Claude and Codex workflows depend on more than the model: context, permissions, quotas, and the harness shape the result.

Mike, Joe, and Josh talk through a practical AI coding agent discussion: testing a new AI coding tool on Windows, deciding how much access to give an agent, how Claude and Codex handle projects differently, and why moving work between harnesses is not always seamless.

Topics include:

- AI coding agents on Windows

- Claude, Codex, and different agent harnesses

- Giving agents repo access without handing over private data

- Owning your context instead of fighting tool-specific project state

- Claude subscription limits, third-party usage, and token burn

- Why learning the harness matters before expecting the workflow to transfer

Chapters:

- 00:00 Opening

- 00:23 Combining the episode and testing an AI tool on Windows

- 11:04 The first magic and reliability gap

- 20:59 Enterprise concerns and agent permissions

- 22:25 Letting Codex work on trusted repos

- 27:33 Context windows and agent harnesses

- 34:23 Claude, subscriptions, and third-party harnesses

- 39:13 Extra usage, token burn, and safety questions

- 43:09 Owning your context

- 44:19 Moving context between different harnesses

- 50:04 Learning how each harness works

More Build It Faster: https://fasterpod.com

Mike and Joe celebrate launching their Google Meet lean coffee app, work through branding and early feedback, reduce friction in the Meet experience, assess mobile usability, and reflect on AI-assisted development patterns they want to reuse.

- 0:00 Celebrating Rapid Development

- 2:47 Exploring Branding and Domain Names

- 5:34 Feedback and User Experience

- 8:37 Mobile Experience and Usability

- 11:49 Lessons Learned from Development

- 17:38 Applying Knowledge to Future Projects

Joe preps for the AI Engineer World’s Fair, Mike and Joe review Google Marketplace rejection fixes, then Mike shares how he used Claude for a first production PR and automated his photography paperwork workflow.

- 0:00 Anticipation for the AI Engineer World's Fair Conference

- 2:53 Goals and Expectations for the Conference

- 6:06 Networking and Knowledge Acquisition Strategies

- 9:04 Exploring AI in Literature and Media

- 12:08 Preparing for Effective Networking

- 14:57 Transitioning to New Topics and Future Discussions

- 16:05 Navigating Rejection: Lessons from Google Marketplace

- 21:00 Personal Growth: Achievements in Project Management

- 33:34 Streamlining Photography: Automating the Creative Process

AI agent skills become much more useful when you treat them as context engineering, not just prompt templates.

Mike and Joe start with launch prep for Meet for Coffee, including naming cleanup, a simple homepage screenshot, support/terms polish, Terraform-managed infrastructure, and what it means to actually launch something instead of releasing it into the void.

Then Joe gives a practical walkthrough of how he is using AI agent skills at work:

- Turning repeated instructions into modular skills

- Keeping the agent instruction budget under control

- Boiling internal onboarding docs into usable agent guidance

- Converting governance processes into prepared inputs

- Using Markdown and Mermaid instead of traditional slide decks

- Planning an AI conference schedule from a sessions JSON file

- Why context engineering can matter more than perfect prompting

- How loose, natural language can help agents understand the spirit of the task

Mike and Joe also talk about AI-assisted knowledge work, clipping useful podcast segments, and the feeling of “context bending” when agents can pull, split, summarize, and apply information across repositories and workflows.

- 00:00 Meet for Coffee launch check-in

- 01:14 Homepage screenshot and product simplicity

- 04:51 Support, terms, and launch blockers

- 06:31 Terraform and infrastructure as code

- 07:31 Launch milestones and invite flow

- 10:38 Product launch, Product Hunt, and growth

- 12:17 Clipping the gold from the podcast

- 13:29 Past product launches and failures

- 16:32 Joe’s AI presentation workflow

- 22:17 What AI agent skills are for

- 26:15 Boiling governance into agent skills

- 29:31 Markdown, Mermaid, and presentation deliverables

- 36:06 Compliant inputs from repository context

- 40:21 Using AI to plan a conference schedule

- 43:34 Context engineering vs prompt engineering

- 45:00 Prompting with natural language and intent

- 52:53 AI tools changing knowledge work

- 55:41 Context bending and performance reviews

Tags: ai agents, agent skills, context engineering, prompt engineering, agentic engineering, claude skills, coding agents, ai workflow, terraform, markdown presentations, meet for coffee, build it faster

Mike and Joe discuss using Claude with GitHub Issues, PRs, repo instructions, local dev, and Google Meet add-on work while refining their lean coffee tool’s UI and workflow.

