Microsoft opened Ignite in San Francisco on Tuesday, November 17, 2026, with a keynote that called Copilot "the new OS for work" and introduced a platform layer named Microsoft IQ. OpenAI released two models and a sales plugin. Google published an interview about a genome model it had already made available in September.
Three different kinds of announcement, one shared question: what happens to the output after it is generated.
Microsoft: the second half of the sentence
Chairman and CEO Satya Nadella framed the week in two parts, and the second is the one worth quoting: "First is how we capitalize on new capabilities that are growing exponentially. And second is how do we build the capability to deal with the consequences of these exponentials."
The announcements attached to that framing are specific. Copilot is described as bringing Chat, Cowork, Code and Autopilot into one experience, with Office built in. Work IQ is described as "the most important database for any organization," holding people, relationships, documents, emails, meetings, chats and projects.
Copilot is also being extended for sales, legal, finance and service roles, with new plugins for business applications "starting with Dynamics 365" and a stated partnership to bring Salesforce into Copilot.
Two further items. Copilot Workspaces is described as a true "multiplayer form factor," where people, their agents and other people share context toward a common goal. Microsoft IQ externalizes the system underneath Copilot as a platform, with "hundreds of new connectors" spanning Fabric, Foundry and Microsoft 365, surfaced as business plugins that can be shared and inherited. Microsoft-Decision-1.1 was announced at that point in the remarks.
All of it is keynote language from an interested party, reviewed as text. No benchmark, availability date, licensing term or customer outcome was reviewed, and none is asserted here.
OpenAI: two models and a plugin that writes to the CRM
OpenAI introduced GPT-6.2 Astra and Sol, described in the announcement summary as offering "better professional outputs, coding, computer use, and efficiency at lower cost across ChatGPT Work, Codex, and the API." No benchmark, price figure or capability comparison is characterized here.
The same day brought a second, quieter release. OpenAI's summary of ChatGPT for Sales describes a "new Sales plugin and 30+ new and expanded partner plugins in ChatGPT to research prospects, prepare pitches, update your CRM, and move deals forward."
"Update your CRM" is the load-bearing phrase, and this part is analysis. A plugin that drafts a pitch produces text a person reads before it goes anywhere. A plugin that writes to a CRM writes to a system of record — the object that forecasting, territory planning, commission and audit all later depend on.
Those are different failure classes. The first scales with how carefully people read. The second scales with how carefully the write path is designed.
Google: an interview, not a launch
Google's November 17, 2026 contribution on this subject is an Ask a Scientist interview by contributor Lindsey Lanquist with Žiga Avsec, a Google DeepMind research scientist who oversees the genomics initiative.
The tool it discusses is not new. The post announcing AlphaGenome Atlas is dated September 8, 2026 and is co-bylined by Pushmeet Kohli, VP Science at Google DeepMind and Chief Scientist at Google Cloud, and by Avsec as Genomics Initiative Lead. Google's October 12, 2026 research roundup already described the Atlas as published and available.
That post's substance stands as company-stated reporting: a database that predicts "the effects of every possible single nucleotide variant in the human genome," built by pre-calculating the regulatory impact of all 9 billion single-letter changes into what Google describes as a 1-petabyte dataset, navigated by a single AlphaGenome Variant Impact score that combines coding and non-coding predictions.
Two named uses are reported in that September post, both vendor-supplied. At the Broad Institute, Laura Covill's team applied the AVI score to unsolved rare disease research and, per Google, flagged a DNM1 variant predicted to create an incorrect splice site, which "provided crucial supporting evidence to successfully solve the case." At the University of Exeter, Gareth Hawkes applied it to data from more than 54,000 UK Biobank participants and, per Google, uncovered 22% more non-coding genetic associations and identified 19 regions linked to body mass index.
What published on November 17 was the conversation around that work, including a limit the September announcement did not lead with.
The caveat arrived two months later
This section is analysis.
In the interview, Avsec describes AlphaGenome 2 as moving beyond tissue-level averages to predict how variants disrupt individual cell types. Then he states the constraint: the team is "actively researching better ways to measure how confident our model is in its predictions so researchers know how much to trust each result," and anyone using AlphaGenome "should interpret its predictions with other sources of information."
The sequencing is the lesson. A model that emits one number across 9 billion candidates was released in September; the explicit instruction to corroborate it arrived in an interview two months later. That is not a criticism of the science. It is an observation about where caveats sit in vendor communication, and a reason to read a launch post and the follow-up conversation as one document.
For a research lab, confidence is method. For an enterprise ranking anything — accounts, candidates, claims, inventory — a single composite score is a decision input whose uncertainty has to come from somewhere else: a second source, a named reviewer, or a threshold above which the output stops being actionable.
Named and not characterized
- Google Wallet on iOS. Google said Wallet is now available on iOS in the U.S. for tickets, boarding passes and receipts. No feature detail reviewed.
- Tasks in Gemini. Google said the feature is free in six countries, with new capabilities starting in the U.S. and deeper Workspace integration. Not tested.
- Temporary threads in AI Mode. Google said AI Mode now offers temporary threads for private searches, without personalization. Not tested.
- EmbeddingGemma 2. Google published a technical report on a multimodal embedding model covering a unified vector space and compact design. Not benchmarked here.
- Gemini 4 Carbon builders. Google published profiles of five builders working with Gemini 4 Carbon. Named only.
- Google Vids avatars. Google said anyone can create a personal avatar from a selfie and voice recording at no cost. Not characterized.
- State AI training. Google said free AI training and career certificates are available across 10 states through libraries and colleges. Terms not reviewed.
