HomeInsightsSignal / NoiseSignal / Noise Nov 18
Signal / Noise

Signal / Noise (Nov 18, 2026): Agent 365 Spending Limits, GPT-4o End Dates, Citi AI-Agent Terms, Vercel Builds, Nvidia Quarter

WWaveo teamNovember 18, 202610 min read
Five glass tiles on metal stands rising along a curved dark platform, a thin red beam running through the first three, which glow red, and stopping before the last two, which are dark and unlit.
The short version

Five notices. Act on three: Microsoft says Agent 365 admins will be able to set spending limits (followed by "In preview."); OpenAI's January 14, 2027 shutdown date for GPT-4o and GPT-4.1; Citi's new section on third-party AI agents. Watch two: a Vercel build price and Nvidia's quarter. One thing to do: ask whoever built your AI tools for a list of the models they use, with end dates.

In plain words

Think of this edition as a spending dial, a wall calendar, a rulebook section and two smaller notices. One page says what Microsoft will let a company cap on AI spending. One page lists when older OpenAI models stop accepting requests through its API. One financial firm has added a section on who is responsible when software acts with your login. The last two are a hosting price and one company's quarterly numbers.

Three Signal items, two Watch items. Signal means you can act on it this week; Watch means worth knowing, with one quick check at most. Every number belongs to whoever published it; the "What it means for you" and "Our take" lines are ours. Previous: November 17.

Which dates matter in this edition?

DateWhat happens
Jan 6, 2027OpenAI's shutdown date for tts-1, tts-1-hd and gpt-4o-mini-tts
Jan 14, 2027OpenAI's shutdown date for GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini
Jan 20, 2027OpenAI's shutdown date for realtime and audio models and one transcription snapshot
Earlier: Nov 30, Dec 1 and Dec 11, 2026OpenAI rows in our November 17 edition

Signal: What can admins do with Microsoft's Agent 365 Cost Management?

Agent 365 is Microsoft's admin product for the AI agents a company runs (our gloss). Cost Management covers usage-based AI services, meaning services billed by how much you use them (our gloss). Microsoft's Book of News for its Ignite event says admins will be able to set spending limits on them.

What does Microsoft's Book of News say?

The item's capability sentences say admins "will be able to" do things, and the labels printed with them differ:

What Microsoft says admins will be able to doLabel printed with it
See what is driving spend; set spending policies, limits and alerts; extend analysis through Microsoft Fabric"In preview." (its own sentence, at the end of this one)
Forecast spend, ask questions in plain language and act on optimization recommendations"Available in the Frontier program this month."
Value insights across Microsoft and non-Microsoft agents, and delegation to business leaders"In December"
Agent 365 Business Analytics, for business leaders (spend, value, success rate, satisfaction)"now available in the Frontier program"

The item prints no date. The page header says Ignite runs November 17-20, 2026, so we read "this month" as November 2026. "In preview." is its own sentence at the end of the first capability sentence; in our reading it labels that sentence only.

What does the Tech Community post add?

Microsoft's Tech Community post of November 17, 2026 lists, under "New Cost Management capabilities", the line "Apply spending limits and alerts at the user, group, organization, or resource level"; the post prints no stage label on that line.

Who needs which license, and what does it cost?

In plain words: the three pages on these capabilities print conditions, not prices, and the pages word one condition in two ways (at least one Microsoft 365 Copilot license, or at least 50 at launch). The Book of News item says Business Analytics insights for Microsoft services "require at least one Microsoft 365 Copilot license in the tenant"; the Tech Community post says the same. A separate Microsoft post on Business Analytics, also dated November 17, words the launch condition differently: "at least 50 Microsoft 365 Copilot licenses" at launch, with availability expanding to tenants with "one or more licenses by mid-December" (no year printed; we read it as 2026). For non-Microsoft services the posts say "Agent 365 and/or ME7 licensing" and the Book of News says an Agent 365 license; the pages we read do not define ME7.

