Lenny’s AI Builders

Lenny’s AI Builders

Download on the App Store

All Lenny’s AI Builders episodes

  • GPT-6.2 just dropped. here's which one to use - and when the 4x price is worth it | LAB 0046
    GPT-6.2 just dropped, and Theo called Astra "the worst model OpenAI has ever released" (to be fair, he later posted that one feature made it usable for him). so is it worth it? if you build but don't code, this one's for you.
    the price: Astra is $8 per million tokens in and $40 out. Sol is $2 in and $10 out. so Astra is 4x the price. Artificial Analysis puts Astra second on its index, and Sol on par with GPT-6 Astra at about 15% of the cost per task.
    i ran both on my own benchmark - ten real builder jobs, like a signup page, a lead magnet and a research brief - and opened the sealed result live on camera (0:42). cheap per token isn't cheap per job, so at the end (11:14) i go through it task by task.
    my rule (13:48): use GPT-6.2 Sol on fast mode for computer tasks and computer use. give Astra a shot for video editing and motion graphics. the work picks, not the price tag - so test new models on the stuff you do day to day.
    also this week:
    - Claude checks Claude (4:43): the Claude Code team says Claude writes and reviews about 90% of their code, and in their own testing a separate grader checking the work lifted task success by up to 10 points.
    - Grok Bots, one job each (5:53): xAI says five Bots now have full-time roles on its accounting team, each owning one job end to end.
    - fold your subagents (7:56): Anthropic's guide on when to fold subagents back into your main agent loop, instead of giving every team its own subagent like a copy of your org chart.
    - Dots get an email (9:03): a post from Samir (@_samirism) says OpenAI is testing an email address for your Dot, so you could cc it to find a time or forward it a receipt. not out yet.
    - Jensen Huang (10:03): he says he'd be honored to come back, and Nvidia's board says it has no plan to bring him back. Nvidia's latest quarter: $115.8B revenue, more than double a year ago.
    today's sponsor is my software, Threadify: your lead generation agent for Threads, X and YouTube. it reads the replies on your posts, flags the people showing real buying signal and drafts a warm opener in your voice. then it waits: you approve every send. free 7-day trial, card required:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0046&utm_content=sponsor&video_slug=lab-0046&cta_slot=podcast_description&entry_angle=lead_signal
    to sponsor a future episode, email [email protected]
    links:
    - OpenAI: GPT-6.2 Astra and Sol: https://openai.com/index/introducing-gpt-6-2-astra-sol
    - Artificial Analysis on GPT-6.2: https://x.com/ArtificialAnlys/status/2122750634210529569
    - Theo, first take: https://x.com/theo/status/2122757647170703550
    - Theo, later: https://x.com/theo/status/2122827076852138024
    - Sam Altman: https://x.com/sama/status/2122798042734633452
    - Matt Shumer: https://x.com/mattshumer_/status/2122752192226406421
    - how the Claude Code team thinks about verification: https://claude.dev/blog/how-the-claude-code-team-thinks-about-verification/
    - xAI: Grok Bots on the accounting team: https://x.ai/news/grok-bot-accounting
    - Anthropic: when to fold subagents into your main agent loop: https://claude.com/resources/articles/building-agents-when-to-fold-subagents-into-your-main-agent-loop
    - Samir on Dot email: https://x.com/_samirism/status/2122481776547516446
    - Jensen Huang: https://x.com/JensenHuang/status/2122726005219737649
    - CNBC: Nvidia's quarter: https://www.cnbc.com/2026/11/17/nvidia-nvda-earnings-report-q3-2027.html
    - The Verge: California sues over the Hugging Face hack: https://www.theverge.com/ai-artificial-intelligence/1034815/california-openai-hugging-face-lawsuit
    - my last benchmark episode: how Claude did (LAB 0040): https://youtu.be/rqeKU4Fr_pQ
    0:00 hook: GPT-6.2, which one to use
    4:25 thesis: cheap per token isn't cheap per job
    4:43 Claude checks Claude: the verification playbook
    5:53 Grok Bots: one job each
    7:05 sponsor: Threadify
    7:56 Anthropic: fold your subagents
    9:03 Dots get an email
    10:03 Jensen asks to come back
    11:14 as promised: task by task
    13:48 my rule
    watch the full episode: https://youtu.be/9s3sF5-YDvs
    15 min
  • make Claude Code use fewer tokens with one new setting - and when it backfires | LAB 0045
    if you build with Claude Code, here's how to make it use fewer tokens with one new setting, and when it backfires.
