Cursor 3, Gemma 4 & AI Taxes

Codex maxxing and the AI-native workplace

Brad’s North Carolina Cook Out mint Oreo shake review, and Ben’s Pacific Northwest work-trip update before the brothers dig into the big idea of the week: what it actually means to become “AI native” at work. Using Jason’s Codex maxxing article as the spark, Brad frames the shift from one-off ChatGPT-style prompts to long-running agent threads with memory, tools, voice input, steering, queued follow-ups, automations, and heartbeats. Ben translates that into a practical workplace lens with the “core four” framing—model, prompt, context, and tools—and compares onboarding an AI agent to training an intern in accounting or engineering. From there, the episode gets very hands-on. Brad explains how he thinks about memory files, Slack MCPs, voice workflows with Wispr Flow, recurring heartbeat tasks that check GitHub PRs, and Codex artifact-style outputs like PDFs, websites, and dashboards. Ben stress-tests the hype with very real questions about token budgets, enterprise ROI, Slack drafts that overcommit on your behalf, and the funny moral gray area of using powerful AI to do tiny jobs you could have done yourself in three clicks. The back half turns into a mobile-agent tour: Ben shares his Termius, VPS, SSH, and Tailscale experiment for running Codex or Claude Code from a phone, while Brad walks through Codex Remote in the ChatGPT app and why phone-to-computer agents feel like the future. They close with Brad’s must-watch mobile-app founder bookmark and a note that Episode 42 wraps season one before a short hiatus and refreshed look. Chapters: 00:00 Intro and Travel Catch-Up 02:06 AI-Native Codex Workflows 04:50 Core Four Agent Framework 07:26 Better Loops and Memory 11:07 Slack Drafts and Memory 14:23 Voice Input and Steering 19:24 Heartbeats for Recurring Work 23:52 Artifacts and Personalized Tools 26:56 Token Budgets and ROI 36:34 Bookmarks and Mobile Codex 44:04 Season One Hiatus Links: - Codex maxxing / AI-native workflow article [https://jxnl.github.io/blog/writing/2026/05/10/codex-maxxing/] - CEO’s journey of building a mobile app [https://www.youtube.com/watch?v=KgOD-_mEBfg] - Codex Remote in the ChatGPT app [https://openai.com/index/work-with-codex-from-anywhere/] #BreakevenBrothers #OpenAI #Codex #CodexRemote #AINative #AIAgents #AICoding #ClaudeCode #MCPServers #AIProductivity Creators & Guests * Bennett Bernard [https://breakevenbrothers.com/people/bennett-bernard] - Host * Bradley Bernard [https://breakevenbrothers.com/people/bradley-bernard] - Host

GPT-5.5 made Codex our daily driver

Episode 41 starts with a very Breakeven Brothers moment: the guys forget what episode they’re on, recap Brad’s family visit, and relive a 150-plus-game Soulcalibur gauntlet before diving into the real headline—why Brad now lives inside the Codex app. He explains what changed with GPT-5.5, why the model feels dramatically better at unblocking itself, and how that has shifted his workflow from small, careful requests to bigger, longer-running tasks. If you’ve been wondering whether Codex is just another AI coding surface or something materially different, this conversation gives a grounded answer from someone using it all day. From there, the episode gets practical fast. Brad walks through plugins, computer use, Git worktrees, the built-in review pane, automations, long chat compaction, and why plan mode has mostly disappeared from his workflow, while Ben stress-tests the ideas from an accounting and knowledge-work angle with Canva brochures, CRM data, email tooling, and plenty of honest beginner questions. They also compare Codex to Claude Code and Claude Co-Work, share Brad’s current default of GPT-5.5 on extra high reasoning, and close with two strong bookmarks: Greg Eisenberg and Riley Brown’s Codex masterclass, plus Evan Bacon’s serve-sim tool for showing an iOS simulator inside Codex’s right-hand pane. Chapters: 00:00 Weekend Recap and Soulcalibur 01:24 GPT-5.5 Powers Up Codex 05:16 Terminal Habits vs App Workflow 08:46 Plugins, MCPs, and Computer Use 11:22 Worktrees and Review Pane 15:05 Life After Plan Mode 20:37 Best Settings for Codex 25:48 Accounting Workflows and Plugins 30:14 Bookmarks and Closing Thoughts Links: - Codex app [https://openai.com/codex/] - Canva MCP [https://www.canva.com/help/mcp-agent-setup/] - Startup Ideas Podcast — “How to Use Codex: The Codex Masterclass” [https://x.com/startupideaspod/status/2048852194406084673] - Augmented Accounting [https://augmenticaccounting.com/] - serve-sim [https://github.com/EvanBacon/serve-sim] #BreakevenBrothers #Codex #GPT55 #AICoding #OpenAI #ClaudeCode #MCP #DeveloperTools #AIProductivity #AccountingTech Creators & Guests * Bennett Bernard [https://breakevenbrothers.com/people/bennett-bernard] - Host * Bradley Bernard [https://breakevenbrothers.com/people/bradley-bernard] - Host