The Book of News item, the Tech Community post and the Learn page print no price for Cost Management, Business Analytics or any license. Microsoft's Tech Community post of May 1, 2026 says Agent 365 is "available as a standalone license at $15 per user per month". That is a license price, not a price for these capabilities. The Book of News item says "Frontier program" without defining it; the other pages say "Frontier preview" (Learn: "Frontier preview program"). We did not read a page saying whether the terms mean the same.

What it means for you: in our reading, if your company uses Microsoft 365 Copilot, ask your admin whether your company's Microsoft setup (its "tenant", our gloss) is in the Frontier program, which licenses it holds and, for Business Analytics, how many Microsoft 365 Copilot licenses it has (Microsoft's launch post names 50). We do not advise anyone to buy or wait.

Our take: Signal, because the check is one question to your admin. Status as of November 18, 2026.

Signal: What are the GPT-4o and GPT-4.1 end dates in OpenAI's API?

An API is the way software calls an AI model (our gloss). OpenAI's deprecations page lists shutdown dates in January 2027 for several models. It defines a shutdown as the state in which "The model no longer accepts requests through the standard API."

Which models, on which dates?

Every row has the status "Deprecated"; the page says "You can still use it until its shutdown date." It lists each model, with its dated snapshots, on its own row; we group rows that share a shutdown date and a replacement. Your developer will know the model names; the date in the first column is the part you need.

Shutdown dateAffected modelRecommended replacementNotice date
Jan 6, 2027tts-1; tts-1-hd; gpt-4o-mini-tts and its two dated snapshotsgpt-realtime-2.1-miniOct 1, 2026
Jan 14, 2027GPT-4o (gpt-4o-2024-08-06, gpt-4o-2024-11-20); GPT-4o mini (gpt-4o-mini-2024-07-18); GPT-4.1 (gpt-4.1-2025-04-14); GPT-4.1 mini (gpt-4.1-mini-2025-04-14)gpt-6-lunaOct 14, 2026
Jan 20, 2027gpt-realtime and gpt-realtime-2025-08-28; gpt-4o-realtimegpt-realtime-2.1Jul 20, 2026
Jan 20, 2027gpt-realtime-mini and gpt-realtime-mini-2025-12-15; gpt-4o-mini-realtimegpt-realtime-2.1-miniJul 20, 2026
Jan 20, 2027gpt-audio and gpt-audio-2025-08-28; gpt-4o-audio; gpt-audio-mini and gpt-audio-mini-2025-12-15; gpt-4o-mini-audiogpt-audio-1.5Jul 20, 2026
Jan 20, 2027gpt-4o-mini-transcribe-2025-03-20gpt-4o-mini-transcribe-2025-12-15Jul 20, 2026

From November 18, January 6 is 49 days away, January 14 is 57 and January 20 is 63 (our arithmetic).

What about fine-tuned models, and what is not on the page?

The page says fine-tuned versions of the listed models "follow the same shutdown schedule as their underlying base models".

The page prints no update date; we read it on November 18, 2026 at about 11:10 PST. It lists later shutdown dates too (February and April 2027), which we do not cover. The page states no grace period; it says developers may in some cases provision dedicated capacity for continued access after a shutdown date, through OpenAI's sales team.

What it means for you: in our reading, if you pay someone to build chatbots, voice features or automations, ask which model names they call and put these dates next to them. We do not advise anyone which replacement to use.

Our take: Signal, because the vendor fixed the dates and the first step is one list. Status as of November 18, 2026: every row is "Deprecated".

Signal: What does Citi's new Section H say about third-party AI agents?

The Online User Agreement governs use of Citi's online and mobile services. The page says it "was last updated November 2026 to add a new Section H" on third-party AI agents.

What does the section say?

The agreement defines an AI agent as software that takes "autonomous, semi-autonomous, or non-autonomous action" for the customer. It first says "You are cautioned against using any third-party AI Agent to access your accounts". It then says that "If you disregard this caution" and give an agent your online sign-in details, or otherwise let it access your accounts or perform transactions for you, "Citi will consider any access and actions taken by the agent as authorized by You". The same holds, it says, if the sign-in details are passed on or otherwise distributed to other people or AI agents.