    the setting is Code Mode: Claude writes code that makes several tool calls in one step. turn it on with one environment variable:
    CLAUDE_CODE_EXPERIMENTAL_CODE_MODE=1
    it's experimental, off by default (v2.1.323 or later) and only runs in the interactive terminal: not claude -p, the Agent SDK, the VS Code panel, the desktop app or cloud sessions.
    my test (0:13): the same jobs with Code Mode off and on, three pairs of runs on Opus 5.5. tokens dropped 36 to 38% in every pair. the catch: my first run cost 83% more at api prices, because it wrote far more to the one-hour cache, and on Opus 5.5 that write costs 40 times a read ($8 vs $0.20 per million tokens). the other two pairs came out 15% and 17% cheaper. so check your cost, not just your tokens. Anthropic's page never mentions tokens, so the token numbers here are from my own test.
    when to switch it on: Anthropic says tasks with many related tool calls, like a change that repeats across many files.
    the practical (10:47): Gabriel Petersson's prompt, "make 10 different increasingly high variance versions and show me, then i can pick the best one". i ran it on the hook of my Threads post with 444K views. number 1 stayed close: "okay, why the hell isn't anyone talking about this?" number 10 went far: "i found out something today and honestly wish i hadn't." try it on designs, landing pages and plans too, then pick your favorite parts of each.
    also this week:
    - Qwen4-Max and Hy4 flash (2:45): Hy4 flash is $0.07 in / $0.28 out per million tokens, and Tencent says it's close to Haiku 5.5 in a blind test with 85 of its own users. Qwen4-Max is $1.50 / $4.50, and Alibaba says cost per task drops about 60% on its own test. Haiku 5.5 is $0.10 / $0.50 (the $0.10 is for prompts up to 100,000 tokens), so Qwen4-Max costs more than Haiku.
    - Anthropic's guide to always-on agents (4:14): an event inbox, timers, memory, permissions and an outbox. the agent sleeps most of the time, your passwords sit in a proxy outside it, and memory only holds what it can't look up again. my version: one rules file it loads every run, plus one progress file per job.
    - iPhone Duo (6:53): Claude Code on desktop can run your app on an iPhone Duo in the iOS simulator, then fold and unfold it to test each layout. beta, Xcode 27.1 or later.
    - Hugging Face and OpenRouter (8:07): one token and one endpoint for every open model on the Hub that a provider serves, no markup from Hugging Face, setup guides for Claude Code and Codex, and $2 of credits a month on PRO.
    - Jensen Huang (9:30): a source told CNBC Nvidia's board is not considering bringing him back, and a petition to reinstate him has 180,000 signatures (per CNBC). after filming, Nvidia reported $115.8B revenue for the quarter, up 103% on a year ago. that was Jensen's quarter.
    today's sponsor is my software, Threadify: your lead generation agent for Threads, X and YouTube. it tracks your links, so you see which posts actually drive clicks, then you do more of what's working. free 7-day trial, card required:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0045&utm_content=sponsor&video_slug=lab-0045&cta_slot=podcast_description&entry_angle=whats_working
    to sponsor a future episode, email [email protected]
    links:
    - Code Mode docs: https://code.claude.com/docs/en/code-mode
    - Anthropic prices (cache, Haiku 5.5): https://platform.claude.com/docs/en/about-claude/pricing
    - Hy4 flash (Tencent): https://x.com/i/status/2122217770314301469
    - Qwen4-Max (Alibaba): https://www.alibabacloud.com/en/press-room/alibaba-unveils-qwen4-max
    - always-on agents guide: https://claude.com/resources/articles/building-always-on-agents
    - iPhone Duo simulator docs: https://code.claude.com/docs/en/desktop-ios-simulator
    - Hugging Face and OpenRouter: https://huggingface.co/blog/openrouter
    - CNBC on Jensen Huang: https://www.cnbc.com/2026/11/16/nvidia-ceo-jensen-huang.html
    - Nvidia's results (SEC): https://www.sec.gov/Archives/edgar/data/0001045810/000104581026000088/q3fy27pr.htm
    - Gabriel Petersson's prompt: https://x.com/i/status/2122083139971416106
    - my Jensen take (LAB 0042): https://youtu.be/JelRGipwZ_Y
    0:00 make Claude Code use fewer tokens with one setting
    0:13 Code Mode: my test, the catch, when to switch it on
    2:10 this is the LAB
    2:28 today's thread: the best model wins
    2:45 Qwen4-Max and Hy4 flash vs Haiku 5.5
    4:14 Anthropic's guide to always-on agents
    6:12 today's sponsor: Threadify
    6:53 iPhone Duo in the Claude Code simulator
    8:07 Hugging Face and OpenRouter: one token for open models
    9:30 Jensen Huang: the board, the petition, the earnings
    10:47 the practical: 10 high-variance versions
    13:26 the best model wins, so test it on your own work
    watch the full episode: https://youtu.be/863KczMCX6w
    14 min
  • stop babysitting your AI agents - Claude Code's coordinator does it now, you set 3 rules | LAB 0044
    if you run AI agents, you know the babysitting. Claude Code's coordinator now does more of it for you, and all you set is three rules.
