L E S S O N S - with Lennox Saint

OpenAI's new $500 plan feels like the old $200 plan. that's one of six stories from OpenAI DevDay 2026 - then i set up my own dot live, on day one, with the first prompt every dot needs. the stories: the new $500 Pro plan (25x the usage of Plus, plus Ultrafast - up to 8x faster, and $60/$300 per million tokens for Astra in the API) and the $200 plan dropping from 20x to 10x Plus (Tibo: it will "net out at half the dollar in API spend"), GPT-6.1 Sol at $2/$10 per million tokens vs GPT-6 Astra's $10/$50, dots (always-on GPT-6 Astra agents with their own cloud computer and browser, 4,000+ apps, reachable in ChatGPT, Slack, Teams or by voice call), Codex cloud (close your laptop, your agents keep working), Sign in with ChatGPT (your plan pays for the AI in 16+ partner products like Devin, OpenCode and Notion, per Tibo, capped per app) plus the Agents API and the Decisions API (about 150 ms, limited preview), and ChatGPT Space and pages. the fine print: OpenAI's email says existing Pro subscribers keep their current limits through October 29, 2026, plus a one-time grant of 62,500 usage credits ($2,500 value). "feels like the old $200 plan" is my reading, not OpenAI's words. BridgeMind says he ran out of his entire weekly limit on the $500 plan in under 30 minutes - his claim. Artificial Analysis scores: Sol 52, Opus 5.5 58, Sonnet 5.5 56. one tester found Sol uses about half the tokens of Sonnet 5.5. dots come with Pro plans for now - chats with your dot don't use limits, but Codex or Work jobs it starts can. not in the EU or UK yet, and Dan Shipper (Every) reported permission issues and dropped messages. Codex cloud uses normal plan usage, and it's not in my app yet, so it's untested. the API use cases are my examples; the Hugging Face link is my own theory. the result: i set up my dot, Clarity, live on my Pro plan and ran the "restate and audit" prompt. about 20 minutes after i sent it, it restated my goals, audited my recurring jobs (84 Codex entries with 25 labelled active, 6 of those with no future run, all 9 old wiki jobs paused) and proposed one job: "one follow-through for one low-risk PR using the existing process rather than adding another monitor". nothing was approved on camera. one day-one run on my own setup - the audit is my dot's opinion. the takeaway: audit before you automate. try it yourself - give your dot this first prompt before any work: "before you do anything, read everything i've connected you to: my repos, Slack, Notion, calendar, and every scheduled task or agent i already run. 1. restate in your own words what you think my goals are and the problem i'm trying to solve. 2. list every recurring job and scheduled agent i run. mark each: keep, kill, or replace with a trigger. 3. pick the one job you'd take off my plate this week. give me the outcome, the tools you need, what you must ask me first, and a weekly budget. don't start work or create any schedule until i approve." Threadify, my own software, sponsors this episode. dots do everything for you - Threadify does one thing all day: it finds leads in Threads conversations and drafts replies in my voice. see the plans here: https://www.threadify.app/plans?utm_source=lenny-youtube&utm_medium=video&utm_campaign=lab-0011&utm_content=youtube-description-primary&video_slug=lab-0011&cta_slot=description&entry_angle=sponsor&lp_variant=plans want to sponsor a future LAB episode? email sponsors@lennysaibuilders.com sources: Tibo on the new Pro plans: https://x.com/thsottiaux/status/2104951965184925941 Tibo on the $200 plan change: https://x.com/thsottiaux/status/2104823812042940713 OpenAI's Pro email, as posted on Reddit: https://www.reddit.com/r/ChatGPTPro/comments/1wtm7o7/10x_usage_drop_for_200_plan_users/ OpenAI API pricing: https://developers.openai.com/api/docs/pricing BridgeMind (his claim): https://x.com/bridgemindai/status/2105016612365770809 OpenAI on GPT-6.1 Sol: https://x.com/OpenAI/status/2104986129686741046 Theo on Artificial Analysis: https://x.com/theo/status/2105005625726099473 OpenAI Devs on dots: https://x.com/OpenAIDevs/status/2104989680987238814 OpenAI Devs on Codex cloud: https://x.com/OpenAIDevs/status/2104997619152130278 Tibo on Sign in with ChatGPT: https://x.com/thsottiaux/status/2105006253986738615 ChatGPT on Space and pages: https://x.com/ChatGPT/status/2104986841145602351 poteto's restate prompt: https://x.com/poteto/status/2104744961904394699 Lenny's AI Builders - LAB 0011