- 00:00 Mike tries the new AI-assisted workflow

- 01:13 Old Git tooling mistakes and a cleaner process

- 03:47 PRs, deploys, and local development

- 06:07 Claude Code limits and session timing

- 07:18 GitHub Issues as agent memory

- 10:58 Infrastructure, deploy labels, and environments

- 14:09 Wrap-up and adversarial check skills

- 17:27 Planning sessions, Slack context, and model choice

- 20:29 Memory files versus repo instructions

- 23:27 Issue granularity and agent-ready work

- 27:04 What changed in the product this week

- 28:41 UI cleanup, menus, confirmations, and controls

- 36:28 Simplifying the start timer behavior

- 39:49 Why standalone browser mode still matters

- 45:01 Seeing the Google Meet add-on working

- 49:40 The pivot, the beachhead, and what comes next

- 52:36 Small UI bugs and mobile-width polish

- 55:08 What is still blocking public release

Tags: claude code, github issues, ai coding agents, software development workflow, pull requests, google meet add-on, lean coffee, product development, build it faster, agent memory, developer workflow, ai assisted coding

Mike and Joe turn a sprawling lean coffee/video side project into a focused Google Meet add-on plan: remove built-in video, target book club first, use ephemeral peer-to-peer boards with export, and create issues from the discussion.

00:00 Setting up a side project session

01:51 Level setting: Instant Coffee and book club

04:25 Asking for a real plan

07:08 Why built-in video is the wrong problem

10:30 Formally cutting video from the app

12:19 Google Meet add-on as the new target

13:35 Free, freemium, and keeping costs low

15:45 Ephemeral boards, exports, and imports

17:35 Premium feature ideas

20:27 First version: peer-to-peer board with export

22:03 Shipping now, premium later

23:34 Do we need a standalone app?

27:53 Going all in on the beachhead

29:33 Lean coffee only, Google add-on only

31:34 Naming and searchability

34:33 Auto-facilitation and future features

36:19 Next step: build the Meet add-on

38:19 Working together through issues

40:08 Cleaning the backlog

41:58 Deployment path and handoff

43:57 The relief of cutting scope

Josh came in with one goal: resist showing us the 3D demo he spent two weeks building. He succeeded. Barely.

Instead we talked about the actual interesting stuff — why he's building systems for AI agents instead of humans, how a mech arena became his favorite benchmarking tool, and what happens when you let Claude playtest your game engine for you.

Also: two Blackwell GPUs, a Glicko rating system, and the most cursed MMORPG premise we've ever heard.

• 00:00 Intro — Mike's late, just Joe and Josh
• 01:04 Six weeks later: redefining success
• 02:34 Enter Matt, the PM partner
• 04:35 Federated agents and the mech arena
• 07:44 BattleBots but autonomous
• 09:20 Why he killed the 3D demo
• 17:32 Mike finally shows up
• 20:58 How the arena actually works
• 24:55 Josh's model stack and hardware ceiling
• 35:48 Monday idea, Friday artifact
• 38:23 BYO LLM: prompt it, throw it in, watch it lose
• 43:15 Tournaments, ratings, Steam for agents
• 51:28 Token fight economics
• 54:16 Three-day POC rule and wrapping up

Mike shares his recent deep dive into AI tools like Claude, exploring how he's integrating them into work processes, building AI-driven workflows, and lessons learned along the way. The conversation covers practical strategies for leveraging LLMs, managing code projects, and enhancing communication and project management with AI.

• 00:00 Introduction and Personal Updates
• 03:17 The Impact of AI on Work Habits
• 06:11 Learning and Teaching AI Tools
• 09:21 Building an AI Guild at Work
• 12:10 Problem Solving with AI
• 18:17 Creating a Google Form with AI
• 24:08 Iterating and Improving Processes with AI
• 28:20 Planning and Context Management in AI Usage
• 30:45 Navigating AI Tools for Project Management
• 32:00 Creating Effective Project Roadmaps
• 34:53 Leveraging AI for Code Analysis
• 38:43 Integrating AI with Data Sources
• 41:43 Focusing AI's Attention on Relevant Information
• 44:46 Optimizing AI for GitHub and Snowflake Access
• 48:57 The Role of AI in PR Management
• 53:39 Trusting AI: The Need for Human Oversight