    Anthropic says the coordinator in Claude Code projects is now more proactive: it answers your questions in the project chat and gives you updates on your threads. two of my rules go straight into the project chat:
    - run at most two threads at a time
    - only post when something finishes or is blocked
    the third is a setting: Project settings > General > Thread model. a new project runs every thread on Opus at high effort, which Anthropic says draws on your plan fastest, so i run my threads on Sonnet 5.5.
    the catch: Anthropic's docs say a project uses your limits faster than a single session does, and a thread limit you ask for is a preference, not a hard cap. stuck threads still need you. and projects are in beta, so you might need to join the waitlist (link below).
    the practical (11:48): the same job twice in Claude Code projects: six pages, one thread each, a short note per page, then a summary. with no rules, all six threads started at once on Opus 5.5. with my three rules, two ran at a time on Sonnet 5.5, and the coordinator told me when one got stuck (it needed GitHub access). both runs used about 3.6M tokens, so the rules didn't save tokens. what they change is pacing: the same work spread over more time. one run each, so it's a first look.
    what i run instead (1:34): T3 Code. it's free and open source, with my machines and accounts across OpenAI and Anthropic in one place (no affiliation). my conductor skill for it keeps one main thread moving the others and updates a review desk every three hours. free on GitHub, link below.
    also this week:
    - Grok Bot can now join meetings (3:30). it joins under its own name, answers in the meeting chat and sends you a summary and action items after. xAI's docs never say it talks. it's on Team, Enterprise and Ultra plans.
    - ChatGPT Wallet (5:07). OpenAI says your dot can now buy things for you with your approval, without getting your actual card details. you tap approve and it pays with a virtual card. it's rolling out to Pro subscribers in the US only.
    - DeepSeek's app features by API (7:34). search, files, links, safety checks and voice are now in the DeepSeek API, plus an open-source demo app. search is $0.75 per 1,000 searches; deepseek-flash is $0.30 in (cache miss) and $1.20 out per million tokens at peak, half that off-peak.
    - ask your coding app to restyle itself (9:10). Anthropic shipped custom themes for Claude Code on desktop, built with mods, and about four hours later Theo posted the same for T3 Code. heads up: a mod runs with your permissions, so only install ones you trust.
    - GPT-6.2 is still a no-show (10:02). when i checked on 14 Nov, OpenAI's developer docs still pointed to GPT-6 Astra for the most demanding work. three episodes in a row.
