Authority Hacker Podcast – AI & Automation for Small biz & Marketers

Send us Fan Mail I've switched from Claude Code to Codex. After a week using OpenAI's Astra model in our business, it's become the tool I reach for first. This week, it rebuilt one of our sales pages inside an awkward drag-and-drop editor with just one round of feedback. But Gael and I haven't had exactly the same experience. In this episode we break down: → Why I switched to Codex for everyday business work → What happened when we put Astra to work on sales pages and 3D design → Where complex video editing still needs human input → Why the way an AI talks to you matters more than you'd think → What changed with image editing and making websites easier for AI agents to use We're sharing what held up after a week of actual business use, and what you should know before switching. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail OpenAI finally released Astra last night, sort of. They announced it, then told almost everyone they can't have it until the weekend. Some people did get early access though, and what they've done with it is worth a look. One handed it a floor plan and got back a photoreal interior render. Another had it prep an entire video edit in Final Cut in three and a half minutes. So we lined it up against Claude Fable 5.1, which also shipped this week, on the things that actually matter for a business: design, writing, computer use, and how much of either model your plan lets you use before it cuts you off. In this episode we break down: → The one-shot design test: Astra vs Fable 5.1 vs Opus on the same prompt → Why computer use is the upgrade most business owners haven't noticed yet → Astra on the $20 plan with no five-hour limits, while Fable is capped at 50% of your usage → Anthropic's "25% permanent raise" that's really a 17% cut from September 14 → Why we're stacking AI subscriptions instead of making our next hire Pick the model you can actually use, not the one that wins the chart. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail A year ago everyone said Claude was the best AI model for writing, so that's what most of us used and never questioned again. Then Gael built a blind test: two pieces of text, no model names, pick the one you like. My top three were all cheap Chinese open source models, and Claude's newest model got knocked out early. One of the winners writes a full SEO article for about 2 cents. Try our free AI writing blind test tool: https://www.authorityhacker.com/tools/ai-writing-model-test/ In this episode we break down: → The blind writing test you can run yourself, free on our site → Why the best model for SEO articles isn't the best for sales emails → How to call these cheap models from inside Claude Code or Codex → AI watermarking, and which models escape it → ChatGPT's new triggers, multi-Gmail support, and password manager logins Stop paying premium prices for mid-table writing. Heads up: the AI Accelerator monthly price goes up Monday August 31 at midnight. Join before then and you lock in the current rate for as long as you stay. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator

Send us Fan Mail rok Bot landed this week and a surprising number of people have already switched to it. Always-on AI agents, each with their own computer in the cloud, that you talk to like colleagues in Slack. It's the closest thing yet to a plug and play AI employee. It's also $200 a month minimum, you can't bring your own API key, and once you've built your setup on it you're locked in. So we put it up against Hermes, the open source agent that does most of the same job for a fraction of the price, and worked out who should actually use which. In this episode we break down: What Grok Bot actually does, and why non-technical founders love it The $200/month lock-in and why we call it a poisoned gift Hermes Bot Mode, the open source alternative (think Webflow vs WordPress) Gael's setup where Codex installs and runs Hermes on a VPS so he never touches the server Should solopreneurs bother, or stick with Claude Code and Codex ChatGPT Computer History, which records everything you do and then tells you what to automate The free mystery model beating Fable on coding tests The tech is good. The terms are the catch. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator (Join before price goes up on Sept 1st) 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail If AI writes your content, that content is now traceable. The EU's watermarking law came into effect this month, and Claude is already leaving a hidden signature in every piece of text it generates. It's not hidden characters you can strip out. It's the word choices themselves. So even if you copy and paste the text, the signature survives. SEOs are panicking that Google will use this to penalize AI content. In this episode we break down: → How the watermark hides in word choices and survives copy/paste → What Google will realistically do with it as a ranking signal → Why Gmail could start flagging AI cold outreach as spam → Why the scariest warnings come from people with something to sell you → The new Authority Hacker MCP that plugs our membership into your AI agent The watermark is real. Most of the fear is manufactured. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail Anthropic just shipped Opus 5, and on paper it beats their best models on nearly every benchmark. Then one of the first big reviews opened with "Opus 5 is here, and I hate working with it." Both things are true. We spent the week testing it inside our business to figure out which one matters for yours. In this episode we break down: → Where Opus 5 genuinely beats Fable (and where it makes more mistakes) → The one real unlock for marketers, interactive pages that replace blog posts → A full 3D game one-shotted into a single HTML file → Cheap specialist models on OpenRouter for volume work → The new ChatGPT voice mode that runs your Mac from your phone Use it where it's strong. Skip it where it's insufferable. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail Gael can write a full blog post using about 7% of his weekly Fable limit. Most people use 50% doing the same job. The difference isn't a better prompt. He stopped letting the smartest model do the work at all. Fable briefs the task, Sonnet handles research, Opus makes the editorial calls, Codex writes the code. Fable just manages. With Anthropic cutting weekly limits by roughly a third, this stops being optional. In this episode we break down: → Why Fable is a project manager and not a worker → How to build a routing table so delegation happens automatically → Why spawning 100 subagents costs you more, not less → Running eight threads at once without losing track of them → OpenAI's new model names and what the ChatGPT and Codex merge means Token efficiency is about to be a job. Start learning it now. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail We run a seven figure business with two people and zero employees, and the reason is none of the flashy AI stuff you see on social media. It is the boring internal work: a support bot that helps members while we sleep, a database that flags which customers are about to cancel, and onboarding that runs itself at 6am every day. In this episode we open up the actual systems and show you exactly how each one is built. Join our free live webinar on Tuesday 16th June: https://authorityhacker.com/webinar In this episode we break down: → The WhatsApp onboarding skill that replaced a daily manual chore → How Gael shipped a full course's notes, images and thumbnails from one prompt → The $12/month server running our support bot 24/7 (and how it is protected from prompt injection) → A Supabase data warehouse that tells us which members are at risk of cancelling → The cancellation flow that drafts personalized save offers in seconds Automate the boring stuff. That is where the money actually is. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail A lot of people refused to leave Claude 4.6. 4.7 felt off, the writing got worse, and their skills stopped behaving like they used to. Opus 4.8 dropped, and that feeling is gone. We ran it through our real marketing skills. Meta ads, carousels, LinkedIn copy, a full one-shot website. Same prompts, same skills, just the new model. Here is how much better it actually is. In this episode we break down: → Why so many people refused to upgrade past 4.6 → Opus 4.8 vs 4.7 on real Meta ads (judge the concept, not the image) → The copywriter who can finally leave 4.6 behind → A one-shot website with GPT-image backgrounds built in → Why it is the best all-rounder now, even if Codex still codes harder Not a revolution. Just the model that finally has taste again.

Send us Fan Mail I let my co-founder roast my Claude Code setup live on this episode. He didn't hold back. Five months in, I thought I had it pretty dialed. Turns out my Claude.md is missing the most important thing, my folder structure is making my life harder, and I'm still manually triggering tasks Claude could just do for me. We go through the whole audit and the specific fixes I'm making this week. In this episode we break down: → Why Gael now says start on the Claude Desktop App, not VS Code → The "drift" problem nobody warns you about at 5 months in → The one rule every Claude.md needs so it maintains its own documentation → Why CLI beats MCP for Notion, Google Workspace, and more → How to stop being the manual trigger and let scheduled routines do the work → The "we're done" pattern that makes your skills get better as you use them If your Claude Code setup feels like it's slowly drifting out of sync, this is the audit you needed. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail GPT-5.5 dropped last week and the internet hasn't shut up about it. Switch now. Drop Claude. Move everything to OpenAI. So Gael actually did it. He swiped completely to Codex and GPT-5.5 for a week, ran the same business prompts on both models, and compared the real outputs side by side. The verdict isn't what you'd expect from the hype cycle. In this episode we break down: → Why GPT-5.5 is more thorough but Claude still wins for marketers → The two-engineers analogy that explains where each model shines → How GPT-5.5 gives you 3x more usage on the $100 plan → Real head-to-head test: same brief, very different ad creative → Why you don't have to "marry" one LLM (and how to use both) → Inside the new Codex desktop app and where it beats Claude → Why people are moving back to Opus 4.6 from 4.7 → Meta finally releases an MCP and CLI for ads → DeepSeek's new V Fall model and why it matters Don't get distracted. Pick the right tool for the work in front of you. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail We just tested OpenAI's new image model and generated a YouTube thumbnail of Gael that looks more real than he does. No cutouts, no Photoshop, one prompt. This isn't a marginal upgrade. On the LLM Arena benchmark, GPT Image 2.0 jumped over 300 Elo in a single release. That's roughly 3x the jump we got from Nano Banana Pro. Production-grade ads, fake Ahrefs dashboards, editable photos of you in places you've never been. All from plain text prompts. In this episode we break down: → The AI-generated thumbnail that looks too real → Making production ads and fake SEO dashboards in one prompt → Why graphic design just hit its ChatGPT moment → The +300 Elo leap and what it actually means for marketing → Claude Design, persistent dashboards, and the GPT 5.5 rumor If you run any AI workflow that generates images, stop and swap your model. This is the shift. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail A month ago Anthropic could do no wrong. Now every AI commentator is piling on Opus 4.7 just hours after it shipped. In this episode we cut through the noise: the model is mostly a side grade, but the new Claude Desktop app, auto mode, and a WordPress integration that unlocks fully automatic SEO are the real upgrades this week for business owners. In this episode: ➤ Our theory on why Opus 4.7 feels weaker than the benchmarks suggest (hint: new "Mythos" safeguards) ➤ Auto mode: the biggest quality of life win for non technical users, safer than bypass permissions ➤ Vibe coded websites that still let your clients manage their blog in WordPress ➤ How Claude Code is about to do real, measurable SEO on your site automatically ➤ The new Claude Desktop app vs VS Code: which one your team should actually use Want to learn how to vibe code websites, run Claude Code properly, and automate the stuff we talk about every week? Join AI Accelerator here: https://www.authorityhacker.com/ai-accelerator/