The section adds these points (our paraphrase unless quoted):

  • Citi has "no responsibility or other obligation to investigate" the authority or validity of such actions, even if they conflict with the customer's instructions to the agent.
  • The customer is responsible for the agent's "incorrect interpretation or execution of your instructions", for unintended actions and for actions beyond the authority the customer granted; that sentence ends "except as otherwise permitted under applicable law".
  • It is "your responsibility and not Citi's to monitor" what the agent does.
  • Citi may limit or end a third-party agent's access "at any time without notice".
  • By letting an agent in, the customer is "authorizing Citi to disclose your information" to the agent.
  • The page says "the Citi Privacy Notice does not apply to the third-party AI Agent"; Citi does not control how the agent handles data, and AI tools that Citi itself provides are outside Section H.

Citi's separate Agreement for Online Funds Transfers has a notice on November 2026 updates and its own Section 4 on third-party AI agents. It says Citi will treat an agent's instructions as authorized by the customer "even if they exceed or otherwise conflict with" the customer's prompts. That is contract wording, not our conclusion.

Who does it cover?

The agreement's opening definitions say "you" means the Citi customer, "including consumers, sole proprietors and other eligible small businesses signers and users". Another section refers to eligible business accounts. The page does not say which business products Section H applies to. We read this agreement and the funds transfer agreement, not Citi's account agreements.

What it means for you: in our reading, if you or your staff use AI tools anywhere near company accounts, ask your own bank in writing what its current online terms say about them, and write down who in your company may authorize a tool. We are not lawyers or financial advisers. This describes a document, not a rule.

Our take: Signal, because the text is on Citi's own page and the question of who may authorize a tool can be settled this week. Status as of November 18, 2026: the page prints the month only.

Watch: What changes for new teams on Vercel's Standard builds?

Vercel is a hosting service for websites (our gloss). A build is the step that turns a project's code into a live site (our gloss). For disclosure, Waveo's own site is hosted on Vercel; this is not a recommendation.

Vercel's changelog entry of 17 Nov 2026 says "Starting today, new Pro and Enterprise teams no longer include a Standard build slot." On Pro, it says, "Standard builds cost $0.014 per build minute", and the monthly usage credit applies to build usage. "Enterprise rates follow your contract." It adds that Standard builds "remain billable after switching from Elastic to Standard" for teams covered by the change.

The entry is headed "for new teams" and speaks of new Pro and Enterprise teams. It does not say whether existing teams are affected, does not define "Standard build slot" or Elastic and does not state the size of the monthly credit. We read this entry only.

What it means for you: in our reading, if your site is hosted on Vercel, ask your developer whether your team is new or existing and how many build minutes a month your deploys use. We do not advise anyone to change host or settings.

Our take: Watch, because it is headed for new teams. Status as of November 18, 2026.

Watch: What did Nvidia report for the quarter ended October 25?

Nvidia makes chips used to run AI (our gloss). Its results release of November 17, 2026 says revenue for the quarter ended October 25, 2026 was $115.8 billion, "up 103% from a year ago". Data Center revenue was $107.6 billion.

For the fourth quarter of fiscal 2027, which the release says "includes an additional week", Nvidia expects revenue of $130.0 billion, plus or minus 3%. It also says "NVIDIA is not assuming any Data Center compute revenue from China in its outlook." We read only this release.

What it means for you: in our reading, nothing to do this week. We are not financial advisers and say nothing about any share price.

Our take: Watch, because it is one company's own figures. Status as of November 18, 2026.

Related reading: our October piece on banks and third-party AI agents; our decode of a Federal Reserve governor's speech on AI agents in customer accounts, a different kind of document; and our Klaviyo Composer and Customer Agent teardown.

If you do one thing this week

In our reading, start with the first line; the other two are quick checks. Not legal or IT advice.

