Alexa for Shopping Moves Into Text Messages, Gemini 4 Carbon Opens to Paying Developers With a Price Clock, and NiCE Rebuilds Cognigy Around Tracing
Amazon began letting invited US customers text Alexa for Shopping, Google opened Gemini 4 Carbon to the paid Gemini API at an introductory price that doubles on December 1, and NiCE launched a next-generation Cognigy with reusable skills, decision tracing and native voice-to-voice. What SMS agents, expiring model prices and per-turn tracing mean for chatbot builders.
This week's chatbot news is less about what a model can say and more about where an agent lives, what it costs next month, and whether you can explain what it did. Amazon put a shopping agent in the text-message thread, Google opened its newest Gemini model to paying developers with a price that changes in two weeks, and NiCE rebuilt its Cognigy agent platform around tracing every turn. Details come from company announcements and launch-week coverage, so treat specifics as provisional.
1. The Agent Moves Into the SMS Thread
On November 6, Amazon began letting a small, invitation-only group of US customers text Alexa for Shopping, the agent it formed in May by merging its Rufus shopping chatbot with Alexa+. A customer accepts the terms and links a phone number, and the agent appears in their contacts. They can ask it to find and compare products, build a cart, watch prices or check an order, and it completes a purchase once the customer approves. Amazon says the conversation and shopping preferences carry across texts, the Amazon app, the website and Alexa devices, and that it will invite millions more customers in the coming weeks. Sponsored Products will appear among the options the agent texts back, free to advertisers for now, though reporting says the agent's own automated purchases are meant to stay free of sponsored substitutions.
Text messaging is the lowest-friction channel there is, with no app to install and no new interface to learn. It also strips away everything a chat UI gives you for free: buttons, rich cards, visible ad labels, an obvious "confirm" step. If you build a messaging agent, the confirmation has to be designed into the conversation itself. Make it explicit, bound to a specific cart and total, and short-lived:
// The agent may propose; only an exact, recent reply from the user commits.
function canCommit(pending: { cartHash: string; total: number; sentAt: number }, reply: string) {
const fresh = Date.now() - pending.sentAt < 10 * 60_000;
return fresh && reply.trim().toUpperCase() === 'YES'; // not "ok", "sure", or a thumbs-up
}
Keep the cart hash and total in the confirmation message, so a later "YES" can't approve a cart that changed after it was quoted, and label sponsored options in the text itself, because there's nowhere else to put the label.
2. A Frontier Model With an Expiry Date on Its Price
Google opened Gemini 4 Carbon to the paid Gemini API in Google AI Studio and to Gemini for business on November 10, after launching it on October 28 to a small group of cyber-defense partners first. Free-tier developers can't use it. Reported introductory pricing is $2 per million input tokens and $10 per million output, matching Gemini 4 Argon, until November 30; from December 1 it rises to $4 and $20. Cached input is discounted 95%. In the consumer apps, Carbon is limited to Google AI Ultra subscribers, who get it in the Gemini app, AI Mode in Search and Antigravity. Third-party listings report a context window of about one million tokens, but Google's announcement reportedly doesn't state one, so confirm limits in the API docs before you rely on them.
The lesson isn't about Carbon specifically. A benchmark run or cost model done this month will be off by half next month. Put the price date in your model config, alert before it passes, and rerun the cost comparison against your current default when it does. If your traffic leans on long, repeated system prompts or retrieved context, the cached-input discount may matter more to your bill than the headline rate, so measure your cache hit rate before deciding.
3. Running Agents, Not Just Building Them
On November 12, NiCE launched the next generation of Cognigy, its enterprise platform for customer-service agents. Teams now assemble agents from reusable skills, tools and personas, with AI-assisted development to build them. Agent Tracing follows every step of a turn, including which skill loaded, each tool and LLM call, token use and the point where the agent handed off to a scripted flow. Native voice-to-voice support runs speech-to-speech models end to end, so an agent can reason and call tools while it keeps talking. The release also changes model pricing: default models are included in the conversation price. NiCE says existing deployments keep running unchanged and customers can adopt the new features when they are ready. Coverage framed the bet plainly: the hard part of agents is now operating them, not building them.
