StarZero All shows Try the podcast clipper →

Oct 27, 2025 · 33m · startup-ideas

Claude Skills Built Me an AI Agent Army (They Run Everything Now)

Greg Isenberg · 5m spoken
0:00 / 0:00
▶ Watch on YouTube →

gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions

Host Greg Isenberg and guest Amir demonstrate how to construct reliable 'digital employees' using Claude Skills, contrasting them with standard Projects and sub-agents. Through technical explanations and live demonstrations, they show how dynamic context management and deterministic Python scripts automate marketing analysis, web app prototyping, and content creation.

How this conversation actually went

Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. Greg holds 18.5% of the talking time here. How this is scored →

Greg as informed peer 2.3 Guest teaching 5.7 Guest disagreement 0.1 Greg pushing back 0.3
05100:0010:0020:0030:001:04–3:33 · Greg as informed peer 0/10 Understanding Claude Projects and Sub-Agents Amir delivers an uninterrupted breakdown distinguishing Claude Projects, Sub-agents, and Skills. Greg listens silently without speaking in this segment.3:33–6:26 · Greg as informed peer 0/10 The Power of Claude Skills and Context Management Amir continues his monologue, explaining how skills load context selectively and referencing research on context rot and performance degradation.6:27–11:08 · Greg as informed peer 2/10 Sponsor Break: IdeaBrowser.com Startup Trends Greg assumes more context prevents hallucinations, prompting Amir to clarify that excessive context degrades performance. Amir then explains how deterministic scripts prevent analytics errors.11:08–17:40 · Greg as informed peer 4/10 Demo: Building Web Artifacts and UTM Generators Amir demos creating a UTM generator web artifact. Greg adds domain context by analogizing AI guardrails to onboarding a junior employee.17:41–20:30 · Greg as informed peer 3/10 Demo: A/B Testing Idea Generator Skill Amir demonstrates an A/B testing skill using Firecrawl web scraping, and Greg suggests extending it into recurring monthly automated reporting.20:31–23:48 · Greg as informed peer 3/10 Demo: Accurate Marketing Data Analytics with Scripts Amir shows how custom Python scripts in skills calculate precise traffic analytics rather than relying on LLM guesses, with Greg verifying output fidelity.23:48–30:29 · Greg as informed peer 4/10 Live Build: Creating a Tweet-to-Newsletter Skill The pair execute a live build converting Greg's tweets into a newsletter format, with Greg evaluating the output quality and suggesting hook refinements.1:04–3:33 · Guest teaching 7/10 Understanding Claude Projects and Sub-Agents Amir delivers an uninterrupted breakdown distinguishing Claude Projects, Sub-agents, and Skills. Greg listens silently without speaking in this segment.3:33–6:26 · Guest teaching 7/10 The Power of Claude Skills and Context Management Amir continues his monologue, explaining how skills load context selectively and referencing research on context rot and performance degradation.6:27–11:08 · Guest teaching 6/10 Sponsor Break: IdeaBrowser.com Startup Trends Greg assumes more context prevents hallucinations, prompting Amir to clarify that excessive context degrades performance. Amir then explains how deterministic scripts prevent analytics errors.11:08–17:40 · Guest teaching 5/10 Demo: Building Web Artifacts and UTM Generators Amir demos creating a UTM generator web artifact. Greg adds domain context by analogizing AI guardrails to onboarding a junior employee.17:41–20:30 · Guest teaching 5/10 Demo: A/B Testing Idea Generator Skill Amir demonstrates an A/B testing skill using Firecrawl web scraping, and Greg suggests extending it into recurring monthly automated reporting.20:31–23:48 · Guest teaching 6/10 Demo: Accurate Marketing Data Analytics with Scripts Amir shows how custom Python scripts in skills calculate precise traffic analytics rather than relying on LLM guesses, with Greg verifying output fidelity.23:48–30:29 · Guest teaching 4/10 Live Build: Creating a Tweet-to-Newsletter Skill The pair execute a live build converting Greg's tweets into a newsletter format, with Greg evaluating the output quality and suggesting hook refinements.1:04–3:33 · Guest disagreement 0/10 Understanding Claude Projects and Sub-Agents Amir delivers an uninterrupted breakdown distinguishing Claude Projects, Sub-agents, and Skills. Greg listens silently without speaking in this segment.3:33–6:26 · Guest disagreement 0/10 The Power of Claude Skills and Context Management Amir continues his monologue, explaining how skills load context selectively and referencing research on context rot and performance degradation.6:27–11:08 · Guest disagreement 1/10 Sponsor Break: IdeaBrowser.com Startup Trends Greg assumes more context prevents hallucinations, prompting Amir to clarify that excessive context degrades performance. Amir then explains how deterministic scripts prevent analytics errors.11:08–17:40 · Guest disagreement 0/10 Demo: Building Web Artifacts and UTM Generators Amir demos creating a UTM generator web artifact. Greg adds domain context by analogizing AI guardrails to onboarding a junior employee.17:41–20:30 · Guest disagreement 0/10 Demo: A/B Testing Idea Generator Skill Amir demonstrates an A/B testing skill using Firecrawl web scraping, and Greg suggests extending it into recurring monthly automated reporting.20:31–23:48 · Guest disagreement 0/10 Demo: Accurate Marketing Data Analytics with Scripts Amir shows how custom Python scripts in skills calculate precise traffic analytics rather than relying on LLM guesses, with Greg verifying output fidelity.23:48–30:29 · Guest disagreement 0/10 Live Build: Creating a Tweet-to-Newsletter Skill The pair execute a live build converting Greg's tweets into a newsletter format, with Greg evaluating the output quality and suggesting hook refinements.1:04–3:33 · Greg pushing back 0/10 Understanding Claude Projects and Sub-Agents Amir delivers an uninterrupted breakdown distinguishing Claude Projects, Sub-agents, and Skills. Greg listens silently without speaking in this segment.3:33–6:26 · Greg pushing back 0/10 The Power of Claude Skills and Context Management Amir continues his monologue, explaining how skills load context selectively and referencing research on context rot and performance degradation.6:27–11:08 · Greg pushing back 1/10 Sponsor Break: IdeaBrowser.com Startup Trends Greg assumes more context prevents hallucinations, prompting Amir to clarify that excessive context degrades performance. Amir then explains how deterministic scripts prevent analytics errors.11:08–17:40 · Greg pushing back 1/10 Demo: Building Web Artifacts and UTM Generators Amir demos creating a UTM generator web artifact. Greg adds domain context by analogizing AI guardrails to onboarding a junior employee.17:41–20:30 · Greg pushing back 0/10 Demo: A/B Testing Idea Generator Skill Amir demonstrates an A/B testing skill using Firecrawl web scraping, and Greg suggests extending it into recurring monthly automated reporting.20:31–23:48 · Greg pushing back 0/10 Demo: Accurate Marketing Data Analytics with Scripts Amir shows how custom Python scripts in skills calculate precise traffic analytics rather than relying on LLM guesses, with Greg verifying output fidelity.23:48–30:29 · Greg pushing back 0/10 Live Build: Creating a Tweet-to-Newsletter Skill The pair execute a live build converting Greg's tweets into a newsletter format, with Greg evaluating the output quality and suggesting hook refinements.

