Blog / DevDay, Custom GPTs, & AI Best Practices

DevDay, Custom GPTs, & AI Best Practices

AUTHOR: Jen Taylor

Oct 07, 2026

4 Min Read

Listen

In this edition: OpenAI’s 2026 DevDay takeaways, Custom GPTs, and the importance of prompt context and human agency in day to day AI use.

This is Q&AI, our blog series aimed at keeping you in the know on updates in the rapidly evolving world of AI. Sometimes, these will be quick updates on new developments in the field. Sometimes, they’ll be tips on tactics, features, or functionality. If you haven’t met me yet, hi: I’m Jen Taylor, Capacity’s Vice President, Strategic Growth and AI Activation.

Q: Is it true that OpenAI is deprecating Custom GPTs? What does this mean for my workflows?

A: Yes, ChatGPT is deprecating Custom GPTs. OpenAI’s built-in migration turns a GPT’s instructions into a Skill inside a Plugin, knowledge files become reference files, and connected apps can come along within the Plugin. The work you’ve put into building your GPTs still has value.

However, the built-in Plugin migration may not be the best fit for every workflow. A Skill contains the instructions and reference materials for a particular task. A Plugin is a container that can bundle multiple Skills and connected apps.

For a focused workflow, such as a copyeditor or report formatter, my preference is to recreate it as a standalone Skill that lives alongside your other Skills. A Plugin makes more sense when we want to package related workflows together.

Sharing and access also matter. In our Team/Business account, I’ve encountered a permission restriction when trying to share a personally uploaded Plugin. Sharing standalone Skills between team members has been straightforward in our experience.

This is why I recommend always documenting your workflows and instructions. At Capacity, having our Custom GPT instructions saved has made it easy to turn them into Skills. We still need to test each replacement, but we’re building on work we’ve already done.

Q: How do I ensure that AI doesn’t turn my work into generic output?

A: It truly is all about context. In theory, we know about the importance of context in prompting, but I’ve really been seeing that lesson reinforced lately.

Last month at HubSpot’s UNBOUND conference, this was a big theme. As everyone gets access to increasingly capable AI, your unique institutional knowledge and context are what help differentiate the output and make it uniquely yours. That context can be what you put directly into a prompt, the reference materials you provide, your customer knowledge, brand, strategy, history, previous decisions, expertise, etc.

None of that is new. It’s really a reminder of something we’ve been talking about all along. The better the underlying AI gets, the more important what you bring to it becomes.

I also caught up on some Ethan Mollick posts last month. If you’re not familiar with him, he’s a Wharton professor who has become one of the most prominent voices studying and writing about how people actually use AI at work. His latest piece gets at an adjacent idea that I think is important: human agency.

His argument is essentially that, as AI agents get more capable of doing work independently, our goal shouldn’t necessarily be removing humans from the process entirely. He talks about four reasons AI should proactively bring humans back in: approval, expertise, diversity of ideas/perspectives, and because some decisions are actually interesting and valuable for humans to make.

That really resonated with me. The reminder for me this week is that better AI doesn’t make our role less important. We still need to give it the context that makes the work distinctive and keep human judgment intentionally in the places where it matters.

Q: What were the major takeaways from OpenAI’s DevDay 2026?

A: OpenAI covered a lot (here’s the keynote, if anyone wants to watch), including new models and changes to how people can build and distribute capabilities. A few announcements that felt particularly relevant to the future of the platform:

Dots: Always-on agents that can take ongoing responsibility, continue working between conversations, and bring results or decisions back to you. This includes personal dots as well as specialist dots designed for particular types of work. OpenAI is beginning with a limited rollout that includes eligible Enterprise and Pro accounts.
Space: A place to bring together pages, files, and shared work so people and ChatGPT can collaborate in one space.
Sign in with ChatGPT: This will allow people to sign into third-party products using their ChatGPT accounts and, when eligible, use their existing ChatGPT plan and usage within those products. OpenAI is starting with a group of partners.
Plugin Extensions: Plugins can now function more like applications inside ChatGPT, with elements such as sidebar experiences, conversation panels, file viewers, editors, and forms—not just tools ChatGPT calls behind the scenes.