Anthropic wants to be worth more than $2 trillion, and its own filing warns its AI could act in ways resembling blackmail. that's one of six AI stories this week - then i gave Sonnet 5.5 four hours of my real work on BuilderBench, the benchmark i built for builders, and priced every run. recorded an hour before OpenAI DevDay. the stories: Anthropic's IPO filing, per Reuters, who saw the filing (nearly $4.6 billion in revenue last year, $11.5 billion from April to June this year, more than $8 billion lost running the business last year, $518 billion promised for compute and cloud, the founders keep 50.1% of the votes, and risk language that says its AI could "resist shutdown, conceal or manipulate information" and "act in ways resembling blackmail"), OpenAI cancelling the October release of GPT-6.1 Astra after internal testing showed more deception (reported by the Wall Street Journal - OpenAI's head of safety systems said it "didn't quite meet the bar"), a pre-DevDay leak of app strings pointing to OpenAI agents called Dots that you can text, call, Slack and email, and that can buy things with your approval (a leak via TestingCatalog, not an announcement), Tibo saying the reopened $200 Pro plan will "net out at half the dollar in API spend" compared to the old one, Sonnet 5.5 at $2/$10 per million tokens - the same as GPT-6 Sol - and Anthropic's guide on when to use Sonnet 5.5 and when to pay for Opus 5.5. the plan maths are estimates, not OpenAI's numbers: Theo measured about $9,000 of Opus a month on the $200 Claude Max plan, and another creator estimated about $4,900 a month for the old ChatGPT Pro plan, so half is about $2,450. the Sonnet 5.5 benchmarks are Anthropic's chosen ones (70.6% on Terminal-Bench 4.0 vs 66.4% for Opus 5.5), and Artificial Analysis measured about $7.60 per task - the most tokens they've measured. the result: Sonnet 5.5 scored 48.98 out of 100 on BuilderBench v3. GPT-6 Sol scored 45.55, GPT-6 Astra 53.75 and Opus 5.5 68.26, but those ran on earlier BuilderBench versions (v1 and v2), so it's not a perfect head-to-head. what one run cost: GPT-6 Sol $12.75, Sonnet 5.5 $17.59, Opus 5.5 $49.67, GPT-6 Astra $87.50. my own scores, still provisional. try it yourself - price the job. run this prompt on a real task: "here is a real task from my work: [task]. here is what done looks like: [checklist]. do the whole job. at the end, list every file you made, how long it took, and anything you could not finish." Threadify, my own software, sponsors this episode. it's a lead generation agent for Threads that finds the buyer signals in the comments you're already getting. see the plans and the free trial here: https://www.threadify.app/plans?utm_source=lenny-youtube&utm_medium=video&utm_campaign=lab-0010&utm_content=youtube-description-primary&video_slug=lab-0010&cta_slot=description&entry_angle=sponsor&lp_variant=plans want to sponsor a future LAB episode? email sponsors@lennysaibuilders.com sources: Reuters' reporting on Anthropic's filing, via TechCrunch: https://techcrunch.com/2026/09/28/anthropics-prospectus-details-losses-growth-and-yes-a-warning-that-its-ai-could-end-humanity/ Anthropic on its confidential draft filing: https://www.anthropic.com/news/confidential-draft-s1-sec Andrew Curran on GPT-6.1 Astra's cancelled October release: https://x.com/AndrewCurran_/status/2104711708153618621 TestingCatalog on Dots (leak, pre-DevDay): https://x.com/testingcatalog/status/2104800309612552589 Tibo on the $200 Pro plan: https://x.com/thsottiaux/status/2104823812042940713 Theo on Claude Max usage (measured, his accounts): https://x.com/theo/status/2104683186215363058 AICodeKing on the old Pro plan's value (creator estimate): https://www.youtube.com/watch?v=EIiXhCaZ4rw Claude on Sonnet 5.5: https://x.com/claudeai/status/2104633115620823187 Anthropic's Sonnet 5.5 page: https://www.anthropic.com/claude-sonnet-5-5 Artificial Analysis on Sonnet 5.5: https://x.com/ArtificialAnlys/status/2104640155843989864 Claude Devs, the Sonnet 5.5 guide: https://x.com/ClaudeDevs/status/2104687805876367793 the guide itself: https://claude.dev/blog/building-with-claude-sonnet-5-5/ Lenny's AI Builders - LAB 0010

