You know that drawer in your kitchen? The one stuffed with batteries, takeout menus, and a phone charger from 2019? That's what most people's AI setup looks like right now.
Both Claude and ChatGPT have shipped lots of powerful customization features over the past few months, and almost nobody outside of the tech bubble in Silicon Valley and X.com have organized theirs yet (although you could argue they've mostly moved on to OpenClaw). Meanwhile, Anthropic just launched inline interactive visuals today, letting Claude build charts and diagrams right inside your conversation. It's one more piece of a bigger puzzle: AI tools are becoming platforms you configure, not just chatbots you prompt.
The good news? Once you understand how these pieces fit together, you can turn a generic AI assistant into something that actually knows your workflow. Here's the practical guide.
First up, the TL;DR
Your AI chatbot now has a settings menu. Here's how to actually use it.
If you've opened Claude or ChatGPT lately and felt like there were twelve new buttons you didn't ask for, you're not alone. Both platforms are racing to become customizable AI workspaces, and the terminology can feel like alphabet soup.
Here's the cheat sheet. Claude's new Customize tab (Settings → Customize) groups four things in one place:
- Skills teach Claude how to do something. They're instruction files that Claude loads when your task matches the skill's description. Think: "when I ask about code review, follow this process."
- Connectors are the plumbing. They link Claude to external tools like Google Drive, Slack, or Salesforce via MCP (Model Context Protocol), Anthropic's open standard for AI-to-tool communication (video).
- Plugins bundle skills, connectors, and slash commands into one installable package. A sales plugin might connect to your CRM, include your prospecting playbook, and give you a
/research-leadcommand. Anthropic open-sourced 11 starter plugins in January. - The Customize tab is the dashboard where you browse, install, and manage all of the above. Enterprise admins can build private plugin marketplaces.
On the OpenAI side, Codex has evolved from a coding agent into a full command center. Codex Skills work similarly to Claude's: bundled instructions and scripts that extend what the agent can do. Automations schedule Codex to run recurring tasks in the background, like daily issue triage or CI failure summaries. And the new ChatGPT Apps, powered by the Apps SDK, let developers build interactive experiences (maps, playlists, slide decks) directly inside your ChatGPT conversation. Spotify, Zillow, Canva, and Coursera are already live.
The real unlock? Start small. Connect one tool you use daily. Install one plugin that matches your job function. Build one automation for a task you repeat weekly. Ask your current chat window to help you build your own sill, "using your skill creator skill." These platforms reward incremental setup over a weekend overhaul.
Oh, and that new inline visual feature from Claude? It's on by default for all plans. Just ask Claude to "visualize" or "diagram" something, and it'll build an interactive chart right in the chat. It's available today.
Both companies clearly got the same memo: the chatbot era is over; the platform era starts now.
The Customization Arms Race
A year ago, asking ChatGPT or Claude a question looked roughly the same. You typed. It answered. Maybe you uploaded a PDF. That was the whole interaction model.
Now both platforms look more like operating systems than chatbots. Anthropic's Cowork launched in research preview on January 30, 2026, with plugin support that immediately rattled software stocks (Thomson Reuters had its worst trading day on record). OpenAI's Codex desktop app arrived in February 2026 with multi-agent workflows, skills, and automations. And ChatGPT Apps, announced at DevDay 2025, turned the chatbot into something closer to an app store.
The pattern is identical on both sides: give users control over what the AI knows, what tools it can reach, and what tasks it can automate.
Claude's Customize Tab, Explained
If you open Claude Desktop and go to Settings → Customize, you'll see everything organized in one place. Here's what each piece does and when you'd reach for it.
Skills: What Claude Knows How to Do
A skill is a folder with a SKILL.md file containing instructions, optional scripts, and reference materials. Claude uses progressive disclosure to stay efficient; it reads just the name and description of each skill until one matches your task, then loads the full instructions.
You can create skills for anything repeatable: your team's code review checklist, your company's writing style guide, a deployment workflow, a research methodology. The key distinction is that skills are procedural knowledge. They answer the question "how should this be done?"
You can install skills from Anthropic's library, upload your own (.skill, .zip, or .md files), or build them from scratch in the Customize tab.
Good skill ideas for non-developers:
- A weekly report generator that follows your company's format
- A meeting notes template that structures action items consistently
- An email drafting skill tuned to your voice and sign-off preferences
- A competitive analysis framework your team uses regularly
Connectors: What Claude Can Reach
Connectors use MCP (Model Context Protocol), the open standard Anthropic introduced in late 2024, to link Claude to external services. When you browse connectors in Settings, you'll see options for Google Workspace (Calendar, Drive, Gmail), Slack, Figma, Canva, Salesforce, Docusign, Apollo, FactSet, and many more.
The connector handles authentication (usually OAuth) and the data pipeline. Once connected, Claude can pull information from that service and act on it within your workflow. A connected Google Drive, for instance, means Claude can search your docs directly rather than waiting for you to paste content in.
The key thing to understand: connectors provide access to data. Skills provide instructions on what to do with it. You usually want both. A "quarterly report" skill without a Google Sheets connector would require you to paste all the data manually.