    today's sponsor is my software, Threadify: your lead generation agent for Threads, X and YouTube. it takes the content admin off your plate so you can double down on what works. the agent drafts and schedules your posts, and you approve them. this is for you if you have an audience on Threads and something to sell. free 7-day trial, card required:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0044&utm_content=sponsor&video_slug=lab-0044&cta_slot=podcast_description&entry_angle=post_engine
    to sponsor a future episode, email [email protected]
    links:
    - Anthropic's post on the coordinator update: https://x.com/i/status/2121375309287301360
    - Claude Code projects docs: the rules, limits and settings (Anthropic): https://code.claude.com/docs/en/claude-projects
    - join the waitlist for Claude Code projects (Anthropic): https://claude.com/form/projects
    - my T3 Code conductor skill (free on GitHub): https://github.com/lennoxsaint/Skills/tree/main/t3-code-conductor
    - T3 Code: https://t3.codes
    - Grok Bot's meetings docs (xAI): https://docs.x.ai/grok-bot/meetings
    - OpenAI's ChatGPT Wallet post: https://x.com/i/status/2121283151729496235
    - DeepSeek's API news: https://api-docs.deepseek.com/news/news261113
    - Claude Code mods docs (Anthropic): https://code.claude.com/docs/en/plugins/mods/overview
    - OpenAI's latest-model guide: https://developers.openai.com/api/docs/guides/latest-model
    - the episode before this one (LAB 0043): https://youtu.be/uZzfiAMyG8o
    0:00 stop babysitting your AI agents: 3 rules
    0:18 Claude Code's coordinator: the update, the rules, the catch
    1:34 what i run instead: T3 Code and my conductor skill
    2:22 this is the LAB
    2:51 today's thread: the hard part is getting abstracted away
    3:30 Grok Bot can now join meetings
    5:07 ChatGPT Wallet
    6:38 today's sponsor: Threadify
    7:34 DeepSeek's app features by API
    9:10 ask your coding app to restyle itself
    10:02 GPT-6.2 no-show
    11:48 the practical: 3 rules, with and without
    13:28 for now, you still set the rules
    watch the full episode: https://youtu.be/8Dj5OXFTWww
    14 min
  • how to cut your AI bill with $1.50 Gemini 4 Hydrogen - and the jobs i'd never give it | LAB 0043
    here's how to cut your AI bill with Google's $1.50 model, Gemini 4 Hydrogen. it shipped at $1.50 per million input tokens and $7.50 per million output tokens. Opus 5.5 is $4 and $20. so the token price is about 60% lower.
    then i ran the same two boring jobs on both models, three times each. both models passed every check i ran. at list prices:
    - cleaning a customer list: 25c on Hydrogen, 33c on Opus 5.5
    - matching invoices: 13c on Hydrogen, 8c on Opus 5.5
    - both jobs once: 38c against just under 42c
    that's about 8% cheaper, not 60. and the jobs i'd never give it are the short, quick ones: Hydrogen took about a minute to write a really short answer, and that job cost more than it did on Opus 5.5. cheap per token isn't the same as cheap per job.
    how i tested it: 13 November, on made-up data. Hydrogen ran through Google's API with thinking set to high. Opus 5.5 ran with effort set to high. each model got the same brief and the same data. 12 runs, and all 12 passed my checks. the bills are list prices, worked out from the tokens each run used. two jobs is a small test, so run your own job before you switch.
    the practical (9:29): xAI's X Subscription page lists Cursor in its Plus and Premium plans, so if you pay for one of those you already have Cursor access. connect Cursor inside T3 Code, pick Gemini 4 Hydrogen and use it as your worker model when you want a cheaper version of Opus. heads up: Hydrogen isn't in the plan's included usage. it bills at Google's list price through top-ups. i hadn't run this route when i filmed, and i say so on camera.
    also this week:
    - Grok Bot gets your Mac apps (2:08). the Grok Bot account says Grok Bot on Mac can now use seven Apple apps and services: Messages, Mail, Notes, Reminders, Calendar, Location and Contacts. it also has a vault, so passwords and API keys stay out of the chat.
    - Claude Meetings (3:53). it's rolling out in beta on Team and Enterprise plans, Google Meet only for now. by default it answers in writing in the Meet chat.
    - GPT-6.2 was a no-show (5:49). OpenAI hadn't announced it anywhere i could find when i filmed. what did land is a banked reset: a one-time Codex limit reset.
    - browser-use now works with 1Password (7:26). you approve the login in 1Password and it fills it in. their docs say the password is never passed to the agent. the catch is service account mode, where the docs say the agent receives the value.
    - Nvidia told the SEC that Jensen Huang has resigned as president, CEO and board member, effective immediately (8:31). CFO Colette Kress is interim CEO, and the filing gives no reason. i also read out press reports: the Financial Times, CNBC, and a Reuters report i've only seen quoted. those are reports, and none of them is in Nvidia's filing.
    - Gemini 4 Hydrogen on my own benchmark, BuilderBench, revealed on camera (10:26).