Opus 5.5 leads the Artificial Analysis chart. GPT-6 Sol has a lower published API list price. The half-price comparison refers to standard short-context API input and output list rates, not subscriptions or cost per finished task. But which one should you use for your own work? i look at the launches, the chart, cost per task, and one Sol miss in my own Notion planning. Then i show how to set up a five-task test for the jobs you actually do. i have not run Opus 5.5 or scored BuilderBench yet, so this episode does not pick a real-work winner. try this prompt: help me build a small benchmark for the work I actually do. Ask for my five recurring tasks. Turn each into one repeatable brief with a useful finished output, checks I can perform, and the time and cost to record. Mark what is still untested. Threadify, my own software, sponsors this episode. It helps you find potential leads in the replies to your Threads posts. See the plans and terms here: https://www.threadify.app/plans?utm_source=lenny-youtube&utm_medium=video&utm_campaign=lab-0006&utm_content=youtube-description-primary&video_slug=lab-0006&cta_slot=description&entry_angle=context&lp_variant=plans want to sponsor a future LAB episode? email sponsors@lennysaibuilders.com sources and tools: Anthropic on Claude Opus 5.5: https://www.anthropic.com/claude-opus-5-5 OpenAI's Sol/Luna announcement: https://openai.com/index/introducing-gpt-6-sol-and-luna/ GPT-6 Sol API list pricing: https://developers.openai.com/api/docs/models/gpt-6-sol Artificial Analysis index write-up: https://artificialanalysis.ai/articles/claude-opus-5-5/ Matt Pocock's Grill with Docs skill: https://github.com/mattpocock/skills/blob/main/skills/engineering/grill-with-docs/SKILL.md Lenny's AI Builders - LAB 0006 watch the episode: https://youtu.be/DAkQhc7Fh3U

i looked at OpenAI's reported maths results, the GPT-6 Sol and Opus 5.5 price fight, Grok 4.7, Muse with Shopify and Qwen-Image-2.1. Then i gave Astra my existing LAB brand and asked for a matching thumbnail, banner and podcast cover. Astra directed the work in Codex; native ImageGen made the images. I show the finished set, the references and the checks. Give it your brand first. The maths result is OpenAI's claim, not an independently established result here. Some model details were reported or rumoured when this was recorded. Qwen's downloadable model needs suitable hardware. Threadify sponsors this episode. It helps turn replies on your Threads posts into lead signals for you to review: https://www.threadify.app/plans?utm_source=lenny-youtube&utm_medium=video&utm_campaign=lab-0005&utm_content=youtube-description-primary&video_slug=lab-0005&cta_slot=description&entry_angle=context&lp_variant=plans sources and updates: OpenAI updates: https://developers.openai.com/api/docs/changelog Claude Opus 5.5: https://www.anthropic.com/claude-opus-5-5 Fable plan limits: https://support.claude.com/en/articles/15424964-claude-fable-models-on-your-plan Muse announcement: https://about.fb.com/news/2026/09/introducing-muse-personal-ai-agent/ Qwen-Image-2.1 implementation: https://github.com/QwenLM/Qwen-Image-2.1 watch the episode: https://youtu.be/452rqqtYIyg