- AWS Heroes. Amazon recognized a new group of AWS Heroes ahead of re:Invent. Named only.
JWA interpretation
This section is the publisher's analysis, held separate from the reporting above.
Read together, the day's three announcements widen the surface on which software produces a judgment — a Copilot workspace context, a CRM field, a variant score — faster than the surrounding check gets redesigned. None of the three is unusual for doing that. All three are unusual for how explicitly they name the check as the open item.
Microsoft put "the consequences of these exponentials" in the second sentence of its keynote, then spent the rest of the visible remarks on form factors and connectors. Google put the confidence question in an interview two months after releasing the dataset. OpenAI described a plugin that updates the CRM without, in the summary reviewed, describing what approval it needs first.
For a buyer, that argues for one question applied consistently: what happens between the output and the consequence. Who reads it, what they compare it against, and what record survives the decision. An assistant that drafts can be judged by reading carefully. An assistant that writes can only be judged by the design of the write path.
Also on the site today
The most-read article in this publication on November 17 was an analysis of the rule that converts a demand forecast into a purchase order, and where that rule lives when it does not fit standard ERP configuration. It carried 15 counted visits at the time of this edition — a pageview count, not a count of unique readers.
What remains unconfirmed
- Availability, licensing terms and delivery dates for Copilot Workspaces, Microsoft IQ, Microsoft-Decision-1.1 and the business-application plugins named in the keynote. None was reviewed.
- Whether the stated plan to bring Salesforce into Copilot is signed, scoped or shipped. Only the on-stage statement was reviewed.
- Any benchmark, price change or capability comparison for GPT-6.2 Astra and Sol. None is characterized here.
- What permissions a Sales plugin needs to write to a CRM, and which of the partner plugins are new versus expanded.
- Independent reproduction of Google's reported research results, including the 22% figure and the DNM1 case. Both are company-stated in vendor posts.
- Availability, benchmarks or independent evaluation of AlphaGenome 2, which the interview describes but does not quantify.
- Any measured effect of these releases on enterprise procurement or deployment behavior. None is asserted.
- Market data. No stock, index or valuation figures were reviewed for this edition, including Nvidia's scheduled earnings for November 17, which are not characterized here.
Evidence notes and limits
Evidence for this edition was reviewed through 23:40 UTC on Tuesday, November 17, 2026.
Microsoft's Ignite 2026 post, dated November 17, 2026 and attributed to Chairman and CEO Satya Nadella, was retrieved and read through the models section. From it this article draws: the quoted two-part framing, including the full sentence ending "the consequences of these exponentials"; the description of Copilot as "the new OS for work" and as combining Chat, Cowork, Code and Autopilot with Office built in; the quoted "most important database for any organization" description of Work IQ; the role extensions for sales, legal, finance and service; the new business-application plugins "starting with Dynamics 365"; the stated Salesforce partnership; the "multiplayer form factor" description of Copilot Workspaces; the Microsoft IQ description including "hundreds of new connectors" across Fabric, Foundry and Microsoft 365 and shareable, inheritable business plugins; and the announcement of Microsoft-Decision-1.1. The retrieved text ends mid-sentence at that announcement, so no further item from the remarks is characterized, and no availability date, benchmark or customer result is drawn from the post.
Google DeepMind's AlphaGenome Atlas post is dated September 8, 2026 and is co-bylined by Pushmeet Kohli, VP Science at Google DeepMind and Chief Scientist at Google Cloud, and Žiga Avsec, Genomics Initiative Lead. From it this article draws, as company-stated: the quoted description of the Atlas as "a database that predicts the effects of every possible single nucleotide variant in the human genome"; the pre-calculation of the regulatory impact of all 9 billion single-letter changes into a 1-petabyte dataset; the 2%/98% coding versus non-coding framing; the AlphaGenome Variant Impact score as a single score combining coding and non-coding predictions; the Broad Institute work attributed to Laura Covill and the DNM1 variant, including the quoted "provided crucial supporting evidence to successfully solve the case"; the University of Exeter work attributed to Gareth Hawkes spanning more than 54,000 UK Biobank participants, the 22% figure and the 19 regions linked to body mass index; and the statement that the Atlas is available through a website portal requiring no coding skills. All research results are reproduced as vendor-reported rather than independently verified.
Google's October 12, 2026 research roundup was reviewed only to establish that the Atlas was already described as published and available before November 17. Its other figures — publication counts, citation counts, screening results and forecasting rankings — are not reported here.
Google's Ask a Scientist interview, by contributor Lindsey Lanquist with Žiga Avsec, is dated November 17, 2026 and was retrieved. It is the source for the description of AlphaGenome 2 addressing cell-type specificity beyond tissue-level averages; for the two quoted statements on measuring confidence and corroborating predictions with other sources; and for the reference to strongly predicted effects performing well in early analyses. No accuracy figure is asserted from it.
OpenAI's GPT-6.2 Astra and Sol announcement and its ChatGPT for Sales announcement, both dated November 17, 2026, are cited only for the descriptions in their announcement summaries: better professional outputs, coding, computer use and efficiency at lower cost across ChatGPT Work, Codex and the API; and a new Sales plugin plus 30 or more new and expanded partner plugins able to research prospects, prepare pitches, update CRM records and move deals forward. No benchmark, price figure, permission scope, availability term or customer outcome is characterized from either.
The reading of a CRM write as a different failure class from a drafted document, the observation about where vendor caveats appear relative to release dates, and the connection between written CRM fields and downstream forecast and commission records are this publication's analysis. Nothing here establishes that any organization adopted these tools, changed a procurement decision, or experienced a specific error.
The advertisement above is separate from this analysis and is not a source for any claim in it.