Send us Fan Mail Anthropic just announced Mythos. A model so powerful they won't release it to the public. It scored 93.9% on coding benchmarks where every other model has been stuck around 80. It hacked Firefox 84% of the time. And they're only giving it to companies like Apple and Amazon to patch their security before the rest of us get anywhere near it. We sat down to figure out what this actually means for small businesses, not the hype, but the practical reality of where AI is heading in the next 3 to 6 months. In this episode we break down: → Why Anthropic built an AI too dangerous to release publicly → What 93.9% on coding benchmarks actually means for non-developers → How Gael went from 2 weeks building an app to one-shotting it in a single prompt → The shift from giving AI tasks to giving it entire projects → Why Anthropic's revenue just jumped $10B in a single month If you're not building systems to use these models when they arrive, you're already behind. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator 💻 Learn Claude Code: https://www.authorityhacker.com/resources/learn-claude-code/

Send us Fan Mail Anthropic accidentally leaked the entire Claude Code source code. 512,000 lines of code, 1,906 files, every system prompt and sub-agent architecture exposed. We dug through the leak and found unreleased features including an always-on background mode that works while you sleep, a new model tier above Opus called Mythos, and more. We also break down the Claude token limit crisis, share practical ways to reduce your token usage, take a first look at EmDash (Cloudflare's open source WordPress replacement), dig into OpenAI's new self-serve ads launching this month, and look at Gemini's play to steal ChatGPT users. In this episode we break down: → What was inside the Claude Code leak (Kairos, Ultra Plan, Mythos) → The silent peak-hour token limit cut and how to work around it → 5 ways to use Claude Code tokens more efficiently right now → EmDash: Cloudflare's serverless WordPress killer with free hosting → OpenAI's $100M ad business and why early adopters should pay attention → Gemini's one-click ChatGPT data import play The free token party is ending. Start optimizing now or pay for it later. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-acccelerator/ 💻 Learn Claude Code: https://www.authorityhacker.com/learn-claude-code/

Send us Fan Mail Last Sunday my Gael texted me that he'd connected Claude Code to our Notion task board. You tick a checkbox, and Claude Code picks up the task, runs it headlessly, and comments back with the results. No terminal open. No VS Code. Just Notion. In this episode we break down: → How Claude's Dispatch feature lets you run Claude Code from your phone → The custom Notion system that turns task checkboxes into Claude Code executions → Scheduling tasks with due dates so Claude runs them at 3am while you sleep → Why Claude Computer Use sounds cool but is underwhelming in practice → How non-technical team members can piggyback on one Claude Code setup Notion just went from task manager to task executor. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator

Send us Fan Mail We've all been riding double usage promos on Claude and Codex for weeks. Most people don't even know these promos exist. They end March 31, and your bills are going up. In this episode we break down: → Why your AI usage costs are about to spike (and what you can do) → What the 1 million token context window actually means for quality → GPT 5.4 Mini and Nano: cheaper models that are surprisingly close to the flagship → Cowork Dispatch: controlling Claude from your phone → OpenAI's leaked internal pivot to enterprise and coding → Ahrefs data showing AI traffic barely replaces lost Google traffic → Why the "AI content is bad" stigma is officially dead If you're using AI tools for your business, pay attention to the next two weeks. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator ⏱️ Timestamps: 00:00 - The usage crunch is coming 02:00 - Double promos ending on Claude and Codex 04:04 - Reasoning effort levels explained 05:40 - 1 million token context window on Opus 07:54 - What context window actually means 10:20 - Why bigger context can mean worse quality 13:44 - Are they watering down your AI model? 15:49 - GPT 5.4 Mini and Nano models 21:30 - Cowork Dispatch: Claude on your phone 25:00 - Who should use Cowork vs Claude Code 28:40 - OpenAI pivots to enterprise (Wall Street Journal leak) 37:00 - Ahrefs: Google traffic down 20%, AI barely replacing it 39:36 - GEO won't replace lost SEO traffic 44:00 - Ahrefs opens API access to everyone 46:50 - Firehose: free web monitoring tool 48:30 - AI content is finally good enough (Ryan Law) 52:40 - The content creator job has changed forever 55:30 - AI Accelerator and getting started

Send us Fan Mail OpenAI just dropped GPT 5.4 and it's not just another model update. The first heading of their announcement wasn't coding. It was knowledge work. That's a signal. We tested it head to head against Claude Code for real business use cases, and the results might change how you work. We also walk through an AI article generator skill that finally solves the quality problem that's plagued content creators for years, Claude's new scheduled tasks feature, and Nano Banana 2, Google's rethink of AI image generation. In this episode: -> GPT 5.4's shift from coding tool to knowledge work platform -> Head to head comparison: GPT 5.4 vs Claude Code for real workflows -> Why OpenAI is targeting 3 billion knowledge workers, not just developers -> The Codex app vs Claude Code. Lock-in strategies and switching costs -> An AI article generator skill that produces genuinely good SEO content -> Claude's scheduled tasks. What they actually do (and their limitations) -> Nano Banana 2. How Google's new image model works differently -> Upcoming Gemini releases and what to expect Whether you're already deep into AI workflows or just getting started, this episode breaks down what actually matters and what you can safely ignore. Try our AI skills, workflows, and community inside AI Accelerator: https://www.authorityhacker.com/ai-accelerator

Send us Fan Mail Everyone's trying to automate social media with AI. Just plug in a prompt, hit post, watch the followers roll in. Sounds great. Except it doesn't work. It's actually one of the biggest traps in AI right now. You end up with generic, soulless posts that get zero engagement and attract zero customers. We've tested this extensively and in this episode we break down exactly where it goes wrong and what actually works. In this episode we break down: → Why fully automating social media with AI is a trap → The "emotion first" framework that actually gets engagement → How to use AI as a social media assistant (not replacement) → X's "Operation Kill the Bots" and what it means for AI content → How we cracked profitable Meta ad creatives with AI → Claude remote control, enterprise plugins, and desktop vibe coding If you're letting AI run your social media on autopilot, this is your wake-up call. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator/ ⏱️ Timestamps: 00:00 - AI + social media is a trap 02:00 - Why views and traffic don't equal money 07:00 - The emotion-first approach to social media 20:00 - Live demo: AI social media skill in action 27:00 - Auto-replies and X's Operation Kill the Bots 29:00 - Profitable Meta ads with AI (live demo) 37:00 - Claude remote control feature 43:00 - Enterprise plugins in Cowork 50:00 - Claude Desktop vibe coding 55:00 - Notion custom AI agents

Send us Fan Mail For 6 months my co-founder told me to use Claude Code. I brushed it off as it looked like a developer tool and I'm not a developer. Then I spent one morning with it and built a full financial dashboard pulling live data from two Stripe accounts, calculating VAT, mapping customer locations, and breaking down every processing fee. I haven't gone back since. In this episode we break down: → Why Claude Code is NOT just for coders → How "skills" turn AI into repeatable business workflows → A live demo building a full Meta ad campaign from scratch → Why chatbots alone won't cut it anymore This is the adjustment. If you're not making it, you're falling behind. 🔗 AI Accelerator: https://www.authorityhacker.com/ai-accelerator ⏱️ Timestamps: 00:00 - I was wrong about Claude Code 03:23 - What Claude Code actually is 07:37 - Building a dashboard with no coding knowledge 14:52 - What are "skills" and why they matter 19:00 - Newsletters, SOPs, and real business workflows 32:33 - Demo: Full Meta ad campaign from one URL 42:00 - AI Accelerator & getting started 45:00 - Should you wait for Cowork instead? 49:30 - Last call: just do it