OpenAI is quietly moving Custom GPTs into maintenance mode and steering everyone towards its new Workspace Agents instead. For business, enterprise and education accounts, Custom GPTs are already being phased out. Individual accounts keep theirs for now, but every new feature is going into the new system, not the old one.
If you’ve spent any period of time building Custom GPTs to handle bits of your marketing, that probably landed with a bit of a thud. I get it. You did the work, you got them humming, and now the ground’s shifting under you.
But please realise that this isn’t really a story about OpenAI changing direction. That happens all the time. The far more interesting story, the one that actually changes how you should build with AI from here on, is what’s replaced Custom GPTs as the best-practice way to package up your AI know-how. And it’s good news for you.
Let me walk you through it.
What’s actually going on at OpenAI
The quick version: on 22 April 2026, OpenAI introduced Workspace Agents, its new Codex-powered system that runs inside ChatGPT, connects to tools like Slack, Salesforce and Gmail, and carries out multi-step jobs on a schedule. It’s clearly where OpenAI is putting its energy now.
Which means Custom GPTs are on the way out. At the time of writing this, business, enterprise, education and teacher accounts are being moved off them, with a conversion tool reportedly on the way. Individual users on Plus, Pro and free plans can hang onto their Custom GPTs for the moment, but make no mistake, all the new development is going into Workspace Agents.
I’m not telling you this to panic you. I’m telling you because it’s a perfect, real-world example of something I bang on about constantly… When you build your AI Workforce inside someone else’s platform, you don’t actually own the foundation. One product update, one pricing change, one announcement like this, and the setup you relied on can shift overnight.
That’s not a reason to down tools and give up on AI. It’s a reason to build smart.
So what replaces Custom GPTs? Skills.
If you’ve been following along, you’ll know I’ve already written a full plain-English guide to this, so if Skills are brand new to you, start there: What Are Skills? That post breaks down exactly what a Skill is and how it’s different from a prompt or a template.
Here’s the short version. A Skill is a saved set of instructions that teaches your AI how to do a specific task properly, the same way, every single time. It lives in a simple file called SKILL.md, it knows when to switch itself on, and it carries the full instructions for the job rather than just a clever opening line. A prompt is a conversation. A template is a saved conversation. A Skill is a trained team member who already knows exactly how you like that job done.
Skills have quickly become the grown-up replacement for Custom GPTs and Gemini Gems. Why? Because they don’t have the character limits that boxed in Custom GPTs, so you can build genuinely detailed, multi-step workflows. And crucially, they’re something you build, understand and own, rather than borrow off the internet and hope for the best.
Where you can build Skills right now
Let me start by saying that the good news about Skills is that you don’t need to code, and you’ve got several great options to be able to create them easily yourself.
You can build Skills directly in Claude (claude.ai) by creating your SKILL.md file and uploading it through Settings. Claude then uses it automatically whenever the relevant task comes up. You can also build them in Claude Code for the more technical among you, and through the Claude API if you’re building something into your own systems. And because a SKILL.md file is just markdown (plain text with a few simple symbols), you can ask Claude itself to help you write one. You bring the thinking, it handles the formatting. That’s my preferred approach.
This is exactly what I showed how to do in my How to Build Your AI Workforce MegaClass. We didn’t just talk about Skills in theory. I walked through building Skills in Claude step-by-step, the kind you can then take and use across any platform that supports the format. Which brings me to the bit that makes all of this matter.
The real story: your foundation can now outlive the platform
Here’s what didn’t make the headlines, and it probably should’ve.
In December 2025, Anthropic (the team behind Claude) published the Agent Skills format as an open standard. Not a closed, Claude-only feature. An open spec that anyone can adopt. And adopt it they have. The same Skills format is now supported across 30-plus tools, including OpenAI’s own Codex, Google’s Gemini CLI, Cursor, GitHub Copilot and Microsoft’s VS Code.
Sit with that for a second, because it’s a big deal for how you build.
It means the instructions you write once, in a single SKILL.md file, aren’t trapped inside one company’s app. They’re portable. The same file can work across competing platforms, with little or no change. So when one provider decides to deprecate a feature (hello, Custom GPTs), you’re not back to square one. You lift your Skills and carry on.
This is the difference between renting and owning. Custom GPTs and Gems were always rented. They lived on someone else’s land, and you played by their rules. Skills, built on an open standard, are far closer to something you genuinely own.
Why this is a win for Claude (and for you)
Claude is my main tool, and this move is a big part of why.
Anthropic didn’t just build a better version of Custom GPTs. They built the format and then opened it up for the whole industry to use. That’s a confident, long-game play, and it’s the opposite of locking you in. It tells you they’re backing themselves on the strength of the tool, not on trapping your work inside it.
For you, that’s the best of both worlds. You get to work in a platform that’s genuinely excellent for building and running Skills, while knowing the assets you create aren’t held hostage there. If the landscape shifts again next year (and it will), your foundation comes with you.
What to actually do about this
So here’s your nudge for today, and it’s the same one I’ll give you every time the platforms reshuffle.
Don’t keep your best work trapped inside one tool. Your best prompts, your brand voice, your positioning, your stories, the instructions that make your AI sound like you and work the way you work, all of it needs to live somewhere you control. Not buried inside a Custom GPT. Not locked in a Gem. In your own files, saved in your own folders, in a format you own.
This is what I call your Authority Knowledge Core: your voice, your expertise, your positioning and your stories, captured in a format you control. When you’ve got that saved somewhere that’s yours, it doesn’t matter which platform wins this year. You can rebuild your AI assistants in minutes, in whatever tool works best, because the foundation never left your hands.
A few practical first steps:
- Pull your instructions out of any single platform. If you’ve got brilliant set-up notes living only inside a Custom GPT, copy them into your own documents now, while you still can.
- Build your repeatable tasks as Skills. Start with the one repetitive job that drains you most, and build a Skill for it in Claude.
- Save everything externally. Keep your Skills and knowledge files in your own storage, organised and backed up, so they’re yours regardless of what any platform does next.
The tool is the amplifier. Your knowledge is the foundation. Platforms will keep changing, and the smart move was never to bet everything on one of them. It was to own the part that actually matters, so you can plug it into whatever comes next.
Custom GPTs are dying. Skills are what replace them. And the businesses that come through this shift in the strongest position will be the ones who saw it coming and built on ground they own.
If you want to see exactly how to do this, building Skills in Claude that you can take anywhere, that’s precisely what I teach in the How to Build Your AI Workforce MegaClass. Go check it out!
