top of page

This Week in AI: October 27 – November 2, 2026 — The Tools Got Cheaper to Run and Easier to Check

Writer: Zh Dast Sari
Zh Dast Sari
Nov 2
7 min read

What changed in AI this week that a small business owner should actually notice? The document tools learned to finish the file without a download, the price of routine machine work dropped again, and a careful costing of AI receptionists put a number on the pause callers will tolerate: about half a second. Here is what each one means if you run a shop, a clinic, a restaurant, or a crew.


I run NexAIS (https://www.nexais.ca/) in Richmond Hill. We put AVA, our AI receptionist, on small businesses' phone lines, so I read this week's launches the way an owner would: does it save a real half-hour, does it lower a real bill, and can I check its work? Three of this week's items pass that test. Two are warnings dressed as features. One is a survey that explains a lot about why AI spending and AI grumbling rise at the same time.


1. Gemini can now finish the file inside the chat — October 29


Until last week, asking Gemini for a document went one way: prompt in, finished file out, and then you downloaded it, opened it somewhere else, fixed the two lines that were wrong, and saved it again. On October 29, Google closed that loop. You can now create a document, spreadsheet, slide deck, or PDF inside the Gemini chat, preview it there, type edits into it yourself, or highlight a passage and ask for that part to be redone. PDF editing works on web and on your phone; the other formats are web only for now. There is no extra charge for the feature itself.


The honest limits matter as much as the launch. Comments and version history still live in the full Docs, Sheets, and Slides apps, not in the chat window. So the chat is a drafting desk, not the filing cabinet. Use it for the handout, the one-page quote, the staff schedule you keep remaking. When the file becomes the version your customers or your accountant will rely on, open it in the real app and read it there once, properly.


2. Sheets shows its sources, and Slides can redo the whole deck — October 28 and 30


Two smaller Google items belong together because they point the same way: AI output is getting easier to audit.


On October 28, Google said Gemini in Sheets now attaches a web source to individual AI-filled cells, and can handle requests of up to 3,000 cells at a time. A bigger spreadsheet was never the problem. Knowing which three cells you should not trust was the problem. A source sitting inside the cell does not prove the number is right — it gives you a place to check, which is what was missing.


On October 30, Google described Gemini editing across a whole presentation at once — content and visual style together — instead of making you fix slide 4, then slide 5, then slide 6 the same way. The rollout began October 29 and may take more than 15 days, English first. Practical advice: try it on a copy of a deck you already own, ask for one plain job ("make the headings consistent"), and read every slide after. A tool that can fix forty slides in a minute can also spread one wrong price across forty slides in the same minute.


3. OpenAI cut the price of routine work again — October 26


OpenAI introduced GPT-6.1 Terra on October 26, a mid-sized model built for writing, reading images and forms, and doing tasks on a computer. The price is the news: $1 per million input tokens and $5 per million output tokens through the API, with cheaper rates for repeated, cached material. It is not sitting in the ordinary ChatGPT chat window — it reaches businesses through the API and through paid work tools.


Most owners will never buy a token. They will feel this as the tools they already pay for getting cheaper to run, or doing more for the same monthly fee over the next few months. There is a second lesson in it, and SAP underlined it the same week by moving some of its own work to cheaper models on purpose: the expensive model is not automatically the right one. Reading an invoice, sorting an inbox, drafting a routine reply — that is routine work, and routine work should be priced like it. Save the costly model, and your own attention, for the jobs where a mistake costs money.


4. Most owners use AI. Most spend $100 a month or less on it.


A Goldman Sachs survey of 2,474 graduates of its 10,000 Small Businesses program, reported on October 30, found 86% use AI at work. Then the number that should shape every AI pitch, including mine: 73% spend $100 or less a month on it. Writing help and marketing content were the common uses. The sample is program graduates, not every small business in the country, so treat it as a strong signal rather than a census.


That gap — near-universal use, very small budgets — is not hypocrisy. It is owners saying: show me the one job it does measurably well, at a price I can predict, or I will keep it at the free tier. It is also why I distrust per-minute pricing for the phone line. An owner with a $100-a-month habit does not need a meter that turns her best month into her biggest bill. She needs a flat number she can put in the budget in January and stop thinking about.


5. The receptionist costing got honest: half a second, and $0.11 to $0.31 a minute


A detailed build-versus-buy costing of AI receptionists, updated this weekend, did two pieces of arithmetic the marketing pages skip.


First, the pause. In ordinary conversation, people leave about 200 milliseconds between turns. They notice a delay after roughly 300 milliseconds, feel it plainly after 500, and past 800 milliseconds to a second they start talking over the machine because they assume it did not hear them. So the working budget for an AI receptionist is under 800 milliseconds, with 500 as the target. A prettier voice that answers a second and a half late loses to a plainer voice that answers on time. When you test any receptionist — ours included — call it and listen to the gaps, not the accent.