    Threadify is my lead generation agent for Threads, X and YouTube. feed its Brain your documents, transcripts and past hits, and the agent drafts posts, threads and replies that already sound like you. nothing goes out until you say yes. connect the Threadify MCP to Claude to see it for yourself. free 7-day trial, card required:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0043&utm_content=sponsor&video_slug=lab-0043&cta_slot=podcast_description&entry_angle=voice_brain
    to sponsor a future episode, email [email protected]
    0:00 how to cut your AI bill with Gemini 4 Hydrogen
    0:14 both bills: Hydrogen v Opus 5.5
    1:33 this is the LAB
    1:50 today's thread: cheap per token isn't cheap per job
    2:08 Grok Bot gets your Mac apps
    3:53 Claude Meetings
    4:55 today's sponsor: Threadify
    5:49 GPT-6.2 no-show, banked reset live
    7:26 browser-use + 1Password
    8:31 Nvidia's SEC filing: Jensen Huang resigns
    9:29 the practical: Hydrogen as your worker model
    10:26 Gemini 4 Hydrogen on BuilderBench: the reveal
    11:42 tell me if that score sits right
    watch the full episode: https://youtu.be/uZzfiAMyG8o
    13 min
  • Grok Bot 1.0 isn't new - here's the one upgrade that actually matters | LAB 0042
    Grok Bot 1.0 isn't new. the bot has been out since August, and xAI's own page calls 1.0 a major update to its AI teammates. the headline feature, team bots, is for Team and Enterprise plans. so if you're solo, you got a number, not really an upgrade.
    the upgrade that actually matters is the quiet one: Grok Bot can now join your Tailscale network. i put my bot on my private network, where four read-only doors lead into my Mac. that morning it counted 42 active agent tasks and listed my four local models in 3 minutes 27 seconds. one read out of four timed out twice. then i ran it again live at 1:38 and it failed on camera. i left that in.
    also in this episode: Gemini 4 Carbon's score on my own benchmark, BuilderBench, revealed live (3:16). Fable 5.2 takes number one in Agent Arena (30.4% net improvement at $7.32 a task, against 19.5% and $5.33 for Opus 5.5 - Arena doesn't define "net improvement" on the pages i read). Gemini Decision 1, a model that only picks an answer off a list (2 to 255 choices, $0.10 per million input tokens, output free). the GPT-6.2 rumour, which was still a rumour when i filmed. and Nvidia's board and Jensen Huang.
    update since filming: on November 12 Nvidia announced that Jensen Huang resigned as CEO, and CFO Colette Kress is interim CEO. when i filmed, it was still a report.
    Threadify is my lead generation agent for Threads, X and YouTube. one of my posts got more replies than i could read, so i had Threadify show me the ones that matter most. free 7-day trial, card required:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0042&utm_content=sponsor&video_slug=lab-0042&cta_slot=podcast_description&entry_angle=lead_signal
    want to sponsor a future episode? [email protected]
    0:00 Grok Bot 1.0 isn't new
    0:43 so what is 1.0?
    1:14 the real upgrade: Tailscale and four doors
    1:38 the live re-run (it failed)
    2:37 this is the LAB
    2:57 today's question: does it beat Opus 5.5?
    3:16 Gemini 4 Carbon on BuilderBench: the reveal
    5:35 Fable 5.2 takes number one in Agent Arena
    6:42 today's sponsor: Threadify
    7:43 Gemini Decision 1
    8:55 the GPT-6.2 rumour
    9:49 Nvidia's board and Jensen
    10:47 the practical: the finishing work paragraph
    12:17 so, does it beat Opus 5.5?
    watch the full episode: https://youtu.be/JelRGipwZ_Y
    13 min
  • i ran my viral Fable 5.2 conductor for 40 hours - here's what it actually built | LAB 0041
    i ran one Claude Fable 5.2 conductor on my real app, Eddy Studio, for 40 hours. it messaged 73 worker threads, and 25 merged pull requests link back to the run. my weekly Fable limit was gone in 13 hours.
    the new meta: Fable 5.2 costs twice what Opus 5.5 does, so Fable directs the show and Opus does the work.
    the practical, four steps: open Claude Code, type /model and pick Fable 5.2, set your sub-agents to Opus 5.5, then give it a massive job and add the conductor paragraph. start with three workers, not 73. the paragraph and the full skill: https://github.com/lennoxsaint/Skills
    also this episode: Gemini 4 Carbon, Claude Sheets, Data and Whiteboards, Anthropic's free full reset, Businesses in ChatGPT and what its terms say, and Anthropic's IPO filing with the Astra rumour (one account's post - a rumour).
