You're Wasting 40% Of Your AI Time On Something Fixable

1. SUMMARY Nate Jones argues that the real barrier to productive AI use in 2026 is not model capability but the "scaffolding" around LLMs — prompts, skills, plugins, MCPs, hooks, and scripts. He maps when to use each layer: prompts for one-offs, skills for reusable processes, plugins for packaged workflows with tools and connectors, and MCPs/hooks for live data access and deterministic checks. The core claim is that non-technical users can now build this scaffolding themselves, which was not practical in 2025. 2. KEY FACTS FACT: OpenAI describes GPT 5.5 as better at "messy multi-part work" like planning, using tools, checking work, and navigating ambiguity. | EVIDENCE: "OpenAI describes Chat GPT 5.5 is better at messy multi-art work like planning and using tools and checking its work and navigating ambiguity." | CONFIDENCE: HIGH FACT: A skill is a markdown document that describes a reusable process in detail, and it works across different LLM tools. | EVIDENCE: "a skill, amazingly enough, is just a clear markdown document that describes in good detail how you do that work" and "You just write the skill and then you use it with whatever tool you end up using it with" | CONFIDENCE: HIGH FACT: A plugin can include skills, app integrations, MCP servers, hooks, assets, commands, and metadata. | EVIDENCE: "A plug-in can include skills, but it can also include app integrations, MCP servers, hooks, assets, commands, metadata." | CONFIDENCE: HIGH FACT: The speaker observed non-technical people building plugins in 2026, which he says was not possible in 2025. | EVIDENCE: "In 2025, I couldn't make this video. In 2026, I can make this video because plugins are now something where I have literally seen people who do not have coding knowledge figure out how to build a plug-in" | CONFIDENCE: HIGH FACT: Claude Design, released a few weeks before the video, is described by the speaker as "a fancy plugin with a UI for Claude for design." | EVIDENCE: "Arguably, Claude design, which dropped just a couple of weeks ago, is a fancy plugin with a UI for claude for design." | CONFIDENCE: HIGH FACT: Hooks and scripts are for deterministic workflow steps that should not be left to the model's judgment, such as formatting code, validating schemas, running tests, or checking JSON structure. | EVIDENCE: "Hooks and scripts are for the parts of your workflow where you should not rely on the model remembering to be careful" and "If the code needs formatting, run a formatter... If the schema needs validation, don't ask the model to think about it. Actually validate the schema." | CONFIDENCE: HIGH FACT: The speaker claims that most people waste significant time by over-relying on prompts for repeatable work. | EVIDENCE: "most people index really highly here. And they they don't realize that they're putting so much into the prompt that generates hours and hours and hours a week of wasted effort" | CONFIDENCE: HIGH FACT: The speaker spoke with two non-technical people working on "complicated retail scale workflows" who had built custom plugins. | EVIDENCE: "I talked to two of them this morning and they're working on like complicated retail scale workflows because they figured out what works for them." | CONFIDENCE: HIGH 3. KEY IDEAS IDEA: Agentic scaffolding should be understood as a layered hierarchy — prompts < skills < plugins — with MCPs, hooks, and scripts as composable components within plugins. | REASONING: The speaker systematically contrasts each layer by scope and reusability, showing how they nest together like Lego bricks rather than competing. | IMPLICATION: Teams need a decision framework for which layer to use, or they default to prompts and waste effort. IDEA: The "mech suit" metaphor — LLMs are the intelligence inside, but the scaffolding (skills, plugins, connectors) is what enables real work. | REASONING: The speaker repeatedly compares LLMs to Darth Vader or Transformers needing external suits to function, emphasizing that the model alone is insufficient. | IMPLICATION: Investment in AI should shift from model access to scaffolding construction. IDEA: Non-technical domain experts are now capable of building plugins, which democratizes workflow automation beyond engineering teams. | REASONING: The speaker cites multiple first-hand observations of non-coders building plugins in 2026, contrasted with 2025 when this was not feasible. | IMPLICATION: Organizations should empower domain experts to encode their knowledge directly into agentic systems rather than routing everything through engineering. IDEA: Deterministic checks (hooks/scripts) must be separated from model reasoning to ensure reliability. | REASONING: The speaker argues that certain tasks (validation, testing, formatting) should never be left to the model's judgment because models can forget or hallucinate. | IMPLICATION: Robust agent systems require explicit deterministic guardrails, not just better prompting. IDEA: Workflows should be decomposed into bounded, single-purpose plugins rather than monolithic multi-job plugins. | REASONING: The speaker uses customer service as an example where refund handling, activation, and upgrades should be separate plugins because "a workflow has one job." | IMPLICATION: Plugin design is an emerging high-value skill involving semantic boundary-drawing around work units. 