Uploaded June 2026 | Updated September 2026, 2 weeks ago
*Workflow for building skills with Claude Code & Codex:* clickhubspot.com/lfqr
Peter Yang has a blunt piece of advice: stop using ChatGPT and Claude as chat boxes and move your real work into Codex and Claude Code. In this episode he walks Kieran and Kipp through the exact system he uses to run his newsletter and podcast on autopilot, including the one trick most people skip that makes AI output actually good: evals.
We get a live look at how Peter builds reusable skills, chains them together to prep a podcast, and then runs a pass/fail eval where a second AI agent grades the draft and rewrites it until it passes every check. We also get into the uncomfortable side: AI brain fatigue, losing the muscle to write from scratch, and why human taste is still the thing that separates good work from slop.
EP 431 Mentions
Peter Yang youtube.com/@PeterYangYT
Codex openai.com/codex/
Claude Code claude.com/product/claude-code
📌 WHAT WE COVER:
→ Why Peter says ChatGPT and Claude chat are holding you back
→ The 3-step method to turn any annoying weekly task into a system
→ What a "skill" actually is (it's just a text file)
→ How to build a skill by talking to the AI, not coding
→ Chaining skills to run a full podcast prep workflow
→ Evals explained: getting AI to grade and fix its own work
→ Why simple pass/fail checks beat AI scoring every time
→ The "skill editor" trick that kills AI slop and keeps skills to one page
→ Why your skills are only as good as the context you feed them
→ AI brain fatigue and the risk of offloading your thinking
→ Why the idea still has to come from you
→ How a non-technical person can start this in one week
🎙️ Hosts: Kieran Flanagan & Kipp Bodnar
Guest: Peter Yang, creator and operator (newsletter + podcast)
⏱️ CHAPTERS:
00:00 — The Hot Take: Stop Using ChatGPT, Switch to Codex & Claude Code
00:01 — What Evals Actually Are (And Why They Matter)
00:02 — The 3 Steps to Turn Any Manual Task Into a System
00:04 — What a Workflow Really Is (Most People Can't See Their Own)
00:05 — Inside a Personal OS: Skills as Plain Text Files
00:06 — Why Peter Switched From Claude Projects to Codex
00:07 — Browser Use, Computer Use & Why the Harness Beats the Model
00:08 — How to Build a Skill by Talking to the AI
00:09 — Killing AI Slop With a Skill Editor
00:10 — The Real Impact: A Thought Partner, Not Just a Tool
00:11 — AI Psychosis and Diluting Your Taste
00:12 — AI Brain Fatigue and Running Too Many Tasks at Once
00:14 — Why Writing From Scratch Still Matters
00:15 — Live Demo: Writing a Draft, Then Running an Eval
00:17 — How to Build a Good Eval From Your Best Work
00:18 — Why Pass/Fail Beats AI Scoring
00:20 — Why AI Always Prefers Its Own Writing
00:21 — Why AI Shaming Is Dumb (Judge the Outcome)
00:23 — How a Non-Technical Person Starts This in One Week
00:24 — Letting Codex Suggest What to Automate Next
📺 Subscribe to Marketing Against the Grain for weekly tactics on AI, marketing, and growth.
#AImarketing #ClaudeCode #Codex #AIagents #AIworkflows #marketingpodcast #AIforcreators #evals #AItools #MarketingAgainstTheGrain #contentstrategy #aiproductivity
Host Links:
📲Kipp Bodnar, twitter.com/kippbodnar
📲Kieran Flanagan, twitter.com/searchbrat
‘Marketing Against The Grain’ is a HubSpot Original Podcast // Brought to you by The HubSpot Podcast Network // Produced by Darren Clarke.
About the Show
Kipp Bodnar, HubSpot’s CMO and Kieran Flanagan Hubspot's SVP of Marketing, lead you down the rabbit hole of marketing trends, growth tactics and innovation. On the way you’ll pick up undiscovered strategies to give you that slight edge for success. These are not your typical twitter thread regurgitated marketing tactics that everyone is doing. These are new methods, with unfiltered examination of successful fresh ideas.
