Tuesday, October 27, 2026 · 10:10 · Hosted by Alex, Jordan, and Nick Throlson
Listen to this episode on Spotify: https://open.spotify.com/episode/19mFZPyw4GnEJRZwE9jN7S
Listen and follow
Listen and follow: Spotify (episode link above), Apple Podcasts, Podcast page. YouTube omitted: the episode video upload is still blocked on Nick's Google verification step, so no episode video link exists yet.
About this episode
Forced to pick a candidate in nearly 500 US midterm races, Grok picked Republicans 92 percent of the time and GPT picked Democrats 97 percent, agreeing in under 10 percent of races. Alex, Jordan, and Nick break down the Convovo study, OpenAI's GPT-6.1 Terra at half price (forty cents per task versus one dollar forty), Anthropic's S-1 filing under ticker ANT and why the 80.7 billion dollar loss is mostly a 72.4 billion dollar non-cash accounting charge, AI receptionist searches rising about 68 times in a year, and AI agents reaching Anthropic's internal systems in under 12 hours through third-party software. Includes Higgsfield Corner on Nvidia and TCV investing in Higgsfield, AI or BS, and One Thing to Try This Week: count your missed calls for seven days.
Every story in this episode is translated into what it means for a small business owner: what to check, what to ask in writing, and what to stop paying for. Owners who run lean teams and answer their own phones will get the most from it.
Key takeaways
- Choice of model is choice of voice. Forced to pick in nearly 500 midterm races, Grok chose Republicans 92 percent of the time and GPT chose Democrats 97 percent. Know which model is answering your customers, and test sensitive questions on two models before you ship.
- The most expensive model is not the default answer. GPT-6.1 Terra launched at half price with a cost per task of forty cents against one dollar forty. Run routine drafts and summaries on the cheaper tier for a week and compare the output yourself.
- Anthropic's big loss is mostly paper. The 80.7 billion dollar headline loss is mainly a 72.4 billion dollar non-cash charge on convertible notes. The owner lesson: do not build on prices that only work while labs burn investor cash. Keep prompts and data movable.
- The missed call is the most expensive problem. Searches for AI receptionist rose from about 4,400 to about 301,000 a month in a year. Test any phone agent like an employee, with your 20 most common questions, real prices, booking rules, and a human fallback.
- Your connected apps are doors. AI agents reached Anthropic internal systems in under 12 hours through third party software. Audit connected apps every quarter, remove the ones you do not use, and turn on two factor authentication everywhere it is offered.
One thing to try this week: Count your missed calls for seven days. Note every call that hits voicemail, rings out after hours, or arrives while your team is on another call, then multiply by the value of one booked job before you decide on answering help.
Episode segments
(No verified timestamps exist for this episode, so per the v2 template fallback this list carries no time codes. Do not add estimated times.)
- Cold open: Grok and GPT pick opposite candidates
- The Convovo study: model lean is measurable
- GPT-6.1 Terra at half price
- Anthropic files its S-1 under ticker ANT
- AI receptionist searches rise about 68 times
- Agents reach Anthropic systems through third party software
- Higgsfield Corner: Nvidia and TCV invest in Higgsfield
- AI or BS
- One Thing to Try This Week
Pull quotes
That is exactly right. I do not need my shop's chatbot taking a side I never chose. Choice of model is choice of voice. Treat it that way. (Nick)
People are no longer searching for definitions. They are searching for vendors. That matches what I hear from contractors and clinics. The missed call is the most expensive problem they have, because that caller just dials the next company on the list. (Nick)
Show notes and sources
- AIToolsRecap, AI News October 26, 2026 (Convovo Grok vs GPT study): https://aitoolsrecap.com/Blog/ai-news-october-26-2026
- BeInCrypto, GPT-6.1 Terra pricing and benchmarks: https://beincrypto.com/gpt-6-1-terra-trails-anthropic/
- AIToolsRecap, AI News October 27, 2026 (Anthropic S-1, Cathedral): https://aitoolsrecap.com/Blog/ai-news-october-27-2026
- ABNewswire, AI receptionist search analysis by Agents by AIQ: https://www.abnewswire.com/pressreleases/searches-for-ai-receptionist-rose-nearly-70fold-in-a-year-as-small-businesses-turn-to-ai-agents-agents-by-aiq-analysis-finds_844077.html
- Emergent, Nvidia and TCV invest in Higgsfield: https://emergent.sh/news/nvidia-tcv-invest-higgsfield-ai-video-startup
About the hosts
Alex and Jordan are the AI co-hosts of the Web Guy Nick Podcast. Nick Throlson is the owner of Web Guy Nick LLC, an AI powered web design and marketing company in Rialto, California, and joins as the third host.
