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.
Anthropic priced the largest IPO in US history at $83 a share, above its $70 to $78 range, and starts trading today on Nasdaq under ANT. Today: Google ships Gemini 4 Neon at $3 per million input tokens. Citi, PNC and Citizens put third-party AI agent losses on the customer, even when the agent is hacked. OpenAI uses its models to improve chip designs by up to 14 percent. More than 100 agents help audit a listed company at KPMG. And Google opens Genie 4 interactive worlds to Ultra subscribers.
Show notes
- Anthropic IPO priced at $83 (Nov 18): 980 million shares for about $81.3 billion in gross proceeds, the largest US IPO on record according to Renaissance Capital. Anthropic itself sells about 956 million shares, trading begins November 19 under ANT, and the offering is expected to close November 20.
- Gemini 4 Neon (Google, Nov 18): The middle tier of the Gemini 4 family for coding, knowledge work and science, at $3 per million input tokens and $15 output, with a 1 million token context window and general availability in GitHub Copilot.
- Banks shift agent losses to customers: Citi, PNC and Citizens updated terms this week so third-party AI agent losses, even after a hack, sit with the account holder. JPMorgan and Bank of America made similar moves in October. Citi exempts its own agents.
- OpenAI designs chips with AI: Up to 14 percent improvement in power, performance and area on optimized AMD hardware, plus work on its Jalapeno inference chip. One multiplier redesign cut estimated area by about 15 percent and a calculation from two cycles to one.
- KPMG Project Exponential: More than 100 agents ran a parallel audit of an unnamed FTSE 250 company across every audit stage. They sometimes dug deeper than the human team, a human corrected a regulation-matching error, and the formal opinion still came from the conventional audit.
- Genie 4 opens to Ultra subscribers: Interactive worlds you can generate, explore, edit with text prompts during exploration, and hear responsive audio in, through Project Genie.
- Quick hit, ChatGPT for Accounting: A plugin for connected accounting apps to prepare invoices, reconcile records and draft financial reports, with human review.
- Higgsfield Corner: Short Drama Studio turns a script into a short series with consistent cast, locations and props across episodes. Early preview on desktop.
- One Thing to Try This Week: The 15-minute agent money rules setup.
Key takeaways
- Public markets force AI vendors to show real customer results every quarter. Borrow that discipline: judge every tool by cost, finished work produced, and your hand fallback.
- If a third-party agent can touch your money, you now carry the loss at major banks. Use a separate low-limit account, an approval threshold, and a weekly review.
- Match the model to the job. Neon-class middle tiers handle most coding and analysis without top-tier prices.
- Agents work best on the wide first pass. A named human still owns the judgment call and the signature.
- Improve one small, measurable job at a time, the same way AI improved one chip circuit at a time.
Nick’s Take
A trillion dollar valuation does not buy a single tool for your shop, and it does not answer a single customer call. What the listing does is force a new kind of honesty into this market. Public companies have to show, every quarter, whether customers keep paying and what the work actually costs to deliver. Before you renew any AI tool this month, write down three numbers: what it cost you, what finished work it produced, and what you would do by hand if it vanished tomorrow. If you cannot fill in the middle number, you do not have a tool, you have a subscription you hope is helping.
One Thing to Try This Week
Build your agent money rules in 15 minutes. One, use a separate account or card for agent activity, never your main operating account. Two, set a low hard limit and an approval threshold. Three, list which agents and tools have access today and remove any you no longer use. Four, put a weekly 10-minute review of agent activity on your calendar.
Sources
- The Neuron, Everything That Happened in AI Today, Thursday November 19, 2026
- AI Weekly, AI news for November 19, 2026
- Precise Impact AI, AI Daily Brief, 19 November 2026
- SXF Brief, AI News Nov 19, 2026
- Higgsfield changelog, Short Drama Studio, November 16, 2026
Find your next AI tool at BotPickr, our directory of more than 170 AI tools and bots: botpickr.com
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 for today. A company that sells A I software just priced the largest initial public offering in United States history, at eighty-three dollars a share, and it starts trading today. That is story one, and it changes how every A I vendor you buy from will have to prove its worth.
