Tricks And Treats

Happy Halloween. The kids get the candy today. You get a newsletter about software pricing, which I'll admit is the box of raisins of trick or treating. Stick with me anyway. There's chocolate further down the bag.

This week the big labs kept fighting over your wallet. OpenAI cut prices with a new model, Google turned Gemini into something that hands you actual files, Microsoft pointed Copilot straight at your books, and Anthropic filed paperwork to go public. Meanwhile a French lab gave away a trillion parameter model, and an OpenAI backed holding company bought a home security brand you've seen on a hundred front lawns.

Six stories below. For each one you get what happened, why a small shop should care, and what I'd do about it before Monday.

Pour something brown. Let's get into it.

OpenAI launched GPT-6.1 Terra on October 26, a fast, efficient model it says is better at visual reasoning and using a computer. Per OpenAI's release notes, standard API prices are 50% lower than GPT-5.6 Terra: $1 per million input tokens, $5 per million output tokens and a nickel per million for cached input. In ChatGPT it's live for Plus, Pro, Enterprise and Edu users in Work and Codex, though not in regular Chat.

Two smaller changes landed the same week. iCloud users can now connect Mail and Calendar to ChatGPT on every plan, so it can find emails, check your schedule and update events. And on Enterprise and Edu, creating new custom GPTs is switched off in affected workspaces. Existing GPTs keep working until their retirement date, and OpenAI is steering people toward its Plugin Creator instead.

Why it matters. Half price on the API changes what's worth automating. A workflow that didn't pencil out at last month's rates might work fine at this month's. The iCloud piece matters for the huge number of owners who run the whole company off an iPhone and never moved to Google or Microsoft. And the custom GPT change is a warning shot. If your team's favorite internal tool is a custom GPT, it has an expiration date, even if nobody's mentioned it to you yet.

What I'd do.

Rerun the math on the jobs you shelved. Pull up the automation you killed for cost reasons, like summarizing every support ticket or drafting every quote. Price it again at Terra rates before you assume it's still too expensive.

Connect your calendar and ask it something real. If you live in iCloud, hook it up to ChatGPT and ask for every client email from the last month that's still waiting on a reply from you. That tells you more than asking it to write a limerick about your accountant.

Make a list of your custom GPTs. Write down what each one does and copy out the instructions inside it. When they retire, you'll want that text to rebuild them somewhere else.

Google spent the week turning Gemini from a chat window into something that hands you finished work. As of October 29, the Gemini app can create, preview and edit files right inside the chat: slide decks, spreadsheets, documents and PDFs. Decks open in Google Slides or download as PPTX or PDF, and spreadsheets open straight into Sheets. PDF editing works on web and mobile, and the other formats are web only for now. It's available on Workspace accounts and personal Google accounts alike.

Inside Sheets, Fill with Gemini now shows the web sources behind each cell it fills in and handles up to 3,000 cells in one request. And Gemini in Slides can now edit an entire deck from one prompt, restyling or condensing slides, with changes you can undo or review in version history. That one's on Business Standard and Plus, Enterprise Standard and Plus, and Google AI Pro and Ultra.

Why it matters. Most small businesses don't need an AI to chat with. They need the proposal, the price sheet and the client deck done by Thursday, and Google is aiming right at that. The citations in Sheets are the sleeper feature. When AI fills in 500 rows of supplier details or competitor pricing, the hard part is checking its work. Now every cell tells you where it got the answer.

What I'd do.

Build your next proposal in the Gemini app. Feed it last month's winning proposal plus your notes on the new client and have it produce the deck. Then open it in Slides and fix what it got wrong. You'll still edit, but you'll be editing a draft instead of staring at a blank page.

Test the citations on something you know cold. Run Fill with Gemini on a list you could recite in your sleep, like your top 20 vendors. Click the sources on five cells. If it's wrong about stuff you know, don't trust it on stuff you don't.

Clean up the old sales deck. One prompt can now bring a tired deck in line with your current branding. Do it on a copy first, because you'll want something to compare against.

On October 29 Microsoft announced Copilot for finance, adding more than 25 new finance skills and partner plugins across Copilot. The accounting features handle account reconciliations, journal entry prep and performance reporting, with an optional audit trail that records what Copilot changed and why. It can trace a workbook's change history and reverse individual edits. Partner plugins, Intuit QuickBooks among them, are starting to roll out through Microsoft's plugin registry.

Availability is a mixed bag. The finance skills in Copilot in Excel are generally available, while the accounting capabilities sit in Microsoft's Frontier early access program. Microsoft says that in seven days of limited Insiders testing, more than 8,000 people used the accounting skills over 44,000 times.

Why it matters. Month end close is where owners lose their weekends. Reconciliations are tedious and repetitive, and they live in Excel and QuickBooks. That's exactly the kind of work AI handles well, provided somebody checks it. The audit trail is the part I care about most. An AI that touches your books and doesn't show its work is a liability. One that logs every change and lets you roll back a single edit is something your bookkeeper might actually put up with.

What I'd do.

Talk to your bookkeeper before you flip anything on. They own the books. If Copilot is going to touch reconciliations, they should be the one reviewing its work, with the audit trail turned on from day one.

Start with reporting, then entries. Have Copilot build the monthly performance summary from numbers that are already closed. The risk is low, the time savings are real, and you'll learn how it thinks before it touches anything that posts.

Test the QuickBooks plugin on old data. It's rolling out, so it may not show up for you right away. When it does, point it at last month's numbers first.