4. KEY QUOTES "If you do it once it's a prompt, if you do it repeatedly it's a skill, if the workflow needs to travel or other people need to install it. If it needs tools or assets or connectors, guess what? It's a plugin." "Think of these as Lego bricks that taken together make something bigger and more useful in terms of your workflow. And by the way, in that analogy, a plugin is a bunch of Legos all built up together into a structure." "If you think of plugins as add-ons, you ask, 'Well, what can I install?' If you think of plugins as workflow packaging, you're going to ask a much sharper question because you're going to ask yourself, 'What part of my work has enough repeatable structure that the agent should be able to inherit it and use it?'" "A good agent workflow is designed so that the parts that are deterministic are correctly framed as scripts or correctly framed as hooks into services that are deterministic." "The goal is not to turn your workspace into a gigantic museum of plugins you never use. The goal is to simply understand the parts of your work that are repeated and valuable and structured enough to package and figure out the right solution so that you can do that." 5. SIGNAL POINTS The real AI leverage in 2026 is not model intelligence but scaffolding design — knowing when to use prompts vs skills vs plugins vs MCPs. Skills are just markdown files describing reusable processes and are portable across LLM tools; no coding required. Plugins are not app store add-ons but complete workflow packages that can include skills, scripts, hooks, MCP connectors, and assets. Deterministic validation (hooks/scripts) must be explicitly built into agent workflows — never trust the model to check its own work. Non-technical domain experts can now build plugins; the barrier dropped between 2025 and 2026. Workflow decomposition into bounded, single-purpose plugins is a high-value emerging skill. Most organizations waste ~40% of AI time by overusing prompts for repeatable work instead of building reusable scaffolding. 6. SOURCES MENTIONED OpenAI / GPT 5.5 — described as better at "messy multi-part work" including planning, tool use, self-checking, and ambiguity navigation. Codex (OpenAI) — referenced as having plugins/extensions and a plugin directory; used as the primary example system. Claude / Claude Code / Claude Design — mentioned as having extensions and connectors; Claude Design specifically called "a fancy plugin with a UI for Claude for design." MCP (Model Context Protocol) — described as universal data connectors, like "internet plugs" for live system access. Salesforce — used as example data source for outbound email workflows via MCP/connector. Figma — mentioned as a design tool that can be plugged into for design language and asset access. Slack, GitHub — mentioned as data sources that plugins can connect to. Substack (natesnewsletter.substack.com) — the speaker's own publication where deeper guides, starter plugins, and workflow audits are published. 7. VERDICT This video carries moderate signal for AI practitioners but high signal for operators and team leads struggling to operationalize LLMs. The core contribution is a clear taxonomy and decision framework for agentic scaffolding — something engineering-oriented coverage often assumes readers already understand. The speaker's claim that non-technical users can now build plugins is notable if true, though it rests on anecdotal observation rather than systematic evidence. The video contains no demonstrations, no code, no data, and no product announcements; it is entirely conceptual. For someone tracking AI infrastructure and agent tooling, the "Lego bricks" mental model and the explicit hierarchy (prompt → skill → plugin → MCP/hook/script) are useful structuring devices. However, the ~40% waste claim is unsupported by data, and the repeated self-promotion of the Substack dilutes signal density. Worth watching for the framework, but not for technical depth or novel information. --- COUNT: Facts: 8 | Assumptions: 2 (the 40% waste figure; the claim that non-technical plugin building is now broadly accessible) | Demonstrations: 0 SIGNAL DENSITY: 55/100 — The conceptual framework is solid and well-articulated, but the lack of evidence for quantitative claims, absence of demonstrations, and embedded marketing for the speaker's paid content reduce the ratio of high-signal to noise.

1. 1

   The real AI leverage in 2026 is not model intelligence but scaffolding design — knowing when to use prompts vs skills vs plugins vs MCPs.

2. 2

   Skills are just markdown files describing reusable processes and are portable across LLM tools; no coding required.

3. 3

   Plugins are not app store add-ons but complete workflow packages that can include skills, scripts, hooks, MCP connectors, and assets.

4. 4

   Deterministic validation (hooks/scripts) must be explicitly built into agent workflows — never trust the model to check its own work.

5. 5

   Non-technical domain experts can now build plugins; the barrier dropped between 2025 and 2026.

6. 6

   Workflow decomposition into bounded, single-purpose plugins is a high-value emerging skill.

7. 7

   Most organizations waste ~40% of AI time by overusing prompts for repeatable work instead of building reusable scaffolding.

8. 8

   6. SOURCES MENTIONED