Isenberg defines skills, sub-agents, and MCPs in agent architecture
Greg Isenberg · OpenAI Releases ChatGPT AI Agent Skills · Dec 22, 2025 · at 2:20
Greg Isenberg breaks down the functional architectural differences between agent skills, sub-agents, and Model Context Protocol (MCP) integrations.
0:00 / 0:55exact quote · 55.2s
▶ Watch the full episode on YouTube → 720p mp4 · rendered on demand · StarZero watermark
“A skill is like a written guide that tells, you know, Claude or You know, Codex or ChatGPT exactly how to do a specific task. For example, analyze spreadsheets this way, or write emails in this voice, or design you know, design the interface like this. So you can basically call and reuse the same skill again and it just, it allows for more consistent output. So what is a sub-agent? Well, a sub-agent is making a few extra copies Of, you know, the LLM that you're using, you know, Claude or Codex or ChatGPT, each with its own job, such as you review the code, you write tests, you update docs. The main LLM can hand out pieces of a big task to help these workers. So work happens in parallel and stays organized. And an MCP is just a universal power plug that lets the LLM access other tools.”
quote is from the automated transcript, cleaned for reading: filler sounds and stutters are removed, nothing is rephrased. names can be misheard (the analysis reads context, assessments check outside sources). how →
More from Greg Isenberg
Isenberg: AI Apps Owning Workflows and Network Effects Will Be Highly Valuable
“I think that you're discounting the opportunity that there is to actually create these, as you call it, flash in the pan apps to actually get data, to actually own the workflow, to actually create network effects. And I think that that those companies Provided…”
Isenberg: Top AI wrappers will survive OpenAI updates within 18 months
“There's, of course, going to be a bunch of startups that get hit by OpenAI and the foundational models in 12 to 18 months, but I think that, you know, the great AI wrappers are going to build valuable companies.”
Isenberg: Current LLMs will beat human beings on 95% of tasks
“Maybe this is obvious, but artificial intelligence is smarter than human beings. We're at a stage where human beings versus the current LLNs are gonna lose on 95% of tasks.”
Solo founders can build $50K MRR businesses using AI tools like Cursor
“I do believe that we live in a world that it's completely possible to use things like cursor to use some of these tools to develop something that you can make 50,000 dollars a month for.”