speaking balance: gold is Greg, purple is the guest (3 minute bins)

0:00 · Greg 20.9% · guest 79.1%0:00 · Greg 20.9% · guest 79.1%3:00 · Greg 0% · guest 100%3:00 · Greg 0% · guest 100%6:00 · Greg 41.7% · guest 58.3%6:00 · Greg 41.7% · guest 58.3%9:00 · Greg 0.2% · guest 99.8%9:00 · Greg 0.2% · guest 99.8%12:00 · Greg 29% · guest 71%12:00 · Greg 29% · guest 71%15:00 · Greg 4.7% · guest 95.3%15:00 · Greg 4.7% · guest 95.3%18:00 · Greg 3.8% · guest 96.2%18:00 · Greg 3.8% · guest 96.2%21:00 · Greg 20.1% · guest 79.9%21:00 · Greg 20.1% · guest 79.9%24:00 · Greg 48.6% · guest 51.4%24:00 · Greg 48.6% · guest 51.4%27:00 · Greg 27% · guest 73%27:00 · Greg 27% · guest 73%30:00 · Greg 9.9% · guest 90.1%30:00 · Greg 9.9% · guest 90.1%33:00 · Greg 34.3% · guest 65.7%33:00 · Greg 34.3% · guest 65.7%
Sharpest disagreement ▶ 7:36 Polite Correction on Context Windows

Amir immediately counters Greg's premise that more context prevents hallucinations, explaining that overloading models causes context rot and errors.

 Hardest push from Greg ▶ 16:50 Distinguishing Plugins from Skills

Greg stops Amir's pitch about monetizing skills to explicitly challenge him to clarify the technical distinction between Anthropic plugins and skills.

 Biggest teaching moment ▶ 9:30 Why RevOps Projects Fail Without Deterministic Scripts

Amir explains to Greg how standard Claude projects produce non-deterministic marketing math, and why embedding explicit Python scripts within skills solves the issue.

 Greg holds their own ▶ 13:54 Framework for Drip-Feeding Context to AI

Greg demonstrates operational expertise by synthesizing Amir's technical demo into a concrete operational framework for managing AI like a junior hire.
the scores for every segment, with the reasoning behind each
ChapterTopicGreg as informed peerGuest teachingGuest disagreementGreg pushing backWhy
Understanding Claude Projects and Sub-Agents 0700 Amir delivers an uninterrupted breakdown distinguishing Claude Projects, Sub-agents, and Skills. Greg listens silently without speaking in this segment.
The Power of Claude Skills and Context Management 0700 Amir continues his monologue, explaining how skills load context selectively and referencing research on context rot and performance degradation.
Sponsor Break: IdeaBrowser.com Startup Trends 2611 Greg assumes more context prevents hallucinations, prompting Amir to clarify that excessive context degrades performance. Amir then explains how deterministic scripts prevent analytics errors.
Demo: Building Web Artifacts and UTM Generators 4501 Amir demos creating a UTM generator web artifact. Greg adds domain context by analogizing AI guardrails to onboarding a junior employee.
Demo: A/B Testing Idea Generator Skill 3500 Amir demonstrates an A/B testing skill using Firecrawl web scraping, and Greg suggests extending it into recurring monthly automated reporting.
Demo: Accurate Marketing Data Analytics with Scripts 3600 Amir shows how custom Python scripts in skills calculate precise traffic analytics rather than relying on LLM guesses, with Greg verifying output fidelity.
Live Build: Creating a Tweet-to-Newsletter Skill 4400 The pair execute a live build converting Greg's tweets into a newsletter format, with Greg evaluating the output quality and suggesting hook refinements.

Statements from this episode (0)

Nothing in this episode matches those filters. clear them

Made with StarZero

Turn any episode into a week of clips.

This entire site, over 300 episodes transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.

Try the podcast clipper Learn more