Cursor 3, Gemma 4 & AI Taxes

Episode 40 opens with Brad’s dirty-30 recap from Big Sur: hiking, Japanese hot baths, unlimited food, and a first falconry experience before the brothers jump into two meaningful AI releases. They break down Cursor 3’s move to a fully agent-first interface, why that changes the feel of an AI IDE, and how Google’s Gemma 4 models look in the small-model and on-device race. There’s also a practical discussion about Apple, local inference in iOS 26, and what better on-device models could unlock for products like SplitMyExpenses. Then the episode gets into one of the most interesting collisions in tech right now: AI tax prep. Using Daniel Vassallo’s thread as the spark, Ben brings the CPA perspective, Brad brings the builder perspective, and together they unpack customer experience, trust, knowledge work, and what AI still misses. The back half covers why MCP servers suddenly feel useful again, the Axios and LiteLLM supply-chain scares, practical OWASP-style security audits, Gary Tan’s gstack claims, a joking SplitMyExpenses sponsor read, and quick bookmarks on iOS reverse-engineering tools and Google’s latest quantum-and-crypto warning. Chapters: 00:00 Big Sur birthday and falconry 02:38 Cursor 3 goes agent-first 04:04 Gemma 4 on-device push 08:42 AI taxes and CPA backlash 18:09 G-Stack versus code quality 25:23 MCP servers make a comeback 33:10 Axios breach and AI audits 40:50 TBPN jokes and sponsor banter 44:04 Bookmarks: iPhone hacks and quantum Links: - SplitMyExpenses [splitmyexpenses.com] - Feross on the Axios attack [https://x.com/feross/status/2038807290422370479] - Daniel Vassallo AI tax-prep thread [https://x.com/dvassallo/status/2039450517143974105] - gstack by Gary Tan [https://github.com/garrytan/gstack] - App Store Connect CLI [https://github.com/rudrankriyam/App-Store-Connect-CLI] - Hopper disassembler MCP [https://github.com/dflatline/hopperpymcp] #BreakevenBrothers #AICoding #CursorAI #Gemma4 #LocalAI #MCPServers #TaxTech #SupplyChainAttack #OWASP #DeveloperTools Creators & Guests * Bennett Bernard [https://breakevenbrothers.com/people/bennett-bernard] - Host * Bradley Bernard [https://breakevenbrothers.com/people/bradley-bernard] - Host