Second, the minute. The headline rates owners see, around $0.05 a minute, cover only the platform that organises the call. Add speech recognition, the language model, the voice, and the phone line, and the same minute costs $0.11 to $0.31 all-in. The costing's practical rule is the one I give owners myself: under roughly 1,000 ordinary calls a month, buy a finished service; build your own only when volume, margin, or a rule about where your data must live makes the engineering worth it. Underneath it sits the survey figure that started this whole product category: in a September 2025 CallRail survey of 1,000 US consumers, 82% said they would call a competitor when a business does not pick up, and only 42% leave a voicemail.


That is why AVA is built the way it is. Each one is designed around the business it answers for — its services, its booking questions, its rules for what goes straight to a person — and it remembers returning callers, with their call and message history, instead of meeting every customer as a stranger. Speed gets the call answered. Memory is what makes the answer useful.


6. A quiet warning: check what your automations missed on October 26–28


Anthropic launched a cost-analytics tool on October 29 that breaks Claude spending down by job and account, which is genuinely useful if your automations run while you sleep. The same week carried the reason you need it. Anthropic marked two connector problems resolved on October 28: calendar listing and search were affected from October 26 to 28, and Gmail reading was affected from late October 27 into October 28. If an assistant watches your inbox or calendar, a clean run today does not prove last Tuesday's work was complete. Pick five messages or appointments from those dates and check them against what your tool reported. It takes ten minutes, and it is the difference between supervising an automation and merely owning one.


What this week adds up to


Nothing this week asks you to buy something new. The pattern is quieter than that: the draft now finishes inside the chat, the spreadsheet shows where its numbers came from, routine machine work costs less, and the real price of answering a phone — in milliseconds and in cents — is finally written down where owners can read it. The businesses getting value from all this are still the ones doing one unfashionable thing: checking the output and counting what comes back.


Our numbers stay in the open for the same reason. Starter is $197 a month (https://crm.nexais.ca/pricing/ai-receptionist) and is the receptionist on its own — answering, taking messages, asking your booking questions, passing urgent calls straight to a person. Professional is $399 a month and adds the CRM and the app, where bookings and customer history sit in one place; that is where restaurants start, because a restaurant needs the booking system, not just a voice. Grow is $599 a month and adds website and Instagram management on top. There is no smaller package under Starter, and the price does not move with your call volume.


The trial is seven days, free, and it runs your own AVA on your line. Your card is taken at signup, nothing is charged during the seven days, and charging starts on day 8 only if you continue. If you would rather just listen first, call the demo line, +1 (437) 529-0400, and time the pauses yourself. It answers at noon and at 2 a.m. A missed call can be a missed customer — this week, at least, the tools that answer got cheaper to run and easier to check. Questions about your own line? Our contact page (https://www.nexais.ca/contact) reaches a real person on our team.


Sources


- Lilach Bullock, "AI news this week for small business: 1 November 2026" (November 1, 2026) — Gemini file editing (Oct 29), Slides whole-deck edits (Oct 30), GPT-6.1 Terra (Oct 26), Sheets source citations (Oct 28), Claude cost analytics (Oct 29) and connector incidents (resolved Oct 28) — https://www.lilachbullock.com/ai-news-this-week-for-small-business-1-november-2026/

- Pondero, "Gemini can now create, preview, and edit files without leaving the chat" — details from Google Workspace Updates, October 29, 2026: formats, PDF on mobile, other formats web only, no separate admin control — https://pondero.ai/agents/guides/gemini-in-chat-file-creation-guide/

- MarketingProfs, "AI Update, October 30, 2026: AI News and Views From the Past Week" — Goldman Sachs survey of 2,474 program participants (86% use AI; 73% spend $100 or less monthly); GPT-6.1 Terra API pricing $1 / $5 per million tokens; SAP shift to cheaper models — https://www.marketingprofs.com/opinions/2026/56256/ai-update-october-30-2026-ai-news-and-views-from-the-past-week

- ForaSoft, "An AI Receptionist Costs $0.035 a Minute to Run. Most Businesses Should Still Buy One." (updated November 2026) — latency budget under ~800 ms with ~500 ms target; all-in cost $0.11–$0.31/min vs $0.05/min headline platform rates; buy a finished service under ~1,000 calls/month; CallRail September 2025 survey (1,000 US consumers: 82% would call a competitor, 42% leave a voicemail) — https://forasoft.medium.com/an-ai-receptionist-costs-0-035-a-minute-to-run-most-businesses-should-still-buy-one-3bc8d7942a35

 
 
 

Recent Posts

See All

Comments


nexais-logo-full-horizontal-C1ABDukH.png

© 2026 NexAIS Inc. All rights reserved.

bottom of page