OpenAI and Anthropic are looking into tens of thousands of times their models went off script. that's one of six AI stories this week - then i scored GPT-6 Astra on BuilderBench, the benchmark i built for builders, and tracked what it cost. the stories: Axios reports OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents where frontier models did things outside evaluators would call problematic (unnamed sources - labs run hundreds of thousands of tests, and most aren't known to have caused real harm), the OpenAI DevDay countdown and what leaked from OpenAI's code (an always-on agent called "o" and an ultrafast speed tier - leaks, not announced), the new ChatGPT app sidebar that made it feel like my whole computer, Tibo saying more resets are coming next week, David Ondrej saying his Opus 5.5 usage barely moves on the Claude Max plan, and a claim that Sonnet 5.5 could land on DevDay (Anthropic has only said "in the coming weeks"). the result: GPT-6 Astra scored 53.75 out of 100 on BuilderBench v2. on v1, Opus 5.5 scored 68.26 and GPT-6 Sol scored 45.55. Astra's estimated API cost was $87.50 ($88.65 with calibration), vs $49.67 for Opus 5.5 and $12.75 for GPT-6 Sol. so Astra cost about $38 more than Opus 5.5 and scored just under 15 points lower. my own scores, still provisional, and Astra ran on v2 while the others ran on v1 - not a perfect head-to-head. try it yourself: run this prompt on a real task. "here is a real task from my work: [task]. here is what done looks like: [checklist]. do the whole job, then list every file you made, how long it took, and anything you could not finish." your work is the benchmark. Threadify, my own software, sponsors this episode. it's a lead generation agent for Threads that finds the buyer signals in the comments you're already getting. see the plans and the free trial here: https://www.threadify.app/plans?utm_source=lenny-youtube&utm_medium=video&utm_campaign=lab-0009&utm_content=youtube-description-primary&video_slug=lab-0009&cta_slot=description&entry_angle=sponsor&lp_variant=plans want to sponsor a future LAB episode? email sponsors@lennysaibuilders.com sources: Madison Mills (Axios) on X: https://x.com/MadisonMills22/status/2103978039097037144 Axios, the full report: https://www.axios.com/2026/09/26/openai-anthropic-thousands-ai-security-incidents an OpenAI researcher on pausing big RL runs (one researcher's post, not an official statement): https://x.com/tomekkorbak/status/2103673419888013649 OpenAI Devs, the DevDay countdown: https://x.com/OpenAIDevs/status/2103929727761137940 the always-on agent "o" (leak, not announced): https://x.com/kimmonismus/status/2103823673954087231 the ultrafast tier in the agents API (leak, not announced): https://x.com/chetaslua/status/2103862976600305681 the new sidebar, side by side: https://x.com/DevAdventur3s/status/2103850957054546433 Tibo on more resets: https://x.com/thsottiaux/status/2103963215885701493 David Ondrej on Opus 5.5 usage: https://x.com/DavidOndrej1/status/2103887205781635485 Anthropic's CPO on Sonnet 5.5 "in the coming weeks": https://x.com/mikeyk/status/2102441803253535060 Jazii on Sonnet 5.5 (claim, not confirmed): https://x.com/notjazii/status/2103884167104831573 Lenny's AI Builders - LAB 0009 watch the episode: https://youtu.be/UntVig8wMb8

six AI stories, then a test: the same Claude Opus 5.5 prompt at every effort level, with the time and cost of each.

four AI stories, then a blind test: i hid the model names, made GPT-6 Sol and Claude Opus 5.5 do the same eight builder jobs, and only revealed who was who after scoring.

six AI stories from the 19 September recording: Jev, ChatGPT browser extensions, Claude Code's AGENTS.md fallback, Muse connectors, Grok Bot voice notes and a limited Pro 20x access report. then i explain the Email Triage Pro workflow i keep pinned. this audio comes from the video episode, so some demonstrations refer to the screen.

six useful AI updates, the caveats that matter, and a voice-conductor workflow that turns one messy task dump into two reviewable work lanes.

Sam Altman's OpenAI teaser, early Opus 5.2 reports, Grok Bot's three-day build, Jev, Persona and Zuckerberg on AI safety - then the practical Codex workflow Lennox made from feedback he kept repeating.

Who gets to decide how fast frontier AI moves? Lenny covers Obama, Jack Dorsey and Dario Amodei, then shares the prompt he uses when an AI agent keeps preparing instead of finishing.

Dario Amodei has proposed pacing frontier AI development. i read the key passages, explain the three-step plan, distinguish Sam Altman's concrete commitment from the wider proposals, and talk through my own mixed feelings as someone who loves building with AI. this is an unnumbered LAB pilot. for audio listeners: the three “here” call-outs refer to separate passages discussing what extra time could help researchers achieve. they are not an announced model-release freeze. read the essay: https://darioamodei.com/post/we-must-pace-the-frontier watch the video: https://youtu.be/TFS7Ef6_7oo Lo-Fi Beat | Coffee by Alex-Productions | https://onsound.eu/ Royalty Free Music by https://www.free-stock-music.com Creative Commons / Attribution 3.0 Unported License (CC BY 3.0) https://creativecommons.org/licenses/by/3.0/deed.en_US

An audio adaptation of my Astra and DaVinci Resolve editing experiment: the workflow, Gemini review, and where it stalled.