Plugins: The Finished Product
Plugins bundle skills, connectors, slash commands, and sub-agents into one installable unit. If skills are the playbook and connectors are the plumbing, plugins are the fully furnished apartment.
Anthropic's 11 open-source starter plugins cover marketing, legal, finance, data analysis, biology, and general utilities. Each is customizable; click "Customize" on any installed plugin and Claude walks you through adjusting the skills, commands, and connectors to match your stack.
Enterprise admins on Team or Enterprise plans can now build private plugin marketplaces to distribute vetted plugins across their org, with per-user provisioning and auto-install capabilities.
When to use what:
- Skill → You need Claude to follow a specific process ("review code using our standards")
- Connector → You need Claude to access an external tool ("pull data from Salesforce")
- Plugin → You want to package both for easy distribution ("install the sales toolkit and everything works")
Today's New Feature: Inline Visuals
Announced today, Claude can now create interactive charts, diagrams, and visualizations directly in the chat. These render inline (not in a side panel), are interactive (click around, adjust inputs), and change as the conversation evolves.
Unlike artifacts (which are designed to be downloaded or shared as polished files), these inline visuals exist to aid understanding in the moment. Ask Claude to visualize compound interest, explain a system architecture, or diagram a workflow, and it'll build something you can interact with on the spot.
This is on by default for all plan types. You can prompt it directly ("draw this as a diagram," "visualize how this changes over time") or Claude will proactively create visuals when they'd help explain something.
It's a small feature with big implications. Combined with skills, connectors, and plugins, it makes Claude less of a text box and more of a workspace where information is both processed and presented.
OpenAI's Customization Stack
OpenAI has built a parallel (and in some ways more ambitious) customization ecosystem across Codex and ChatGPT.
Codex Skills and Automations
Codex Skills are conceptually identical to Claude's. They're directories with a SKILL.md file plus optional scripts and resources. Codex reads each skill's metadata first, then loads full instructions only when it decides to use one. You can invoke skills explicitly ($skill-name) or let Codex match them automatically based on your task.
Where Codex goes further is Automations. You can schedule Codex to run tasks in the background on a recurring basis. OpenAI's own team uses automations for daily issue triage, CI failure summaries, release briefs, and even self-improving skill generation (an automation that scans recent sessions and updates skills that aren't performing well).
Setting up an automation: pair your instructions with optional skills, pick a schedule, and let it run. Results land in a review queue so you can jump in when needed. Before scheduling, OpenAI recommends testing the prompt manually in a regular thread first to make sure it behaves.
Good automation ideas:
- Daily standup summaries pulled from your issue tracker
- Weekly code review digest across your team's PRs
- Recurring dependency updates with automated test runs
- Morning news briefs on topics relevant to your project
ChatGPT Apps: The App Store Inside the Chat
ChatGPT Apps are the most consumer-visible piece of OpenAI's platform play. Built on the Apps SDK (which itself is built on MCP), these are third-party interactive experiences that live inside your ChatGPT conversation.
When you message "Spotify, make a playlist for my party," the Spotify app activates within the chat, renders an interactive playlist UI, and lets you customize it through conversation. Canva can turn an outline into a slide deck. Zillow can show home listings on a map. Coursera can stream a video while you ask ChatGPT questions about the lesson.
For developers, the Apps SDK makes it possible to build MCP-powered apps with custom UIs that reach ChatGPT's 800M+ user base. App submissions are now open for review and publication.
For regular users, you can browse and connect apps through Settings → Apps & Connectors. They activate in conversation automatically when relevant, or you can call them by name.
The Practical Playbook: Where to Start
Here's the simplest way to get value from all of this without spending a weekend on configuration:
- On Claude:
- Go to Settings → Customize. Browse the connectors. Connect one tool you use every day (Google Calendar, Slack, or your CRM are good first picks).
- Install one plugin that matches your job function. Marketing, legal, finance, and data analysis are all covered by Anthropic's starter plugins.
- Try the new inline visuals: ask Claude to "visualize" or "diagram" something you're working on.
- On ChatGPT / Codex:
- Open Settings → Apps & Connectors. Connect a partner app you already use (Spotify, Canva, Figma).
- If you use Codex, create one skill for a process you repeat weekly. Then set up one automation to run it on a schedule.
- Browse the ChatGPT app directory to see what's available.
- On both platforms:
- The best automation ideas come from tasks that make you think "I do this the same way every time." If the process is repeatable, it's a candidate.
- Don't try to customize everything at once. One skill, one connector, one automation. Build from there.
What This Means for You
The underlying message from both Anthropic and OpenAI is the same: the era of one-size-fits-all AI is ending. These platforms are becoming as configurable as your phone, and the people who invest 15 minutes setting up their stack will get dramatically more value than those who keep using the default settings.
The terminology is different (Anthropic says "connectors," OpenAI says "apps"), but the architecture is converging on the same open standard: MCP. That means skills, plugins, and integrations are increasingly portable between platforms, and the ecosystem is growing fast.
If there's one thing to do today, it's this: open the settings menu on whichever AI tool you use most, and connect one thing. The customization era has arrived, and it rewards the curious.