*Workflow for building skills with Claude Code & Codex:* clickhubspot.com/lfqr
Peter Yang has a blunt piece of advice: stop using ChatGPT and Claude as chat boxes and move your real work into Codex and Claude Code. In this episode he walks Kieran and Kipp through the exact system he uses to run his newsletter and podcast on autopilot, including the one trick most people skip that makes AI output actually good: evals.
We get a live look at how Peter builds reusable skills, chains them together to prep a podcast, and then runs a pass/fail eval where a second AI agent grades the draft and rewrites it until it passes every check. We also get into the uncomfortable side: AI brain fatigue, losing the muscle to write from scratch, and why human taste is still the thing that separates good work from slop.
EP 431 Mentions
Peter Yang youtube.com/@PeterYangYT
Codex openai.com/codex/
Claude Code claude.com/product/claude-code
📌 WHAT WE COVER:
→ Why Peter says ChatGPT and Claude chat are holding you back
→ The 3-step method to turn any annoying weekly task into a system
→ What a "skill" actually is (it's just a text file)
→ How to build a skill by talking to the AI, not coding
→ Chaining skills to run a full podcast prep workflow
→ Evals explained: getting AI to grade and fix its own work
→ Why simple pass/fail checks beat AI scoring every time
→ The "skill editor" trick that kills AI slop and keeps skills to one page
→ Why your skills are only as good as the context you feed them
→ AI brain fatigue and the risk of offloading your thinking
→ Why the idea still has to come from you
→ How a non-technical person can start this in one week
🎙️ Hosts: Kieran Flanagan & Kipp Bodnar
Guest: Peter Yang, creator and operator (newsletter + podcast)
⏱️ CHAPTERS:
00:00 — The Hot Take: Stop Using ChatGPT, Switch to Codex & Claude Code
00:01 — What Evals Actually Are (And Why They Matter)
00:02 — The 3 Steps to Turn Any Manual Task Into a System
00:04 — What a Workflow Really Is (Most People Can't See Their Own)
00:05 — Inside a Personal OS: Skills as Plain Text Files
00:06 — Why Peter Switched From Claude Projects to Codex
00:07 — Browser Use, Computer Use & Why the Harness Beats the Model
00:08 — How to Build a Skill by Talking to the AI
00:09 — Killing AI Slop With a Skill Editor
00:10 — The Real Impact: A Thought Partner, Not Just a Tool
00:11 — AI Psychosis and Diluting Your Taste
00:12 — AI Brain Fatigue and Running Too Many Tasks at Once
00:14 — Why Writing From Scratch Still Matters
00:15 — Live Demo: Writing a Draft, Then Running an Eval
00:17 — How to Build a Good Eval From Your Best Work
00:18 — Why Pass/Fail Beats AI Scoring
00:20 — Why AI Always Prefers Its Own Writing
00:21 — Why AI Shaming Is Dumb (Judge the Outcome)
00:23 — How a Non-Technical Person Starts This in One Week
00:24 — Letting Codex Suggest What to Automate Next
📺 Subscribe to Marketing Against the Grain for weekly tactics on AI, marketing, and growth.
#AImarketing #ClaudeCode #Codex #AIagents #AIworkflows #marketingpodcast #AIforcreators #evals #AItools #MarketingAgainstTheGrain #contentstrategy #aiproductivity
Host Links:
📲Kipp Bodnar, twitter.com/kippbodnar
📲Kieran Flanagan, twitter.com/searchbrat
‘Marketing Against The Grain’ is a HubSpot Original Podcast // Brought to you by The HubSpot Podcast Network // Produced by Darren Clarke.
About the Show
Kipp Bodnar, HubSpot’s CMO and Kieran Flanagan Hubspot's SVP of Marketing, lead you down the rabbit hole of marketing trends, growth tactics and innovation. On the way you’ll pick up undiscovered strategies to give you that slight edge for success. These are not your typical twitter thread regurgitated marketing tactics that everyone is doing. These are new methods, with unfiltered examination of successful fresh ideas.