You don't need Cognigy to adopt the idea. Emit one trace record per turn with the skill, every tool call and its arguments, model, tokens and latency, plus why the turn ended or escalated. Without that record, the first customer complaint about a wrong refund becomes an argument rather than a lookup.
Conclusion
The three stories meet at the same point. An agent that takes orders by text, a model whose price moves on a fixed date, and a platform that bills by conversation all push the hard work from the prompt to operations: confirmations that hold up in plain text, cost assumptions with an expiry date, and a trace for every turn. Those are the parts users and finance teams will actually notice.
Suggested visuals: a sequence diagram of an SMS purchase showing proposal, quoted cart and total, explicit "YES" and commit; a chart of monthly cost for a sample workload on Gemini 4 Carbon before and after December 1, with and without cached input; and an annotated example of a single per-turn trace record.
— Maya
Frequently asked questions
Can anyone text Alexa for Shopping?
Not yet. Amazon started on November 6, 2026 with a small, invitation-only group of US customers and says it will invite millions more in the coming weeks. Users accept the terms, link a phone number, and the agent completes purchases after they approve.
How much does Gemini 4 Carbon cost in the API?
Reported introductory pricing is $2 per million input tokens and $10 per million output tokens through November 30, 2026, rising to $4 and $20 on December 1. It is available only on the paid Gemini API tier, opened to paying developers on November 10.
What is new in the next generation of NiCE Cognigy?
Launched November 12, 2026, it adds reusable skills, AI-assisted development, Agent Tracing that records each skill, tool call, LLM call and token use per turn, native voice-to-voice support, and default models included in the conversation price.
I'm Maya — I write most of what you'll read here. I spent years as a copywriter before I got a little obsessed with what these AI tools can actually do, so now I spend my days poking at chatbots, breaking them, and writing up what's worth your time. Everything here is something I've actually tried. If a prompt didn't work for me, it doesn't make the cut.
Want to try any of this?
Smillee's free and there's no signup — open it and paste in whatever you're working on.
Start chatting →More from the blog
- Trends
AT&T Makes Its 500 Agents Earn Their Autonomy, AWS Sells Cited Answers as a Connector, and New Jersey Moves to Treat Chatbots as Products
AT&T says nearly 500 agentic workflows run in production, each supervised by a person until it reaches mid-to-high-90s accuracy. AWS added a Web Answer connector to AgentCore Gateway that returns cited answers instead of raw search results. And a New Jersey bill that would make chatbot providers strictly liable for injuries cleared its first committee. What accuracy-gated autonomy, managed grounding and product liability mean for chatbot builders.
- Trends
Claude Managed Agents Move Inside the Firewall, Gemini Enterprise Agents Sign In as You, and ChatGPT Retires Two Voice Modes
Anthropic opened a private beta that lets companies run the whole Claude Managed Agents service on their own infrastructure, Google made 3-legged OAuth generally available for A2A agents in Gemini Enterprise, and OpenAI set December 11 to retire Advanced and Standard voice in ChatGPT in favor of GPT-Live. What self-hosted agent control planes, user-delegated agent auth and single-model voice mean for chatbot builders.
- Trends
DeepSeek Puts Its App Inside the API, Antigravity Splits Planning From Doing, and CALL-E Hands Agents a Phone Line
DeepSeek opened the search, file, voice and safety tools behind its consumer app to API developers, Google paired a planning model with Gemini 4 Hydrogen in a new Antigravity Orchestrator mode, and AI Rudder launched CALL-E so assistants like ChatGPT can place real phone calls after the user approves a call plan. What bundled tools, planner/executor splits and agents on the phone network mean for chatbot builders.