    Threadify is my lead generation agent for Threads, X and YouTube. connect the Threadify MCP to Claude and it manages my posting schedule, picks the best time for me to post and shows me the most important replies i need to answer. free 7-day trial:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0041&utm_content=sponsor&video_slug=lab-0041&cta_slot=podcast_description&entry_angle=agent_mcp
    0:00 i ran one Fable 5.2 conductor for 40 hours
    0:16 what it actually built
    0:59 the conductor paragraph, word for word
    1:41 the stats and the mess
    2:38 the new meta: Fable directs, Opus works
    3:18 this is the LAB
    3:41 the thread: Anthropic's aura is not a coincidence
    4:00 Gemini 4 Carbon
    5:08 Claude Sheets, Data and Whiteboards
    7:05 today's sponsor: Threadify
    7:52 the free full reset
    9:13 Businesses in ChatGPT
    11:01 Anthropic's IPO and the Astra rumour
    12:59 the practical: four steps
    watch the full episode: https://youtu.be/yw5yOAV3KPI
    15 min
  • Claude Fable 5.2 just dropped. here's the one job i'd actually pay it for | LAB 0040
    Claude Fable 5.2 just dropped, and there's only one job i'd actually pay it for: the boss. it reads the task, picks which worker gets it and knows when the output is good. Opus 5.5 does the hours.
    so i ran Fable 5.2 blind against Opus 5.5 on 10 kinds of my real work, on my own benchmark, BuilderBench. i made my bet before i saw a score. the result is at 10:35.
    the practical: in Claude Code, run /model fable for the main chat, then set your subagents to opus.
    also this week: Fable resets and the catch, Grok Bot resetting weekly limits, OpenAI's dot running on several of your computers, Meta's Muse Ember, and the Jensen Huang, Nvidia and Megaspeed story (reported by the FT and the NYT, Nvidia says it was investigated and debunked, nothing proven).
    Threadify is my lead generation agent for Threads, X and YouTube. it reads the replies on your posts and flags the people showing real buying signal, so you spend your attention where it counts, and nothing goes out until you say yes. free 7-day trial:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0040&utm_content=sponsor&video_slug=lab-0040&cta_slot=podcast_description&entry_angle=lead_signal
    0:00 the one job i'd pay Fable 5.2 for
    0:19 Fable 5.2 is the boss, Opus 5.5 the workers
    3:01 this is the LAB
    3:17 the thread: AI just got a pay scale
    3:50 Fable resets have a catch
    5:22 Grok Bot resets weekly limits
    6:22 today's sponsor: Threadify
    7:06 OpenAI's dot runs on several of your computers
    8:19 Meta's Muse Ember
    9:38 Jensen Huang, Nvidia and the Megaspeed story
    10:35 the bench: Fable 5.2 vs Opus 5.5, blind
    12:37 the catch + outro
    watch the full episode: https://youtu.be/rqeKU4Fr_pQ
    13 min
  • i let claude dream over my agents' memory. here's what it changed | LAB 0039
    i let claude dream over my agents' memory. i gave it six months of my own agent logs, then checked every change it made against my real log. my agent's notes went from 7 to 9 out of 10 on a quiz i wrote before the run. of 125 edits, 54 held up, 58 held up only partly, 2 did not hold up, and 11 had no log day to check.
    the catch: this is my own version of dreaming, not anthropic's feature (i don't have access yet). it was one run, 10 questions, and claude graded its own homework. so read 7 to 9 as a sketch of the idea, not a benchmark.
    the practical: tell your agent to read its own notes and logs between jobs, then ask what's wrong and what it can do better. that is all dreaming is.
    also this week: Claude Code cloud sessions, Claude Duet, Grok Bot joining meetings, Meta's Muse Wonder memory, and Nous Research's Arete-27B.
    Threadify is my lead generation agent for Threads, X and YouTube. it remembers how you write, drafts your posts, threads and replies so they already sound like you, and nothing goes out until you say yes. free 7-day trial:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0039&utm_content=sponsor&video_slug=lab-0039&cta_slot=podcast_description&entry_angle=voice_brain
    0:00 i let claude dream over my agent's memory
    0:15 what happened: dreaming on six months of my logs
    1:31 what anthropic actually reports
    2:48 why this matters
    3:51 the thesis: memory makes agents useful
    4:10 Claude Code cloud sessions
    5:17 Claude Duet
    6:28 today's sponsor: Threadify
    7:29 Grok Bot joins meetings
    8:43 Meta's Muse Wonder memory
    10:05 Nous Arete-27B
    11:27 the practical: dream over your agent's notes
    12:38 wrap-up
    watch the full episode: https://youtu.be/4goanD5lGAo
    14 min
  • i tested the Claude Code skill that reverse-engineers any viral video | LAB 0038
    i tested the Claude Code skill that reverse-engineers any viral video. i fed it my best edit and checked every cut it found against my real edit file. in the first minute it found 7 of my 14 jump cuts, and 15 of the 17 cuts it flagged were real. over the whole 14 minutes it caught about 1 in 3.