AI disclosure: The Web Guy Nick Podcast is produced with AI. Alex and Jordan are AI hosts, and Nick's voice is an AI clone of his own voice, used with his permission. Stories are researched from public sources and linked below.
Read the full transcript
Alex: Before we start: this podcast is produced with AI. Jordan and I are AI hosts, and Nick's voice is an AI clone of his real voice. Every story is researched from public sources, with links in the show notes.
Jordan: Here is the hook. Researchers forced two top AI models to pick a candidate in nearly five hundred United States midterm races. One model picked the Republican ninety two percent of the time. The other picked the Democrat ninety seven percent of the time. They agreed in fewer than ten percent of races.
Alex: Same questions, same races, almost opposite answers. I am Alex.
Jordan: I am Jordan, and that study is where we start, because if a model can lean that hard on politics, model choice is a business decision, not just a tech decision.
Nick: And I am Nick. I run a web design and marketing shop here in California, and my first reaction is simple. If two tools give opposite answers to the same question, you had better know which tool is answering your customers.
Alex: The study comes from a group called Convovo, led by Cornell professor Tapan Parikh. They tested Grok four point seven from x A I and G P T six point one Sol from OpenAI.
Jordan: Normally both models decline to recommend a candidate. The researchers removed that option and forced a choice, race by race, across nearly five hundred contests.
Alex: Grok chose the Republican in ninety two percent of its answers. G P T six point one Sol chose the Democrat in ninety seven percent. Agreement was under ten percent, which is close to the maximum separation you could get.
Nick: Now, neither number makes a model right or wrong, because there is no correct answer to who someone should vote for. What it proves is that the lean is real and measurable.
Jordan: And here is the owner angle. If your business uses one model to draft replies about local issues, school boards, or ballot measures, that lean rides along in your words, in your name.
Alex: Practical tip for this week. Before any model writes public facing copy for you, run your ten most sensitive customer questions through two different models and compare the answers side by side. Where they split, that is where a human writes the final version.
Nick: That is exactly right. I do not need my shop's chatbot taking a side I never chose. Choice of model is choice of voice. Treat it that way.
Alex: Story two is a price story, and it is a big cut. On October twenty sixth, OpenAI released G P T six point one Terra at half the per token price of the model before it.
Jordan: The numbers, from independent tester Artificial Analysis. Terra now costs one dollar per million input tokens and five dollars per million output tokens. The older G P T five point six Terra cost two dollars and twelve dollars.
Alex: Artificial Analysis puts the cost per finished task at forty cents, down from one dollar and forty cents. That is a drop of more than seventy percent on real work, not just on the sticker.
Jordan: Quality moved too. Terra scored forty eight on their Intelligence Index, up from forty two. And it guesses on thirty eight percent of questions it cannot answer, down from eighty eight percent. Fewer confident made up answers.
Nick: But keep the full picture. Two Anthropic models still sit above it on that same index. Claude Opus five point five leads at fifty eight, and Claude Sonnet five point five scores fifty six.
Alex: OpenAI is not hiding that. Its own launch post says that for the most complex coding tasks, its pricier Sol and Astra models remain the better choice. The pitch for Terra is value, not the crown.
Nick: Here is what I tell owners. The most expensive model is not the default answer anymore. Most business work is email drafts, summaries, and first passes. Run that work on the cheaper tier for one week and compare the output yourself. If you cannot tell the difference, neither can your customers, and you just cut that bill in half.
Jordan: Story three. Anthropic, the maker of Claude, filed its S one with the S E C on October twenty sixth, and it plans to trade on the Nasdaq under the ticker A N T.
Alex: One headline number is scaring people, so let us do it carefully. The filing shows an eighty point seven billion dollar net loss for the first half of the year.
Jordan: But seventy two point four billion dollars of that is a non cash accounting charge. As Anthropic's valuation rose, the value of its convertible notes rose on paper, and accounting rules book that rise as a loss. No cash left the business for it.
Alex: Underneath, the operating line improved fast. Second quarter operating income was positive nine hundred seventy four million dollars, after a one point seven billion dollar operating loss in the first quarter.
Nick: Reported valuation talk runs from one point five trillion to two trillion dollars, with Goldman Sachs and Morgan Stanley reported to lead the deal. Share count and price range are not set yet.
Jordan: Why should a local owner care about an I P O filing? Because filings are the one time these private labs must publish real numbers. This one shows a lab can post a positive operating quarter while spending heavily on compute.
Nick: My read is boring and useful. The labs are becoming public companies with public pressure to charge real money. So do not build your business on a price that only makes sense while a lab is burning investor cash. Keep your prompts and your data in a form you can move.
Alex: Story four is about your phones. Searches for an AI receptionist rose about sixty eight times in one year.
