When Your Agents Are Running but Nobody's Thinking

Does AI help experienced workers more than young employees? Kevin Williams says 20+ years of judgment and taste beat being two years out of college.Kevin Williams, founder of Ascend Labs, and co-host Matt talk through what Kevin heard from about 15 mid-market CEOs in Bozeman, Montana. They cover why leadership literacy comes first, why every company needs one named AI champion, how long-running agents handle slow bureaucratic work, and the cost of managing a dozen agents at once.In this episode:- Why staff pushback is showing up even at companies with no plans to cut jobs- The case against bringing personal AI accounts to work- Why Kevin sees more AI resistance from young people than older people- Naming a champion when you have no CTO, CIO, or MSP- Kevin's agent pursuing workforce training registration in all 50 states- Goldman Sachs restricting casual AI use over cognitive surrenderThanks to our Sponsors!AscendAI: https://ascendlabs.aiRapidDev: https://rapidevelopers.com/LiberNovo: https://libernovo.com/Try Muse, get a billion tokens! https://muse.ai/join (use code RHYQ60  in settings)CHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: https://www.linkedin.com/in/kevinguywilliams/Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the https://resources.ascendlabs.ai/openai-vs-anthropic-for-teams...Book a Call: Book a call...CHECK OUT MATT’S STUFF:LinkedIn: https://www.linkedin.com/in/matt-graham-nocode/Company: https://www.linkedin.com/company/rapid-dev/Direct Email: Reach out to Matt at [email protected]

AI agent governance and the bitter lesson: why the rules you write for your AI agents today won't last, plus email falling from 3% to 0.3% response.Kevin Williams and co-host Matt get into what happens when agents start talking to other agents and who decides what they're allowed to share. Kevin connects it to the bitter lesson and to the N8N workflows Claude Code wiped out in January. They also cover security for a bot-driven company and what drives AI adoption inside a team.In this episode:- Why Matt keeps some bots locked so only he can command them- The bitter lesson, and Google's healthcare models losing to general frontier models- Email response rates dropping from about 3% to about 0.3%- Kevin's four security controls: 2FA, limited cards, capped API spend, code words- A client office where everyone had Dot open one week after rollout- Matt's twenty-minute, one-presenter weekly show and tellThanks to our Sponsors!AscendAI: https://ascendlabs.aiRapidDev: https://rapidevelopers.com/LiberNovo: https://libernovo.com/Try Muse, get a billion tokens! https://muse.ai/join (use code RHYQ60  in settings)CHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: https://www.linkedin.com/in/kevinguywilliams/Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the https://resources.ascendlabs.ai/openai-vs-anthropic-for-teams...Book a Call: Book a call...CHECK OUT MATT’S STUFF:LinkedIn: https://www.linkedin.com/in/matt-graham-nocode/Company: https://www.linkedin.com/company/rapid-dev/Direct Email: Reach out to Matt at [email protected]

How much should you trust AI agents to make decisions? Kevin Williams explains why human-in-the-loop reviewers make bad bot sitters, and Matt runs Codex with full autonomy.Kevin Williams and Matt talk through the agentic tools they use every day, from Dot in Slack to a dial-in number Kevin uses to talk through his workday. They argue about guardrails, YOLO mode, and when it makes sense for a team.In this episode:- Why a human reviewing fragments of an automation can't give a meaningful thumbs-up- Matt's setup that tunnels Dot directly to Codex, credit cards included- Kevin's advice on virtual credit cards for agents- Running an hour a day by voice, including sending NDAs- A solo Harvard MBA who coordinated 11 external labs with Claude, and why Anthropic bought him- The risk of restructuring around a small AI-built SaaS vendorThanks to our Sponsors!AscendAI: https://ascendlabs.aiRapidDev: https://rapidevelopers.com/LiberNovo: https://libernovo.com/Try Muse, get a billion tokens! https://muse.ai/join (use code RHYQ60  in settings)CHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: https://www.linkedin.com/in/kevinguywilliams/Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the https://resources.ascendlabs.ai/openai-vs-anthropic-for-teams...Book a Call: Book a call...CHECK OUT MATT’S STUFF:LinkedIn: https://www.linkedin.com/in/matt-graham-nocode/Company: https://www.linkedin.com/company/rapid-dev/Direct Email: Reach out to Matt at [email protected]