AI agents, accounting workflows & social contracts

This week, the brothers dive headfirst into the world of personal AI agents, with Brad detailing his meticulous, multi-step journey of setting up OpenClaw on a virtual private server. The conversation navigates the critical balance between unlocking immense power and managing the significant security risks that come with granting AI access to personal data. From automating mundane tasks like unsubscribing from spam emails to the dream of booking dinner reservations with a simple command, they explore the technical hurdles and trust barriers that stand between us and a truly agentic future. The discussion then shifts to the professional world, tackling the recent panic and excitement in the accounting community as many discover the power of AI tools like Claude. Are accountants officially "cooked," or is this simply the next evolution of the profession? Drawing parallels to software engineering, they debate how AI is modifying jobs rather than eliminating them. Finally, they grapple with a fascinating new social concept: "AIDR" (AI Didn't Read), questioning the authenticity and perceived effort of AI-generated communication and what it means for our human-to-human connections in an increasingly automated world. Chapters: 00:09 Introduction and Personal Catch-Up 03:02 Darren Aronofsky's AI Film Series 06:36 Brad's Experience Setting Up OpenClaw 16:34 The Broader Challenges of AI Security 21:44 Is AI Making Accounting Obsolete? 29:57 Adapting to AI and GPT-4's Power 37:48 AIDR: AI's Impact on Communication Links: - AIDR: AI Didn't Read [https://www.0xsid.com/blog/aidr] - Google Workspace CLI [https://github.com/googleworkspace/cli] - Bradley's blog post [https://bradleybernard.com/blog/one-month-at-openai-what-surprised-me-most] #OpenInterpreter #AISecurity #FutureOfWork #AccountingTech #GPT4 #ClaudeAI #DeveloperTools #DigitalEthics #AIpodcast #PromptInjection Creators & Guests * Bennett Bernard [https://breakevenbrothers.com/people/bennett-bernard] - Host * Bradley Bernard [https://breakevenbrothers.com/people/bradley-bernard] - Host

AI for taxes & the end of App Stores?

This week, the Breakeven Brothers dive into the practical (and sometimes frustrating) applications of AI, starting with a hilarious attempt to use AI for taxes and an experiment to recreate their famous intro jingle with Google's new Lyria 3 audio model. The results might surprise you. They also recap a flurry of major releases from the past few weeks, including OpenAI's Codex 5.3 Spark, Anthropic's Claude Sonnet 4.6 with its blazing fast mode, and Google's Gemini 3.1 Pro, debating the emerging market split between raw intelligence and sheer speed. The conversation then shifts to the powerful world of on-premise AI, exploring how tools like OpenCode and local LLMs can offer unparalleled security for sensitive data in industries like accounting and finance. Is it possible to get the power of modern AI without sending your data to the cloud? The hosts discuss the trade-offs and explore game-changing productivity hacks, like using the Codex app's automation features to analyze your own usage patterns. Finally, they tackle a thought-provoking tweet from Karpathy about the potential 'end of the App Store' and what the explosion of AI-generated software means for developers, legacy systems like COBOL, and companies like Apple. Chapters: 00:07 Can AI Do Your Taxes? 03:36 Recap of Recent AI Releases 05:56 Testing Google's Music AI, Lyria 09:01 Exploring Open Code with Local LLMs 14:31 The Case for On-Premise AI Models 24:55 Unlocking Power with Codex Automations 31:36 Will AI Make App Stores Obsolete? 38:43 Weekly Bookmarks and Security Concerns 43:11 Conclusion and Next Episode Teaser Links: - Replit Animation on X [https://x.com/Replit/status/2024578806208745637] - CNET: Hackers Are Trying to Copy Gemini [https://www.cnet.com/tech/services-and-software/hackers-are-trying-to-copy-gemini-via-thousands-of-ai-prompts-says-google/] - Ollama [https://ollama.com/] #AIPodcast #LocalLLM #OpenCode #AppStore #SoftwareDevelopment #Codex #GoogleGemini #OnPrem #AICyberSecurity #AIMusic Creators & Guests * Bennett Bernard [https://breakevenbrothers.com/people/bennett-bernard] - Host * Bradley Bernard [https://breakevenbrothers.com/people/bradley-bernard] - Host