Nate B. Jones argues that useful agents are not just smarter models. They are models placed inside a working scaffold: skills, plugins, connectors, hooks, and scripts that turn vague capability into repeatable work.
From prompts to reusable systems
A prompt is still the right tool for a one-off request. But when the same task comes back again and again, prompting becomes a fragile interface. People keep rewriting instructions, pasting context, and manually checking the result. A skill captures the reusable method: how a team reviews pull requests, writes marketing documents, handles customer support, or structures outbound emails.
A plugin is a larger package. It can include skills, live-data connectors, validation scripts, hooks, assets, commands, and metadata. It turns a workflow into something installable and shareable instead of something rebuilt by hand every time.
Connectors are not the workflow
Jones separates MCP servers and app connectors from plugins. Connectors give the agent access to the systems where work actually lives: CRM, Slack, Figma, GitHub, documents, dashboards. A plugin can contain those connections, but the plugin is the broader workflow that decides what to do with the data.
Deterministic checks should not be left to the model
Hooks and scripts belong wherever the workflow needs certainty: formatting code, validating a schema, checking JSON, running tests, or forcing a review before the agent stops. A reliable agent does not merely claim it checked something. It runs the check.
The valuable skill is drawing workflow boundaries
The point is not to turn everything into a plugin. The valuable judgment is deciding what should remain a prompt, what should become a skill, what needs a connector, what must be verified by a script, and what should stay human. That makes domain experts central: they know which sources matter, which steps are often missed, and what good output looks like.
In 2026, the practical leverage in agentic AI is moving toward packaging work. Teams that turn important routines into reusable systems will waste less time prompting and gain more reliable automation.
Source
- Chaîne: AI News & Strategy Daily | Nate B Jones
- Vidéo source: https://www.youtube.com/watch?v=647pSnX5H_Y
No comments yet