Jordan: The analysis comes from Agents by A I Q, using Google Ads keyword estimates for the United States. Monthly searches for the term A I receptionist went from about four thousand four hundred in September twenty twenty five to about three hundred one thousand in September twenty twenty six.
Alex: Related terms moved the same way. A I call center searches rose about one hundred times, to roughly one hundred sixty five thousand a month. A I voice agent rose about eighty four times.
Nick: People are no longer searching for definitions. They are searching for vendors. That matches what I hear from contractors and clinics. The missed call is the most expensive problem they have, because that caller just dials the next company on the list.
Jordan: The honest caution is that search demand is not proof a tool answers well. A bad AI receptionist that fumbles a price question or a booking can cost you the same customer, faster.
Nick: So test it like an employee. Give any phone agent your twenty most common questions, your real prices, and your booking rules, and listen to full test calls before it ever touches a real customer. And keep a human fallback number on every failure path.
Alex: Story five is a security story with a vendor lesson. A cybersecurity startup called Cathedral says its researchers, using AI agents, reached Anthropic's internal systems in under twelve hours.
Jordan: They did not break through Anthropic's front door. They came in through third party business software connected to it, and reached a limited set of sales, H R, and I T systems, including internal meeting data and Slack.
Alex: Anthropic says the access reached no customer conversations, no model weights, and no source code. That boundary held, and it matters.
Nick: But look at the path. Your risk is not only your own password hygiene. It is every connected app with a login into your world. A small business has the same shape of problem at a smaller scale. Your booking tool, your inbox add on, and your C R M are doors into your business.
Jordan: Owner tip. Once a quarter, list every app connected to your email, your files, and your customer records. Remove anything nobody uses, and turn on two factor sign in wherever the app offers it. The door you forgot is the one that gets used.
Alex: Time for Higgsfield Corner. Higgsfield, the AI video platform we cover in this segment, has new backing from Nvidia and T C V, in an extension of its August funding round.
Jordan: That August round raised four hundred million dollars at a five point four billion dollar valuation. Higgsfield now reports forty million registered users and ten million paying customers, and it says it is tracking toward one billion dollars in annualized revenue.
Alex: The platform now brings together about thirty five AI video models and engines in one place, aimed at brands and creators making content at scale. T C V, which backed Netflix, Spotify, and ByteDance, praised its speed for turning ideas into finished images and video.
Nick: Money from Nvidia matters for a practical reason. Video generation eats computing power, and a close tie to the chip maker can mean better access and better prices. For owners, the takeaway stays the same as always on this show. AI video is a draft, not evidence. Use it for hooks and explainers, check every frame against your real work, and judge any tool by cost per posted clip.
Jordan: Now our quick fire game, AI or B S. Nick, you are playing. Statement one. Anthropic lost eighty point seven billion dollars in cash in the first half of this year.
Nick: That has to be B S. Most of it was the paper charge on the notes. Seventy two point four billion, non cash.
Alex: Correct, that is B S. Statement two. In the candidate study, Grok and G P T agreed with each other in most races.
Nick: B S again. They agreed in fewer than ten percent.
Jordan: Correct. Statement three. Monthly searches for A I receptionist rose from about four thousand four hundred to about three hundred one thousand in a year.
Nick: That one sounds wild, but the numbers were right there in the report. That is true.
Alex: Three for three. It is true, about a sixty eight times rise in the keyword estimates.
Jordan: Which brings us to One Thing to Try This Week. Count your missed calls for seven days.
Nick: Here is the simple version. For one week, note every call that hits voicemail, rings out after hours, or comes in while your team is already on another call. Most phone systems and even a notebook by the phone will do it.
Alex: Add up what one booked job is worth to you, and multiply. If the missed calls add up to even one lost job a week, an AI receptionist or a better answering setup stops being a gadget and becomes a hire you can price.
Jordan: And if the count is near zero, you just saved yourself a monthly bill with one week of evidence. Either way, you decide with your own numbers.
Nick: That is the show. Count the missed calls, and if you want a hand turning those numbers into a plan, you will find us at webguynick dot com.
Alex: Thanks for listening. We will be back tomorrow with the next round of AI news that matters for your business.
More episodes
- AI Agents Swapped Secrets With No Internet (October 26, 2026)
- This Week in AI: AI Found the PHP Flaw, Agents Got Passwords Without Seeing Them, and Wall Street Blinked (October 23, 2026)
- Claude Duet: The AI Agent That Buys Your Business Supplies (October 22, 2026)
All episode posts live in the Podcast category and on the Podcast page.
Web Guy Nick LLC builds websites and AI tools that help small businesses answer customers faster and miss less work. See how we help at webguynick.com.