Alex: Welcome to the Web Guy Nick Podcast for Thursday, November nineteenth. I am Alex.
Jordan: And I am Jordan. With us, as always, is Nick, the owner in the room, who actually has to pay the A I bills we talk about.
Nick: And read the fine print on them, which is where our third story lives today. A major bank just told its customers that if a third party A I agent with access to their account gets hacked or makes a mess, the loss is on the customer, not the bank. Hold that thought if you are thinking about letting an agent touch your money.
Alex: Story one, the lead. Anthropic, the maker of Claude, priced its initial public offering at eighty-three dollars a share. That is above the seventy to seventy-eight dollar range it had guided to. The offering covers nine hundred and eighty million shares, for about eighty-one point three billion dollars in gross proceeds. Renaissance Capital calls it the largest I P O in United States history.
Jordan: The details matter here. Anthropic itself is selling about nine hundred and fifty-six million of those shares, with existing shareholders selling the rest. The registration statement was declared effective by the Securities and Exchange Commission on November eighteenth, shares are expected to begin trading on Nasdaq today under the ticker A N T, and the offering is expected to close on November twentieth, subject to the usual conditions.
Alex: On a fully diluted basis, the pricing values Anthropic at roughly one point four to one point five trillion dollars, depending on which analysis you use. Either way, this is no longer a private startup story. It is a public scoreboard, updated every trading day.
Nick: Here is why an owner should care about a stock listing. When a vendor goes public, it has to publish real numbers, revenue, costs, and losses, in filings anyone can read. The sales pitch gets quieter and the paperwork gets louder. That is good for buyers like us, because pressure from public investors rewards tools that show measurable results for customers, not just impressive demos.
Jordan: Story two. Google has released Gemini four Neon, the middle tier of the Gemini four family. Carbon handles the hardest long projects, Hydrogen handles fast everyday work, and Neon is built for software engineering, knowledge work, and scientific tasks that need depth without the top tier price.
Alex: Neon costs three dollars per million input tokens and fifteen dollars per million output tokens, with a one million token context window and four thinking levels. Google says that at its highest thinking level it ranks third on the Artificial Analysis Intelligence Index, with a score of sixty-one. It is also now generally available in GitHub Copilot.
Nick: The owner translation is simple. You now have three clear price lanes from one family. Use the fast cheap lane for sorting, summarizing, and first drafts. Use the middle lane, like Neon, for coding help and analysis you will actually check. Save the top lane for the one hard job a month where a wrong answer is expensive. Most shops I work with waste money by sending every job to the biggest model, like hiring an electrician to change a light bulb.
Nick: Break time, and I am spending it on my own site. If you have ever typed ‘best AI tool for’ into Google and drowned, BotPickr is the fix. More than 170 AI tools and bots, organized by the work they do: sales, support, content, scheduling, the lot. Pick the job, get a short list that actually fits, and skip the hype videos. Find your bot in 60 seconds at botpickr.com. Now back to Alex and Jordan, who never need a break, because they are robots.
Alex: Story three, and this is the fine print story Nick flagged. Citi, P N C, and Citizens have updated their customer terms this week to put losses from third party A I agents on the account holder, even if the agent is hacked. TechCrunch reported the change, and the three banks follow JPMorgan and Bank of America, which announced similar policies in October.
Jordan: Read that slowly, because the distinction is sharp. Citi exempts its own A I agents from that shift. So if the bank’s own tool makes the mistake, that is one conversation. If a third party tool you connected makes the mistake, the bank’s position is that you authorized it, so you carry the risk.
Alex: This lands right after two weeks of agent payment news on this show, from ChatGPT Wallet to Gemini Tasks buying on your behalf. The capability is arriving fast. The liability rules are arriving just as fast, and they point at the user.
