is the one thing Claude can import from ChatGPT, and it is a copy and paste. Everything else is rebuilt: custom instructions, projects, custom GPTs, apps and connectors. What carries over, how to move each piece, and a checklist. · 7 min read
- GitHub Copilot SDK: Building Your Own Agent on CopilotThe GitHub Copilot SDK puts the agent runtime behind the Copilot CLI inside your own program: sessions, tools, MCP servers, custom agents and a permission callback, in six languages. What it is, how it signs in, a short example, and where it sits next to Copilot Extensions and Claude’s SDK. · 7 min read
- GitHub Copilot Custom Agent Examples (.agent.md)Three complete custom agent files you can copy, a reviewer, a planner and a docs writer, with the frontmatter explained field by field: tools, MCP servers, handoffs, and where the files go for a repository, an organization or an enterprise. · 8 min read
- GitHub Copilot’s Cloud Agent With Azure BoardsYou can start GitHub Copilot’s cloud agent from an Azure Boards work item and get a draft pull request back, linked to the work item. What it needs, what it sends to GitHub, what it cannot do, and when the Azure DevOps MCP server is the better route. · 7 min read
- Cursor CLI: Using Cursor’s Agent From the TerminalCursor’s agent also runs in a terminal, as a command called agent. How to install it and sign in, the interactive and print modes, models, rules and AGENTS.md, MCP, permissions, the output formats scripts read, and running it in CI. · 7 min read
- Gemini CLI Best Practices: GEMINI.md, Extensions and Approval ModesEight habits that make Gemini CLI safer and more useful: signing in now that personal Google login has ended, a layered GEMINI.md, careful extensions, the right approval mode, sandboxing, checkpoints, trusted folders, and the GitHub Action. · 8 min read
- Goose, Block’s Open-Source Agent: What It Is and How It Uses MCPGoose is an open-source AI agent with a desktop app, a CLI and an API, now part of the Agentic AI Foundation. What it runs on, how its extensions are MCP servers, how recipes package a workflow, which permission mode it starts in, and how it compares with Claude Code. · 7 min read
- JetBrains Junie vs Claude CodeJunie is JetBrains’ own coding agent, in the IDE, in AI Chat and now as a CLI. Claude Code is Anthropic’s agent, which reaches JetBrains IDEs through a plugin. How they differ on IDE depth, models, instructions files, MCP and approvals, which one a JetBrains user should pick, and where Codex fits. · 7 min read
- AGENTS.md vs README.md vs Skills: What Goes WhereA repository now has several places to write things down: the README for people, AGENTS.md for agents, skills for procedures, tool-specific files, docs, and a task board. What belongs in each, the mistakes that make agents worse, and a worked layout. · 7 min read
- Supabase MCP: Setup, Read-Only Mode and the RisksSupabase hosts an official MCP server that signs in with OAuth and can be locked to one project, run read-only and cut down to the tool groups you need. The URL and its three switches, setup in Claude Code, Cursor and VS Code, and the production warnings Supabase gives itself. · 6 min read
- Airtable MCP: Connecting Claude and ChatGPT to a BaseAirtable runs an official MCP server, with a Claude connector and a ChatGPT app on top of it. The URL, the OAuth sign-in and the personal access token route, the scopes it asks for, what its tools can and cannot do, prompts that keep a base safe, and when a base is the wrong home for a task list. · 6 min read
- Docker MCP Toolkit: Running MCP Servers in ContainersDocker’s MCP Toolkit runs MCP servers as containers from a curated catalog, groups them into profiles and hands every client one gateway to connect to. What the Toolkit, Catalog and Gateway each are, how to turn them on, the docker mcp commands, secrets and OAuth, and what the containers do and do not protect you from. · 6 min read
- ClickUp MCP in Cursor and VS CodeConnect ClickUp’s official MCP server to Cursor, VS Code and Codex: the one URL, the browser sign-in, where each editor keeps the entry, and how to stop any of them changing ClickUp without asking. · 7 min read
- Trello MCP in Cursor and VS CodeConnect Trello’s official MCP server to Cursor, VS Code and Codex: the one-click install, the mcp.json and config.toml entries, the workspace you choose at sign-in, and how to let reads run while every change still asks. · 6 min read
- Jira and Confluence MCP in Codex CLI and Gemini CLISet up the Atlassian Rovo MCP Server in Codex CLI and Gemini CLI: the config.toml and settings.json entries, the browser sign-in, API tokens for machines without one, approval settings, and a short section on Copilot CLI. · 7 min read
- Todoist in ChatGPT: App or Developer ModeTodoist has an official plugin in ChatGPT and an official MCP server. Which route to take, how each connects, how ChatGPT asks before changing your tasks, the limits, and the same server in Codex and Gemini CLI. · 6 min read
- MCP Error Handling: Tool Errors vs Protocol ErrorsMCP has two ways to report a failure: a tool result with isError set, which the model reads, and a JSON-RPC error, which the model usually never sees. Which to use when, the error codes in the current specification, how the TypeScript and Python SDKs behave, and how to handle timeouts, cancellation and retries. · 8 min read
- Why MCP Uses JSON-RPC, and What a Message Looks LikeEvery MCP message is a JSON-RPC 2.0 request, response or notification. What the specification says about that choice, why it suits the protocol, and a real exchange, message by message: server/discover, tools/list, tools/call, an error and a notification. · 7 min read
- Logging for MCP Servers: What to Record and WhereA stdio MCP server that logs to standard output breaks its own connection, and the protocol’s built-in logging is now deprecated. Where an MCP server’s logs should go, what one record per tool call should hold, what to keep out of it, and how debug logs, audit records and traces differ. · 7 min read