On October 26 Anthropic filed an S-1 with the SEC for a proposed IPO and plans to list on the Nasdaq under the ticker ANT. The share count and price range haven't been set. AI Weekly, citing Bloomberg and The Information, reported first half 2026 revenue of $16.3 billion, up from roughly $1.2 billion a year earlier, and said about 20% of that came from two unnamed customers. The headline net loss of $80.7 billion is mostly a $72.4 billion noncash accounting charge on convertible notes, per the same report.

Why it matters. If you run part of your business on Claude, your vendor is about to answer to public shareholders every quarter. That cuts both ways. Public companies have to disclose a lot, so you'll know more about where the business is headed than you ever have. They also feel steady pressure to grow margins, and that pressure usually shows up in pricing and plan structure. I wouldn't panic about any of it. I'd just stop treating any single AI vendor like the water company.

What I'd do.

Keep your prompts portable. Store your best prompts, instructions and examples in a plain document you own, outside any one AI app. If pricing changes, you can move in an afternoon instead of a month.

Run your most important workflow on two models. Once a quarter, test it on a second provider. You don't have to switch. You just want to know you could.

Skim the risk factors when the prospectus is final. Companies tell regulators things they'd never put on a pricing page. Twenty minutes with that section will tell you what Anthropic itself thinks could go wrong.

Mistral released the open weights for Mistral Large 4, a trillion parameter model nicknamed Le Chonk, on October 29. The weights are on Hugging Face, and the model is also available through Mistral's API. Mistral says it ranks first among open models on Artificial Analysis' Coding Agent Index and uses 47% fewer tokens per task than the Chinese open model GLM-5.3. API pricing is $0.70 per million input tokens and $2.10 per million output. Those are Mistral's own claims, so treat the benchmarks as a sales pitch until outside testers weigh in.

The other selling point is geography. Mistral's API preview runs on its own infrastructure in Europe, and with open weights a company can run the model on its own servers, where nobody can switch it off. Mistral's VP of Science made exactly that pitch to TechCrunch.

Why it matters. Almost nobody reading this will run a trillion parameter model in the back office. That takes serious hardware. But it matters if you have clients in Europe, or clients in healthcare, legal or finance who ask where their data goes. "It runs on a European company's servers in Europe" is an answer that closes deals. And at those prices, Mistral makes a cheap backup model for any workflow you can't afford to have go dark.

What I'd do.

Have your data answer ready before a client asks. Write one paragraph listing which AI tools touch client data and where they run. Keep it handy for the next security questionnaire.

Ask your software vendors about open models. If a tool you pay for runs on an expensive closed model, ask whether they're testing open weight options. Cheaper inputs for them should eventually mean better pricing for you.

Run a cheap side by side. Push a batch of real but nonsensitive tasks through Mistral's API and compare the results with what you're paying for now.

Thrive Holdings, a holding company spun out of Josh Kushner's Thrive Capital, is buying ADT for $10 a share, about $7.6 billion in equity and $15.3 billion in enterprise value. OpenAI is both an investor and a technology partner. Thrive says it holds more than 75 companies, and its stated plan is to own them long term and revitalize them with AI, with OpenAI supplying models, technology and forward deployed engineers. ADT will be the foundation of Thrive's smart home security business.

Why it matters. This is the AI rollup playbook at full size. Buy a service business with a big customer base and a lot of manual work, then send in the engineers to wring out the cost. I'd expect copycats in smaller industries, run by people with less money and more ambition. Two things follow for you. Your next competitor might be a local shop with a deep pocketed owner and AI engineers on call. And if you've ever thought about selling someday, buyers like this pay more for businesses whose processes are written down and ready to automate.

What I'd do.

Write down how your business actually runs. Every process that lives only in your head is a weakness against a well funded competitor, and a discount when a buyer looks at you. This prompt gets you started.

Automate the front door first. Missed call text back, booking and reminders are the first things a rollup automates. A tool like Go High Level handles all three today, no forward deployed engineers required.

Let your tools pass data to each other. A Make.com scenario that pushes new web form leads into your CRM and a Google Sheet takes about an hour to build, and it retires one more job somebody's doing by hand.

Microsoft offered buyouts to about 60,000 people. That's more than a quarter of its full time staff, at level 70 and below in the US, UK, Canada and India, per FOX Business. The report doesn't tie it to AI, but it lands while Microsoft spends heavily on data centers. Expect experienced people on the job market this winter.

Meta's Muse agent landed in Singapore. Meta brought Muse to Singapore with connectors for Gmail, Google Calendar and Instagram, plus partners like Grab, Shopee and Agoda. US and Canadian owners already have Muse for Small Business, so this is Meta signaling it plans to take the agent global.

Google announced Gemini 4 Carbon. Google calls it its next frontier model, with introductory API pricing of $2 per million input tokens and $10 per million output, rising to $4 and $20 later, per 9to5Google. Cyber defenders get it first. Everyone else gets "coming soon."

Reprice one automation you shelved for cost, using the new GPT-6.1 Terra rates.

Have Gemini build your next proposal deck, then edit it yourself in Slides.

Run Fill with Gemini on a list you know cold and click the sources on five cells.

Ask your bookkeeper whether Copilot should touch reconciliations, and turn on the audit trail if the answer is yes.

Save your best prompts in a document you own, outside any single AI app.

Pick one process that lives in your head and use the prompt above to get it on paper.

Jordan

The AI Newsroom | Practical AI for people with a business to run.