The Startup Ideas Podcast

Episode: Claude Skills: The NEW Way to Build AI Agents (Live Tutorial)
Host: Greg Isenberg
Guest: Amir
Date: October 27, 2025

Episode Overview

In this engaging episode, Greg Isenberg and guest Amir deep-dive into Anthropic's Claude Skills—one of the most exciting recent advancements in building AI digital employees. Listeners will learn what Claude Skills are, how they differ from Projects and Sub Agents in Claude, and why they're a game-changer for automating specialized workflows. Amir gives hands-on demos, including how to craft custom Skills and live-builds a skill that turns tweets into long-form newsletter content. The conversation is packed with practical advice, memorable analogies, and concrete examples for startups, marketers, and anyone interested in leveraging AI agents to boost productivity.

Key Discussion Points & Insights

1. What Are Claude Skills and Why Do They Matter?

[00:32–05:40]

• Definition:

  • Claude Skills are automated workflows and specialized task-automation modules that can be applied globally at a project or individual level in Claude.
  • They provide laser-focused, repeatable instructions for a set of tasks, drawing in just the right amount of context and running scripts/code where needed.

• Distinguishing from Projects and Sub Agents:

  • Projects: Workspaces with custom instructions, context files, and tools; great for collaboration and repeatable task automation, but require continual context file updates.
  • Sub Agents: Used within Claude code to break complex workflows into small, specialized agent tasks. Useful for dividing frontend and backend work, each within their isolated context.
  • Skills: Extend or augment project abilities with highly targeted, scriptable tasks. Context is only loaded when directly relevant, minimizing unnecessary data and hallucination.

• Memorable Analogy:

  • "AI is essentially your coworker or someone that...you want to train, you want to build the guidelines...build the limit, the constraints around it and guidelines on how you want it to respond to you." – Amir [05:20]

• Importance of Context:

  • Too much or too little context can harm outcome quality (referring to "context rot"). The key is providing just enough—the Goldilocks principle for prompt engineering.

2. How Claude Skills Solve Real AI Workflow Problems

[07:57–10:35]

• Custom Skills:
  • Created as markdown files describing tasks; can reference external files (like brand guidelines) and run custom scripts (e.g., Python) to guarantee accuracy and determinism.
  • Unlike projects that leave output to the model’s interpretation (non-deterministic), Skills ensure the process is both specific and repeatable.

"If you wanted to have a very set of strict guidelines on how it should actually run and analyze the data, then you can create that within the skill itself." – Amir [10:18]

• Practical Value:
  • Solves frustrations with inaccurate or inconsistent AI output due to vague instructions or lack of precise context.
  • Example: Marketing teams can get robust, reliable campaign analyses by scripting explicit calculation logic in a Skill.

3. Live Tutorials: Applying and Building Claude Skills

[11:10–32:00]

a. Artifact Builder Skill (Web App for Marketers)

[11:10–13:26]

• Amir demonstrates using a preloaded Skill in Claude to create a UTM link generator—literally a shareable web app for marketing teams.

"It's literally a web app. But what's interesting is we're now creating a set of specific instructions and skills." – Amir [12:56]

b. A/B Testing Ideas Skill

[17:23–20:09]

• By adding a Skill designed to generate A/B test ideas for websites, Claude can ingest a URL, scrape it with tools like Fire Crawl, and suggest experiment frameworks for conversion optimization.
• Example used: humblelytics.com—Claude suggests reordering sections on the homepage, providing practical, actionable insights.

"It was actually a really good one. It told me to...test shifting the case study...to immediate social proof. I was like, damn, that's a good idea." – Amir [19:25]

c. Marketing Insight Agent (Data Analysis Skill)

[20:09–23:37]

• Amir uploads a traffic analytics CSV file; Claude Skill provides clear insights: total spend, revenue, profit, best-performing channels, etc.
• Strongly contrasts with the more error-prone, hallucination-prone outputs from broader project-based prompting.

"If I had done this through a project...without running scripts, it would have probably hallucinated some of the data." – Amir [21:56]

4. Live Build: Tweet-to-Newsletter Ghostwriting Skill

[23:51–30:30]

• Greg asks if it's possible to create a Skill that expands his viral tweets into newsletter content, matching his tone and style.
• Amir walks through:
  • Exporting sample tweets and newsletter posts as reference files.
  • Using the Skill Creator to guide Claude in producing a zip containing all needed references.
  • Uploading and testing—Claude turns a tweet into newsletter copy in one shot.

"Honestly, this is fire...As a starting point, it's not bad." – Greg [28:55] & [29:15]

• Discussion on future possibilities:
  • Expanding the Skill with more context for improved personalization.
  • Adding automated visuals or integrating more data sources.

5. Wider Implications: Adoption, AI Fluency, and the Real Barriers

[30:31–32:43]

• Amir references new data showing a dip in AI tool subscriptions, especially in enterprise.
• He attributes a lack of sustained productivity to insufficient education on prompting and context-setting, not the technology itself.

"There isn't enough AI fluency and education around how to actually do prompting...the reality is like, no, no, you got to give the right amount of context and do some prompt structure." – Amir [31:35]

• Anthropic’s approach—building with intention and listening to feedback—is noted as industry-leading.

Notable Quotes & Memorable Moments

• "Think about AI as a teammate, especially a junior teammate, that you have to really give it guardrails and really give it context because that's how you would, you know, if you hired someone junior..." — Greg [13:54]
• "The prompter, the problem is them." — Amir, humorously blaming incomplete or vague prompts for user frustration [15:31]
• "We talk endlessly about power market fit. We should talk more about pricing market fit...that's actually a banger." — Greg/Amir, on leveraging AI for content and business insights [29:44]
• "Dude, this is crazy..." — Greg, reacting to the ease and effectiveness of the live Skill-building demo [30:06]

Key Timestamps for Segments

| Time | Segment Description | |-------------|-------------------------------------------------------| | 00:32–05:40 | Intro to Claude Skills, Projects, Sub Agents | | 07:57–10:35 | Solving Real Problems with Custom Skills | | 11:10–13:26 | Live: Artifact Builder Skill for Marketers | | 17:23–20:09 | Live: A/B Testing Ideas Skill | | 20:09–23:37 | Live: Marketing Insight Analysis Skill | | 23:51–30:30 | Live Skill Build: Tweet to Newsletter | | 30:31–32:43 | Discussion: AI Adoption, Education, and Fluency |

Takeaways & Action Items

• Claude Skills offer an unprecedented level of control, repeatability, and specificity in automating complex or specialized tasks with AI.
• The real secret to leveraging AI productively: Thoughtful, context-aware prompting and skill curation—treating your AI like a junior teammate you have to explicitly train.
• Massive opportunities exist to package and sell Skills (and Plugins) as products, especially for repetitive business processes.
• Wider AI adoption in enterprises likely hinges on boosting workforce AI fluency and proper prompt engineering.

Follow Amir on X: @amirmxt
Discover more startup ideas from Greg: gregisenberg.com