Nick: My rule for clients is now very plain. Never give a third party agent direct access to your main business account. If you want to test agent payments, use a separate account or a card with a low hard limit, set an approval step for anything above a small amount, and review the activity weekly. An agent should hold a small wallet, not the keys to the whole shop. If a tool cannot work within limits like that, it is not ready for your money yet.
Jordan: Story four. OpenAI says its models helped design better computer chips, including its own Jalapeno inference chip and work with A M D on A M D hardware. OpenAI reports improvements of up to fourteen percent in power, performance, and area measures on optimized A M D hardware.
Alex: One concrete example makes this real. An A I generated redesign of a multiplier circuit cut the estimated chip area by about fifteen percent and reduced a calculation from two clock cycles to one. OpenAI is also working with Synopsys to bring A I into established chip design software that engineers already use.
Jordan: Why does chip design matter to a bakery or a roofing company listening in Rialto? Because cheaper, faster chips are what make the A I tools on your phone cheaper per task over time. When the machines that run A I get more efficient, the price of drafting an email, summarizing a call, or building a quote keeps falling.
Nick: And there is a second lesson. The win did not come from asking a model for a whole chip in one prompt. It came from giving the model one small, measurable part, checking the result, and keeping what improved the numbers. That is exactly how a small business should adopt A I. One job, one measure, keep it only if the number moves.
Alex: Story five. K P M G has run more than one hundred A I agents through a full parallel audit of an unnamed company listed on the F T S E two hundred and fifty index in the United Kingdom. The project, called Project Exponential, ran alongside the regular human audit team, and the agents worked across every stage of the audit, gathering evidence, analyzing information, and drafting conclusions.
Jordan: K P M G’s United Kingdom head of audit, Cath Burnet, said that in several cases the agents investigated issues more deeply, or asked questions the human team had not raised. K P M G plans to extend the trial to as many as twenty larger audits.
Alex: The limit showed up clearly too. In one case an agent could not correctly match a regulatory requirement to the company’s own documents, and a human auditor had to step in. And to be clear, K P M G did not rely on the experiment for the formal audit opinion. The signed accounts still came from the conventional audit.
Nick: That pattern is the healthiest one I have seen all month. Agents do the wide, tireless first pass. Humans own the judgment call and the signature. If you run a service business, copy that split. Let the agent pull the records, flag the odd items, and draft the summary. Keep a named person responsible for the final answer you send to a customer or a regulator. Speed without a signature is just risk moving faster.
Jordan: Story six. Google DeepMind has opened Genie four, its interactive world model, to eligible Google A I Ultra subscribers through Project Genie. You can generate a world, explore it, change the environment with text prompts while you are inside it, and hear responsive audio as you move.
Alex: Sessions run about one minute, and Project Genie adds an agent that helps create worlds plus a community gallery where people share what they built. This is not a business tool yet, but watch the direction. A world you can describe, walk through, and edit in plain language is a rough draft of how store layouts, training scenes, and product demos could be prototyped next.
Nick: For a local shop, the near term use is simple and practical. Imagine showing a customer a remodel, a venue setup, or a product in a scene you generated that morning, instead of describing it for ten minutes. Do not build your business on an Ultra subscriber preview, but do start collecting the photos and notes you would need to describe your space well, because the shops that can describe their work clearly will use these tools first and best.
Jordan: Quick hit, thirty seconds. OpenAI has introduced ChatGPT for Accounting, with an Accounting plugin for connected accounting apps. It is built to help prepare invoices, reconcile records, and draft financial reports, with setup guidance on access permissions and human review. Nick, your one line verdict?
Nick: Useful if the review step stays real. Let it draft and reconcile, keep a human approving anything that leaves the building, and it could save a bookkeeper hours a week. Skip the approval step and you have rebuilt the bank story from earlier, just with invoices.
Alex: That brings us to Nick’s Take, drawn from our lead story, the Anthropic listing.