How to fix your business before you automate it with AI. Sabir Semerkant exited three brands from his program for skipping that step.Kevin Williams sits down with Sabir Semerkant, founder and CEO of Growth by Sabir, who has worked in direct-to-consumer ecommerce since the field was new. They get into why AI scales bad processes and bad data, the multi-model audit workflow Sabir runs in Claude Code, and how to test new models without breaking production.In this episode:- Why Sabir exited about three brands from his Rapid2X program- Garbage in, garbage out, and why wrong attribution data gets dangerous- An orchestrator, Haiku workers, and two read-only auditors scoring work over three rounds- Why Sabir pays monthly for his $100 plans and skips the annual discount- The Meta and Gemini tests that failed in his dev environment- Kevin's 600 hard-coded references to Sonnet 4.6 and the config file fixThanks to our Sponsors!AscendAI: https://ascendlabs.aiRapidDev: https://rapidevelopers.com/LiberNovo: https://libernovo.com/Try Muse, get a billion tokens! https://muse.ai/join (use code RHYQ60  in settings)CHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: LinkedIn... –Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the https://resources.ascendlabs.ai/openai-vs-anthropic-for-teams...Book a Call: Book a call...CHECK OUT SABIR’S STUFF:Website: https://growthbysabir.com/

How to start using AI agents in your business, and why taste matters more than tools. Plus the Charleston room where all 17 CEOs leaned in.Kevin Williams and Matt compare notes on agent setups, why two people automating the same job get wildly different results, and what business owners in paving and HVAC are finally doing about AI. Kevin also shares where he tells every company to start.In this episode:- Why an agent built by someone with taste beats one built by a beginner- Who to pair with your taste people, and why they rarely have the technical skills too- Why large organizations shouldn't let everyone build their own AI solutions- 17 CEOs in Charleston, all engaged with AI for the first time- Kevin's two starting points: expensive human mistakes and misused talent- Living with ChatGPT's Dots, long-tail tasks, and 2FA headachesThanks to our Sponsors!AscendAI: https://ascendlabs.aiRapidDev: https://rapidevelopers.com/LiberNovo: https://libernovo.com/Try Muse, get a billion tokens! https://muse.ai/join (use code RHYQ60  in settings)CHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: LinkedIn... –Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the https://resources.ascendlabs.ai/openai-vs-anthropic-for-teams...Book a Call: Book a call...CHECK OUT MATT’S STUFF:LinkedIn: https://www.linkedin.com/in/matt-graham-nocode/Company: https://www.linkedin.com/company/rapid-dev/Direct Email: Reach out to Matt at [email protected]

How to build with AI when you're not a full-time developer, plus OpenAI's Dot agent doing a month of work in a day. Kevin Williams and AI developer Vlad Larin.Vlad Larin went from managing a 250-person delivery team to building with AI himself, and he argues the skill that separates good AI builders from bad vibe coders is the same one that separated good developers from bad ones: vision, architecture, and knowing what the product is supposed to do. Kevin Williams and Vlad compare build setups, test OpenAI's Dot, and talk about what AI subscriptions are likely to cost.In this episode:- Why a decade of untangling client requirements translates to building with AI- Vlad's setup: Codex plan mode, GitHub, Supabase, and Vercel- Gaslighting Codex as an "unofficial turbo mode" (and when it backfires)- Dot catching a missing meeting invite and bills due, unprompted- Kevin's 100-ad Meta campaign run hands-off with computer use- Why $500 a month is justifiable and $5,000 would stunt the marketThanks to our new Sponsor LiberNovo and their insanely supportive Omni Chair:  https://libernovo.com/To try out Muse and get a BILLION tokens use code: RHYQ60  https://muse.ai/join (full disclosure, we will get a billion tokens too… but aren’t sure how we will use them)Ready to find out where your organization actually stands on AI readiness?https://assessment.ascendlabs.ai/Want to talk through your AI tool decisions before you commit?tidycal.com/kevinwilliamsCHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: LinkedIn... –Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the https://resources.ascendlabs.ai/openai-vs-anthropic-for-teams...Book a Call: Book a call...VLAD LARIN:LinkedIn: https://www.linkedin.com/in/vlad-larin/