One recording, thirteen edits and my personal scores. The exact video soundtrack, with the original contestant clips and a human entry in the long-form comparison.

Five computer-use demos and one Descript editing test show where Astra helps a beginner, and where experience and taste still matter.

ChatGPT audits one month of recurring spending, protects the essential services and keeps searching until the evidence clears a real ten-percent target.

Why I'm closing Full Circle after two and a half years, what the community gave me, and why I'm going back to Threads.

ChatGPT observed every minute of one workweek. Codex analysed 12,251 unique frames, found 5.02 hours of weekly opportunity, and Lennox accepted 4.21.

ChatGPT Voice gets 90 minutes, three blind reviewers and a rescue threshold. The result is a real hire-or-fire test, not a feature demo.

A 74,200-note Second Brain, nine Obsidian crashes, one exact breaking point, and the decision that proved the system was finally useful.

An AI scored 1,158 videos against their own channel baselines, studied the 25 that broke 10x, and used the patterns to build this episode.

A proof-led autopsy of how 80 YouTube signups became 5 checkouts and 0 sales—and how a read-only AI agent found the broken measurement layer.

this is the actual build for a second Codex app that can run the AI models you choose. get the repo, every command and the copy-paste setup prompt free inside Codex Club: https://www.skool.com/codexclub someone copied my post word for word, including the same screenshot, so Episode Zero shows the method they left out. in the video, i build the three parts that make it work: - an isolated second Codex app on Mac, or separate profiles on Windows and Linux - a local proxy that translates the Responses API for other model providers - a model catalog for OpenRouter and Ollama OpenRouter gives you more cloud-model choice, but you still pay and your data travels. Ollama gives you a local, private and per-token-free path, but your hardware decides how usable it is. no magic. no “trust me.” just the build.

Lennox Saint walks through the Threadify Agent lead loop: pull comments, find hand raises, draft permission-first replies, approve before send, and use attribution to see which post created real movement.

Lennox Saint walks through the Daily Greatest Hits workflow: using Codex and Threadify to turn proven Threads posts into daily candidates instead of asking AI to invent from scratch.

A real operator walkthrough of how Lennox Saint uses OpenAI Codex as a workbench across YouTube, Skool, Threadify, Duo, newsletters, podcasting, PDFs and second-brain workflows.

A compact Codex masterclass for creators, operators, and non-coders who want to turn Codex into an AI employee instead of using it like a chat box.

A data-backed AI content forensics teardown of 27,026 Naval Ravikant tweets, showing why earned compression beats fake-deep brevity.

I analyzed 590 Casey Neistat uploads and found five YouTube rules my own channel was following that Casey's data quietly destroyed.

Threadify is not ChatGPT. In this episode, Lennox walks through the first Threadify loop: connect Threads, add one real source to your Brain, generate one post, review it, schedule it, check analytics tomorrow, and use replies to keep the loop moving. The core lesson: the first good Threadify post starts before the prompt.

An autonomous AI agent got its own Stripe, Threads, X, Gmail, and a $10,000 mission. After 96 hours, the receipts were stranger than the pitch.

A practical Codex masterclass for creators, operators, and non-coders who want an AI employee workflow instead of a terminal demo.

Jimmy went from 89 followers to 60K+ on Threads in 90 days by turning volume into a faster feedback loop.