Nick: Here is my take. A trillion dollar valuation does not buy a single tool for your shop, and it does not answer a single customer call. What the listing does is force a new kind of honesty into this market. Public companies have to show, every quarter, whether customers keep paying and what the work actually costs to deliver. As an owner, borrow that discipline for your own buying. Before you renew any A I tool this month, write down three numbers: what it cost you, what finished work it produced, and what you would do by hand if it vanished tomorrow. If you cannot fill in the middle number, you do not have a tool, you have a subscription you hope is helping. Across the forty-four websites I manage, the tools that survive are never the ones with the biggest headlines. They are the ones where the owner can point to a job that got done faster, cheaper, or better, and prove it in one sentence. Demand that proof from every vendor now, because the public market is about to demand it from them.
Jordan: Time for Higgsfield Corner. On November sixteenth, Higgsfield added Short Drama Studio, now in early preview on desktop. It turns a script into a short series. It splits the text into episodes, builds the cast, the locations, and the props, and generates the episodes with the same characters throughout.
Alex: Consistency is the hard part of A I video, and it is exactly what this is aimed at. The same character, in the same place, across multiple short episodes, without starting the casting problem over for every clip.
Nick: For a business owner, think series, not one off ads. A coffee shop could run a five part short series about its morning regulars. A contractor could run a short drama about the remodel that went wrong and then right. Short series are what people actually finish and share. If you try it, write the whole short script first, keep each episode tight, and check every face and logo before you post, because consistency on screen still needs a human eye off screen.
Jordan: Now our game, A I or B S. Three claims, Nick calls each one, and the stakes today are real. If Nick misses two, he has to personally review every agent payment setting on his own accounts this week, on camera for nobody, which he will hate.
Nick: I already review them, because unlike my co-hosts, I am the one the bank says is liable. But go ahead.
Alex: Claim one. Anthropic’s offering raised more than two hundred billion dollars in gross proceeds. A I or B S?
Nick: B S. The number was about eighty-one point three billion dollars. Still the largest I P O on record, just not two hundred billion.
Alex: Correct. Claim two. Gemini four Neon costs three dollars per million input tokens and supports a one million token context window.
Nick: That one is A I, meaning true. Three in, fifteen out, one million token window, and it is live in GitHub Copilot.
Jordan: Correct again. Claim three. Citi says customers are fully protected if a third party A I agent they connected gets hacked and moves their money.
Nick: B S, and it is the most important B S of the day. Citi, P N C, and Citizens put that loss on the customer. Only the bank’s own agents are exempt in Citi’s case. Read your terms before you connect anything to an account.
Jordan: Three for three. Nick keeps his week, and the rest of us learned that the only host who needs sleep is also the only host who reads terms and conditions for fun.
Alex: One Thing to Try This Week. Build your agent money rules, in fifteen minutes, before you connect an agent to anything financial. It has four short steps.
Jordan: Step one, pick the account. Use a separate account or card for agent activity, never your main operating account. Step two, set a low hard limit and an approval threshold, so anything above a small amount needs your tap.
Alex: Step three, write down which agents and tools have access today, and remove any you no longer use. Step four, put a weekly ten minute review on your calendar to read the agent activity, the same way you would read a bank statement.
Nick: Do those four steps once and you get the upside of agents shopping, booking, and reconciling for you, without handing a hacked tool an open line to your business. That is the whole lesson of today in one habit.
Alex: To close, a short recap. Anthropic priced the largest I P O in United States history at eighty-three dollars a share and starts trading today. Google shipped Gemini four Neon at three and fifteen dollars, and opened Genie four worlds to Ultra subscribers. Three more banks put third party agent losses on customers. OpenAI used its models to improve chip designs by up to fourteen percent. And more than one hundred agents helped audit a listed company, with a human still signing the work.
Jordan: Your one action from today is the agent money rules list. Fifteen minutes, four steps, and you are safer for every agent tool you try this year.
Nick: Do the fifteen minute setup this week. It is cheaper than one bad transaction. If you want a hand getting your site and your tools in order, you can find us at webguynick dot com. We will see you tomorrow.
