Week of October 20-26, 2026
Three things moved in AI this week that touch how Alaska service businesses handle leads, phone calls, and the tools they’ve already built automations on. Here’s what changed and what to do about it.
Yelp now charges contractors $30 a month for an AI agent that answers and qualifies leads
On October 20, Yelp announced its 2026 Fall Product Release, and the headline item for service businesses is Smart Lead Reply: a paid AI agent that instantly responds when a consumer submits a Request a Quote, asks follow-up questions, gathers project details, and qualifies the lead before a human ever sees it. Pros can customize it with their own scheduling links, pricing, and promotions. It costs $30 a month, rolls out in November, and is limited to categories that use Request a Quote, including movers, roofers, plumbers, home cleaners, contractors, and electricians. Yelp also expanded integrations with ServiceTitan, Housecall Pro, Jobber, and Workiz so leads and scheduling can flow into tools these trades already run.
This matters because the gap Yelp is closing is a specific and common one: a lead comes in outside business hours or during a busy stretch, nobody replies fast enough, and the homeowner books with whichever pro answered first. An AI agent that responds in the first minute and asks the right qualifying questions changes who wins that lead, not just how it gets logged.
If your business gets Request a Quote leads through Yelp, in the trades this will show up as an option on your account in November. Before paying for it, check whether it duplicates something you already have through ServiceTitan or Housecall Pro, since Yelp built this to plug into both.
Response speed on new leads is now a paid feature on the platforms that generate them. If you're not the business answering fastest, you're increasingly the business losing the job to whoever is.
Mistral released a voice-agent SDK built to plug into real phone systems, not just chat apps
On October 20, Mistral AI launched Voxtral 3, a family of speech models paired with a voice-agent SDK built on LiveKit. The SDK chains listening, reasoning, function calling, and speech generation into one pipeline fast enough to complete a conversational turn in under a second, and developers can swap the underlying language model, voice, or language without rebuilding the agent. The practical detail for dispatch and call-center use is telephony support: calls can run over standard SIP lines, the protocol actual phone systems and IVR setups use, not just browser-based voice chat. Transcription pricing starts at $0.0035 a minute, and Mistral says coverage spans 50-plus languages.
This is a toolkit for developers and vendors, not a product you install directly. The significance is that a major AI lab is now shipping voice infrastructure aimed squarely at phone systems and contact centers, which is exactly the layer that AI answering services, dispatch tools, and after-hours call handlers for trades and logistics are built on top of.
If you use or are evaluating an AI phone answering or dispatch tool, ask the vendor what's actually under the hood and how recently it was updated. The underlying voice models are improving every few months right now, and a vendor running last year's model is giving you a worse product than what's technically available.
OpenAI set a hard shutdown date for GPT-4o and GPT-4.1, the models a lot of existing automations still use
On October 14, OpenAI notified developers that GPT-4o, GPT-4o mini, GPT-4.1, and GPT-4.1 mini, along with every fine-tuned model built on top of them, will be removed from the API on January 14, 2027. OpenAI is pointing everyone toward gpt-6-luna as the replacement. The three-month notice is shorter than the six months OpenAI’s own deprecation policy generally promises, which has drawn some pushback from developers relying on these specific models.
We flagged the rumor of this retirement last week without a confirmed date; it’s now locked in. The practical risk isn’t the model names, it’s that these were some of the most widely used models of the last two years, which means a lot of quiet, working automations, from email drafting tools to internal chatbots to custom integrations built by an outside vendor, are quietly pointed at a model ID with an expiration date now attached.
If someone built you a custom AI tool, chatbot, or automation in the last two years and you haven't touched the underlying setup since, this is worth a five-minute check with whoever built it. January sounds far off, but "find out who built it and whether they're still around" can itself take longer than fixing the actual code.
Check with any vendor or developer who built an AI tool for you on OpenAI's API, and ask directly whether it uses GPT-4o or GPT-4.1. If the answer is yes, get a plan to move it to a current model before January 14, 2027, not after.