AI token costs in 2026: why Luna 6 at 40x cheaper than Fable may end the cost panic, and what Codex computer use actually did in four hours.Kevin Williams and Matt talk through the state of agentic AI right now: the jump from OpenClaw and Hermes to Grokbot and computer-use models, what Meta's Muse is actually for, and why Google paid $10 million for a dead airline's Slack messages.In this episode:- Kevin runs out of Codex tokens for three days and explains the $100 vs $200 plan trap- Luna 6 ships with computer controls at 40x less per token than Anthropic's Fable- Codex logs into Tag Manager, Analytics, Search Console, AdWords, Slack and HubSpot and finishes a week of campaign setup in four hours- Why Kevin refused to give Muse his Amex login, and what it did to his kids' Canvas account- Google buying Spirit Airlines' operating data to build a sandbox digital airline- Whether Anthropic's inference demand in early 2026 cost it the training raceThanks to our new Sponsor LiberNovo and their insanely supportive Omni Chair:  https://libernovo.com/To try out Muse and get a BILLION tokens use code: RHYQ60  https://muse.ai/join (full disclosure, we will get a billion tokens too… but aren’t sure how we will use them)Ready to find out where your organization actually stands on AI readiness?https://assessment.ascendlabs.ai/Want to talk through your AI tool decisions before you commit?tidycal.com/kevinwilliamsCHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: LinkedIn... –Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the https://resources.ascendlabs.ai/openai-vs-anthropic-for-teams...Book a Call: Book a call...CHECK OUT MATT’S STUFF:LinkedIn: https://www.linkedin.com/in/matt-graham-nocode/Company: https://www.linkedin.com/company/rapid-dev/Direct Email: Reach out to Matt at [email protected]

An Anthropic researcher went public with a 10% chance that AI does away with humanity, implied it's a common feeling inside the company, and other researchers backed him up. Dario Amodei followed with an essay saying it's time to talk seriously about slowing down. Sam Altman chimed in, Bernie Sanders wants it banned outright, the administration is calling it a hoax, and there's an election in a few weeks.Kevin Williams and Matt skip the politics and go after the technology behind the panic: recursive self-improvement. The slope is the part nobody is planning around. Once a model can pull its own levers, improvement stops being a steady climb and becomes a rock tipping from potential energy to kinetic. Astra 6 to Astra 20 to Astra 2000 in hours, days, minutes. The 2015 prisoner's dilemma argument holds that the first lab there wins the whole game.Then they get practical. What an RSI-style loop looks like inside a business today, why constantly shifting models make solutions hard to build against, treating context as code, sharing reusable skills, and Kevin's cost work: a resources tool that ran $40 per update until he rebuilt it around an MCP, and thumbnail images at seven and a half cents each.Full episode: https://ascendlabs.ai/podcast/ep-42-recursive-self-improvement-end-of-the-world-or-an-opportunity/In this episode:- The 10% claim, who made it, and why the regulatory capture read is tempting- Why RSI improvement is exponential and what tipping over looks like- The national security version: what if China's model goes to Astra 2000 overnight- Building solutions on a model that quietly becomes Opus 17- Self-improvement loops applied to your own processes, not just the model- Why Kevin downgraded Claude Max from $200 to $100Ready to find out where your organization actually stands on AI readiness?https://assessment.ascendlabs.ai/Want to talk through your AI tool decisions before you commit?tidycal.com/kevinwilliamsCHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: LinkedIn... –Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the OpenAI vs. Anthropic comparison...Book a Call: Book a call...CHECK OUT MATT’S STUFF:LinkedIn: https://www.linkedin.com/in/matt-graham-nocode/Company: https://www.linkedin.com/company/rapid-dev/Direct Email: Reach out to Matt at [email protected]