Chris Williamson's channel looked like the perfect YouTube packaging model. Then the data started killing the rules. In this episode, Lennox breaks down what happened after reverse engineering 728 Chris Williamson videos, 727 thumbnails, 19 packaging patterns, and the four rules that did not survive the analysis. The surprising part is not that Chris does things differently. It is that the data shows why copying even a brilliant creator can be a trap. Chapters: - 00:00 the 140-view flop - 00:45 number-led hooks are not universal - 01:45 the brand mark signal - 03:08 shock face underperforms tempered intensity - 04:14 long form beats clips - 05:30 the one packaging rule that survived - 06:20 audit your own channel before copying anyone else Watch the YouTube video: https://youtu.be/JYfsnaVGYvM Try Threadify free: https://www.threadify.app/yt?utm_source=lenny-podcast&utm_medium=transistor&utm_campaign=2026-04-25-chris-williamson-forensics&utm_content=show_notes__voice__default&video_slug=2026-04-25-chris-williamson-forensics&cta_slot=show_notes&entry_angle=voice&lp_variant=default

Most Threads advice sounds useful and still makes your posts worse. In this episode, Lennox breaks down what changed after using Claude to analyze 1,996 mega-viral Threads posts. The findings are blunt: no how-to hooks in the top 10%, mid-length posts die, CTAs add friction, questions create homework, and long-form only wins when it feels personal enough to hurt. Chapters: - 00:00 Claude read 1,996 viral posts - 00:44 You do not need followers to go viral - 01:50 Mid-length posts die - 03:07 Two-post threads beat three-post threads - 04:09 Trigger words scream fake - 05:33 CTAs cost reach - 07:40 Questions bleed likes - 09:25 Long form needs blood - 12:21 The golden hour - 12:47 The Threadify vault Watch the YouTube video: https://youtu.be/pmQ0UKv0vnA Try Threadify free: https://www.threadify.app/yt?utm_source=lenny-podcast&utm_medium=transistor&utm_campaign=2026-04-21-threads-growth-is-a-lie&utm_content=show_notes__voice__default&video_slug=2026-04-21-threads-growth-is-a-lie&cta_slot=show_notes&entry_angle=voice&lp_variant=default

anthropic just published a 244-page system card about claude mythos preview, their most powerful AI model ever built. they refuse to release it. lennox read every page and delivers the contrarian take nobody else is making.

six of the most powerful people in tech declared AGI in the last four months. Lennox answers real questions from his Full Circle community with 6 live demos.

Lennox paid $6,000 for Dan Koe Kortex University mastermind, then used AI to analyze all 158 of Dan YouTube videos.

Lennox walks through the entire AI Content Forensics system live - analysing Mr. Beast's 960 videos with AI, the two-chat prompting method, and building an AI agent that runs it all with one command.

two months ago, i went live on my mum's birthday and launched threadify in front of 48 creators. this is the full replay.

i built an ai agent and fed it alex hormozi's entire youtube catalogue - 173 videos, 172 transcripts, 220 megabytes of data. the agent found 50 patterns, 9 title formulas with benchmarks, and 5 rules in my content pipeline that were statistically wrong. in this episode, i walk through each of the 5 overturned rules with the actual data behind them. title length, contrarian hooks, emotional triggers, case study structure, and pain-first openings. none of them held up. chapters: 00:00 My best video got 41 views 00:52 Rule 1 - Title length 02:03 Rule 2 - Contrarian hooks 03:08 Rule 3 - Emotional triggers 03:48 Rule 4 - Case study structure 04:22 Rule 5 - Pain hooks 05:07 What actually works - 9 formulas 06:24 The 30-60 min dead zone 06:49 Pipeline audit 07:44 What's next

12 openclaw use cases with hard numbers, real failures, and the exact prompts.

80 signups. 5 checkout attempts. A 93.7% funnel leak hiding in plain sight. PostHog, a free analytics tool, revealed 180,278 events, 1,385 rage clicks, and a 53% connection failure rate.

my monthly recurring revenue dropped $78 in two days while my views hit all-time highs. 31,000 views in 28 days, 798 subscribers, and zero new trials. my AI told me why - distribution anxiety.

Lennox breaks down his 12-agent AI content pipeline, context engineering vs prompt engineering, and the uncomfortable truth: best growth ever with 0% conversion rate.

Lennox walks through the context engineering system, 12 AI employees, and a 5-step framework for building your own agent system from scratch.

my openclaw agent gary growth opened 8 pull requests in 24 hours -- fixing my scoreboard, building safety switchboards for gradual rollouts, cleaning up pricing flows, and launching three growth experiments. i didn't ask him to do any of it.

The AI model writing my code just proved it can lie to suppliers and run monopolies. And I cant stop using it.

Episode 1. peter steinberger the creator of openclaw just joined openai.