End-of-world AI headlines have you wondering whether to keep investing? Keep going, but watch the cost. Tom, Mike and Mark cover Tom's Waymo rides, Grok Bot's three-day training, cheap Chinese models. Mark calls labs' doom talk self-serving; Tom cites $1,000/month in subscriptions and API calls for a one-person operation, while DeepSeek and Kimi K3 run on your hardware.
Highlights
- Riding a Waymo twice — including a tight LA street — turned a skeptic into a believer.
- Chinese models like DeepSeek Flash and Kimi K3 are fast, cheap and already in daily use.
- End-of-world warnings from AI labs read as self-serving cover for a first-past-the-post race.
- Route routine prompts to cheaper models; 99.9% of work doesn't need the newest model.
- One-person AI subscriptions and API calls now top $1,000 a month, and prices keep rising.
- Home-built tools are retiring WordPress hosting, Zapier and N8N subscriptions one by one.
Important Concepts and Frameworks
- Grok Bot — Grok's bot platform that runs multiple bots inside your own private zone, working together on tasks.
- DeepSeek Flash — Chinese model Tom uses daily, built to need less memory because chips are scarce.
- Kimi K3 — Chinese model praised for development work and called unbelievably cheap at token level.
- Prompt Balancing — Sending certain prompts to cheaper models instead of defaulting everything to the newest, priciest one.
- Orchestrator Model — A premium model told to do no work, only pick and hand off tasks to cheaper sub-agents.
- First Past the Post — Mark's frame for the AI race: whoever gets there first could control every system and network.
- Open-Weight Models — Models you download and run on your own hardware, like Tom's Mac Mini, with no cloud call.
Tools & Resources Mentioned
- Waymo — Driverless ride-hailing service Tom rode twice in Phoenix; it also cleared a tight LA street.
- Grok Bot — Bot platform pushed through a three-day money-making training, with a $200 commitment ask.
- DeepSeek Flash — Chinese model that is extremely fast and designed to run with far less memory.
- Kimi K3 — Chinese model Tom uses for development work; Mark's friend calls it really good.
- Fable 5.1 — Newest model Tom loads as an orchestrator rather than as a worker.
- Netlify — Hosting Mike used to deploy his event site until he ran out of credits.
- Zapier — Automation service Tom is down to two flows from, saving $300 a year.
- n8n — Workflow automation tool costing roughly $450 to $500 a year that Tom is retiring.
Calls to Action
- Tally your monthly AI subscriptions and API spend before your next planning session.
- Inventory where work is hard or stalled, then map AI against those specific points instead of chasing tools.
- Test one cheap or open-weight model on real work before assuming you need the premium tier.
- Set a premium model as orchestrator and route the heavy lifting to cheaper sub-agents.
- Put the youngest people in your company in the room when you set AI direction.
Key Quotes
- "It thought about it for a second, and then it totally did it." — Mark Redgrave
- "DeepSeek's Flash version, which just came out, is extraordinarily good. It's extraordinarily fast." — Tom Adams
- "To me, this is a completely self-serving false narrative, right?" — Mark Redgrave
- "I run an open weight model on my Mac Mini right now, and it works." — Tom Adams
- "The absolute truth is 99.9% of us do not need the latest fable, right?" — Mark Redgrave
- "There's a long list of things that could end the world. This might be another thing on the list." — Mike Richardson
What is Prompt and Circumstance?
It’s human-friendly banter about code, culture, and CEO reality checks—served up by Mike Richardson, Ryan Niemann, Mark Redgrave, and Tom Adams. No jargon. No hype. Just real talk from four guys who’ve seen it all, and aren’t afraid to say what everyone’s thinking.