    the catch: the full reverse skill needs HyperFrames Pro ($25 a month), and it can tell you what i did, not why people watched. that gap is the one thing it can't see - stick around, i show you what it is at the end.
    the practical: i ran Cloudflare's free security audit skill on my own Mac app, Eddy Studio. read-only, about 23 minutes. 0 confirmed, 3 leads to test, 2 ruled out. it only read the code, so zero confirmed means not proven yet - not safe.
    also in this episode: Claude Personal Finance (US beta, read-only) and Pompliano's story, Cursor building live apps you can share with a link, Perplexity Teammates, and METR on AI speeding up AI research about 2x.
    Threadify is my lead generation agent for Threads, X and YouTube. it drafts your posts, threads and replies so they already sound like you, schedules them at your best time, and nothing goes out until you say yes. free 7-day trial:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0038&utm_content=sponsor&video_slug=lab-0038&cta_slot=podcast_description&entry_angle=post_engine
    0:00 i tested the skill that reverse-engineers any viral video
    0:12 7 of 14 cuts found
    2:13 this is Lenny's AI Builders
    2:28 AI can now pull apart anything that works
    2:49 Claude Personal Finance
    4:35 Cursor builds live apps
    5:46 today's sponsor: Threadify
    7:02 Perplexity Teammates
    7:58 Cloudflare's free security audit skill
    9:14 METR: AI speeds up AI research about 2x
    10:26 the practical: a free security audit on my own app
    11:37 the one thing it can't see
    watch the full episode: https://youtu.be/1YF5B9dOPAs
    13 min
  • grok bot can make phone calls now. but there's a catch | LAB 0037
    Grok Bot can now make phone calls. so i asked it to ring a hotel in New York and haggle the price. it showed the call card, said it was dialing the New Yorker... then blocked. the call was never even dialed.
    the catch: Grok Bot can only make calls while you're in the US. if you're not, there's no working around it. when it does work, you just ask in chat - it rings the number, talks to a person and tells you what happened. one early user on X said an 85-second call moved his usage from 5.8% to 6.0% (his plan isn't known, so take it with a pinch of salt).
    the practical: i told Claude to save a memory to Notion Memories, then asked ChatGPT to read it back. it worked - ChatGPT found the memory and listed its four rules. but Notion labelled it "created by ChatGPT", even though Claude wrote it. the trick is to be explicit: tell your agent "add this to Notion Memories" or "check my Notion Memories".
    also in this episode: Claude Code background sub-agents now use less of your limit, Notion's one shared brain for your agents, ChatGPT Images 2.6 (plus an AI Lenny test), Kimi K3.1 Pro open weights and what it would take to run them on Macs, and Cat Wu leaving Claude Code for OpenAI.
    Threadify is my lead generation agent for Threads, X and YouTube. it drafts your posts, threads and replies so they already sound like you, and nothing goes out until you say yes. free 7-day trial:
    https://www.threadify.app/home?utm_source=podcast&utm_medium=audio&utm_campaign=lab-0037&utm_content=sponsor&video_slug=lab-0037&cta_slot=podcast_description&entry_angle=voice_brain
    0:00 Grok Bot can make phone calls now. mine was blocked
    2:15 this is Lenny's AI Builders
    2:30 AI just walked out of the chat box
    2:56 Claude Code background sub-agents now use less of your limit
    4:14 Notion gives your agents one shared brain
    6:13 today's sponsor: Threadify
    6:59 ChatGPT Images 2.6, and an AI Lenny test
    8:23 Kimi K3.1 Pro open weights
    9:38 Cat Wu leaves Claude Code for OpenAI
    11:10 the practical: one Notion brain, Claude and ChatGPT
    14:19 see you next time
    watch the full episode: https://youtu.be/4S05I4s0-I0
    15 min