- Google ADK vs OpenAI Agents SDKTwo code-first agent frameworks that run the same loop and disagree about almost everything around it. Languages, multi-agent patterns, sessions and memory, MCP, tracing, evaluation and where each one runs, with the same small agent written in both. · 8 min read
- AI Agents for Small Businesses: Ten Jobs They Can Take OnTen everyday jobs an AI agent can do in a small business, from sorting the inbox to drafting quotes and chasing invoices. For each: what the agent does, what a person still checks, and how much setting up it takes. Plus the data protection basics and how to start small. · 8 min read
- Spec-Driven Development Tools Compared: Spec Kit, Kiro, BMAD, OpenSpecFour tools that make a coding agent write things down before it writes code. What each one produces, the steps it walks you through, which agents it works with, whether it is open source, and how it copes with a codebase that already exists. · 8 min read
- Feature Request vs User Story vs Change RequestA feature request is what someone outside the team asks for, a user story is the team’s own statement of a change worth building, and a change request asks to alter something already agreed. Who writes each, when, what goes in it, and how one turns into another. · 8 min read
- Product Backlog vs Sprint BacklogThe Product Backlog is the ordered list of everything the product might need, owned by the Product Owner. The Sprint Backlog is the Developers’ plan for one Sprint. What each contains, who can change it, the commitment attached to each, and how the Increment fits in. · 7 min read
- Sprint Review vs RetrospectiveThe Sprint Review looks at the product: what was built, with the people it was built for, and what to do next. The Retrospective looks at the team: how it worked, and what to change. Purpose, attendees, inputs, outputs and timebox for each, and the mix-ups that blur them. · 7 min read
- What Is Scrumban? Kanban vs Scrumban vs ScrumScrumban is Scrum’s regular planning and review rhythm combined with Kanban’s pull system and work-in-progress limits. Where the name came from, what it keeps from each side, the usual practices, and how to tell a real Scrumban team from one that has dropped the hard parts of both. · 7 min read
- WIP Limits in Kanban: How to Set ThemA WIP limit caps how many items may be in progress at once, so a team finishes work before it starts more. What the Kanban Guide actually asks for, a first limit for teams of three, six and ten, and what to do when a column is full. · 7 min read
- Kanban Swimlanes: When They Help and When They Hide WorkSwimlanes split a kanban board into horizontal rows, by class of service, person, team or product. Five common schemes with examples, the signs that lanes are hiding work rather than showing it, and the lighter alternatives: filters, colors and categories. · 7 min read
- Cycle Time vs Lead Time: Kanban Metrics for Small TeamsCycle time runs from when work starts to when it finishes; lead time usually runs from when it was asked for. The four flow metrics the Kanban Guide requires, how to work them out from nothing but a task history, and why a percentile beats an average. · 8 min read
- Kanban Board vs To-Do List: When to SwitchA to-do list records what needs doing; a kanban board also shows where each thing is and who has it. The signs a list has outgrown itself, when a list is still the better tool, and how to move to a board in about an hour. · 7 min read
- Do You Need Story Points? Estimation for Small TeamsStory points are a relative unit for estimating work, born in Extreme Programming and never part of the Scrum Guide. What they are for, what they cost, the lighter alternatives, and when a small team is better off with them. · 7 min read
- Daily Standup: What to Say, and When to Skip ItA standup is a short daily meeting where the people doing the work agree what happens next and clear what is in the way. What the Scrum Guide actually asks for, three formats that work, what to say, and the days a board makes the meeting unnecessary. · 7 min read
- Sprint Planning Meeting: Agenda and TemplateSprint Planning answers three questions: why this Sprint is worth doing, what can be Done, and how. The Scrum Guide’s rules, a timed agenda for a two-week Sprint, the inputs and outputs, a template to copy, and the ways the meeting usually goes wrong. · 7 min read
- GitHub Projects as a Kanban Board: Setup and LimitsHow to turn GitHub Projects into a working kanban board from GitHub’s own documentation: the board layout, the Status column field, column limits, the built-in workflows, iterations and the roadmap. Then the honest limits, and when it is the right board. · 8 min read
- Jira Kanban Board Best Practices for Small TeamsSeven practices for a Jira kanban board that a small team can keep up, taken from Atlassian’s own documentation: columns mapped to statuses, column constraints, quick filters, swimlanes, the kanban backlog and the two reports worth reading. Then how to tell when Jira is more than you need. · 7 min read
- Microsoft Planner as a Kanban Board: Where It Works and Where It StopsMicrosoft Planner’s Board view is a kanban board most Microsoft 365 users already have. How the new Planner works today, how to set a plan up as a kanban board, what basic and premium plans each include, and where a dedicated kanban board does more. · 7 min read
- Claude Cowork Not Working on Windows: FixesWhen Cowork will not start, will not show up, or cannot reach your files, the cause is usually one of a short list: the plan, the app version, Windows virtualization, an admin setting, or the merge of Cowork into chat. A fix list by symptom, and how to report what is left. · 7 min read
- Claude Enterprise vs Team vs Pro: Which Plan FitsPro and Max are for one person; Team and Enterprise are for an organization. The choice between the last two turns on a few specific needs: SCIM, audit logs, custom roles, retention and how usage is billed. How to tell which plan your organization needs. · 7 min read