  1. Ask whoever built your AI tools, by Friday, November 20, for one list: every model and tool they use, with its end date next to it. The dates are in our November 17 edition (November 30, December 1 and December 11) and in today's OpenAI table (January 6, 14 and 20). January 6 is 49 days from November 18 (our arithmetic).
  2. If your company uses Microsoft 365 Copilot, ask your admin whether your company's Microsoft setup is in the Frontier program and which licenses it holds.
  3. If your site is hosted on Vercel, ask your developer whether your team is new or existing and how many build minutes a month your deploys use.
Book an AI discovery call

How we know, and what we could not check

We read the seven main sources on November 18, 2026, from about 11:10 PST, and three added ones (Microsoft's Business Analytics post, its May 1 license post and Citi's funds transfer agreement) at about 11:42 PST. We tested nothing. Not read: press coverage, Microsoft's other Agent 365 pricing and licensing pages, OpenAI's pricing, migration and ChatGPT pages, Citi's account agreements, Vercel's pricing documents, and Nvidia's investor pages and filings.

Sources

FAQ

What is Microsoft's Agent 365 Cost Management, and is it available?

Microsoft's Book of News for Ignite (November 17-20, 2026; the item prints no date) describes it as "a centralized place to see and control a growing list of usage-based AI services" in the Microsoft admin center. It says admins will be able to set spending policies, limits and alerts, and, in our reading, "In preview." labels that sentence. It labels forecasting "Available in the Frontier program this month". The Book of News item, a Tech Community post of November 17 and a Learn page print no price; a May 1, 2026 Microsoft post prints a standalone Agent 365 license price of $15 per user per month, which is not a price for these capabilities.

Which OpenAI models have shutdown dates in January 2027?

OpenAI's deprecations page lists January 6, 2027 for tts-1, tts-1-hd and gpt-4o-mini-tts, January 14 for GPT-4o, GPT-4o mini, GPT-4.1 and GPT-4.1 mini, and January 20 for several realtime and audio models and one transcription snapshot. Every row has the status "Deprecated" and a recommended replacement named by OpenAI, for example gpt-6-luna for the January 14 rows. The page prints no update date, and says fine-tuned versions follow the same schedule as their base models. We recommend no replacement.

Does the January 14 shutdown date affect the ChatGPT app?

The OpenAI page we read lists dates for models in the API. We did not read a page about the models in the ChatGPT app, so we cannot say whether those dates affect it.

What did Citi add to its Online User Agreement about AI agents?

Citi's agreement says it "was last updated November 2026 to add a new Section H" on third-party AI agents; the page prints the month only. Section H cautions against using a third-party AI agent to access your accounts. It says that if you disregard the caution and give an agent your online sign-in details or otherwise let it access your accounts, Citi will treat the agent's access and actions as authorized by you. Citi's separate Agreement for Online Funds Transfers has a Section 4 on the same subject. We are not lawyers; this describes a document, not a rule.

Does Citi's new section cover small businesses?

The agreement's opening definitions say "you" means the Citi customer, "including consumers, sole proprietors and other eligible small businesses signers and users". The page we read does not say which business products Section H applies to, and we read only the Online User Agreement and the funds transfer agreement, not Citi's account agreements. Check the agreement that governs your product.

What changed for new teams on Vercel's Standard builds?

Vercel's changelog entry of 17 Nov 2026 says new Pro and Enterprise teams no longer include a Standard build slot. On Pro, it says Standard builds cost $0.014 per build minute and the monthly usage credit applies to build usage; Enterprise rates follow the contract. The entry is headed "for new teams" and does not state the size of the monthly credit. Waveo's own site is hosted on Vercel.

What did Nvidia report for its quarter ended October 25, 2026?

Nvidia's release of November 17, 2026 says revenue was $115.8 billion, up 103% from a year ago, with Data Center revenue of $107.6 billion. For the fourth quarter, a period that the release says includes an additional week, it expects revenue of $130.0 billion, plus or minus 3%, and says it is not assuming any Data Center compute revenue from China. These are the company's own figures and expectations.

Work with us

Not sure which of your tools depend on the January dates?

Tell us which AI tools your team uses and we will help you map which models and vendors each one depends on before the dates arrive.

Book an AI discovery call