Kevin Williams and Matt open Alex Lieberman's 30 Features of an AI Native Company and get through the first handful of items, starting with the one most organizations get wrong. Blueprinting your processes is the right first move. Replicating every existing step inside an AI system is not, because a model like Astra will happily reproduce a five-step handoff that was designed around people and tools that predate it.Before that, Kevin walks through what he actually built with Astra last week: a 45-page presentation deck assembled from client research, downloaded imagery and past decks; a technical SEO project that produced a new page, four updates and seven draft articles in about 45 minutes; and a part-based 3D printable model of a cabin he used to own in West Virginia, generated from two photos and optimized for multi-color printing.They also get into why the intelligence layer has no good off-the-shelf answer, why AI-first organizations that are smaller and founder-led can move when ISO-certified companies can't, and whether the sunk cost fallacy still applies to systems you built nine months ago.Topics covered:- The caveat on blueprinting processes: input and output, not five steps- Why every person in the organization needs a daily driver, plus a literacy layer- Building an intelligence layer across email, Slack, transcripts and documents- What Astra did with a deck, an SEO campaign and a 3D print- Why security-mature organizations struggle to move at all here- When to rebuild a working system and when to leave it aloneReady to find out where your organization actually stands on AI readiness?https://assessment.ascendlabs.ai/Want to talk through your AI tool decisions before you commit?tidycal.com/kevinwilliamsCHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: LinkedIn... –Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the https://resources.ascendlabs.ai/openai-vs-anthropic-for-teams...Book a Call: Book a call...CHECK OUT MATT’S STUFF:LinkedIn: https://www.linkedin.com/in/matt-graham-nocode/Company: https://www.linkedin.com/company/rapid-dev/Direct Email: Reach out to Matt at [email protected]

Kevin spent a rainy weekend running up to 10 simultaneous Codex agents across three internal builds and burned two billion tokens. On a consumer plan, propped up by repeated usage resets, that cost him about $125. At retail API rates on GPT 5.6 high, the same weekend runs north of $8,000 to $10,000 and enterprise teams building in Codex fall outside the consumer plan entirely.Matt and Kevin dig into what that gap means for anyone standing up internal AI infrastructure today. Matt's team built an internal system that captures every call, Slack message, and email on a client project and flags when something drifts. It runs affordably now. His concern isn't the build cost it's what happens if the run cost triples.They also cover why switching between Claude Code and Codex costs you almost nothing if your code lives in GitHub, why that undercuts the case for price increases, and what happened when Kevin turned a swarm loose on 45 repositories to fix a favicon.Topics:- Why AI avatars still fail for self-cloning after three years of trying- The 82/18 split between deterministic and agentic workflow steps- LinkedIn bans, bot-detectable IPs, and 200 profile lookups in five minutes- Codex sidebar chats, work trees, and browser control versus Claude Code- Two billion tokens, $125, and the $8,000 enterprise equivalent- The runaway agent that logged into an abandoned Lovable accountReady to find out where your organization actually stands on AI readiness?https://assessment.ascendlabs.ai/Want to talk through your AI tool decisions before you commit?tidycal.com/kevinwilliamsCHECK OUT KEVIN’S STUFF:Website: https://ascendlabs.ai/LinkedIn: LinkedIn... –Free Tool: Take the https://assessment.ascendlabs.ai/Deep Dive: Read the https://resources.ascendlabs.ai/openai-vs-anthropic-for-teams...Book a Call: Book a call...CHECK OUT MATT’S STUFF:LinkedIn: https://www.linkedin.com/in/matt-graham-nocode/Company: https://www.linkedin.com/company/rapid-dev/Direct Email: Reach out to Matt at [email protected]