Good day and welcome to Eye on AI. On this version…A breakdown of the AI information from Salesforce’s Dreamforce…OpenAI debuts its browser…Google makes a most cancers analysis breakthrough…and backlash in opposition to the data-center increase builds.
I spent final week on a reporting project in San Francisco. Whereas there, I dropped by Salesforce’s big Dreamforce convention. Salesforce founder and CEO Marc Benioff made a number of headlines—however in all probability not for the rationale he needed.
En path to the convention, Benioff had given an interview to the New York Occasions during which he’d been complimentary of President Trump and stated he would welcome the Nationwide Guard being deployed to San Francisco. By the top of the week, Benioff had apologized for that stance—saying he had spoken out of an abundance of concern in regards to the security of these attending his convention—after being sharply criticized by famed Silicon Valley investor and advisor Ron Conway, who resigned from the board of Benioff’s charitable basis in protest, in addition to Lauren Powell Jobs, Steve Jobs’ widow and president of the Emerson Collective, who lambasted Benioff’s place in an op-ed within the Wall Avenue Journal.
All of this was a distraction from what was really going down at Dreamforce, the place Salesforce introduced a slew of latest instruments to make it simpler for its clients to construct and deploy AI brokers. Few tech firm CEOs have staked as a lot on the promise of “agentic AI” as Benioff. How is that wager going? Effectively, to evaluate from Dreamforce, the reply is—like a lot that’s taking place in AI proper now—one thing of a blended image.
On the one hand, Salesforce says that its “Agentforce” options have seen the quickest adoption of any product the corporate has ever launched. However, then again, not counting brokers in Slack—which I’ll cowl additional in a second—the corporate has solely about 12,500 clients (or simply over 8% of its buyer base) having adopted Agentforce previously 12 months, and solely 6,000 of these are paid engagements. (Salesforce permits clients to experiment with Agentforce totally free, up till a sure utilization threshold.) These comparatively low numbers—which helped clarify Salesforce’s relatively tepid current income progress—had been weighing on Salesforce’s inventory.
However on the convention, Salesforce additionally issued improved guidance for its future revenues, saying that, because of Agentforce, natural gross sales progress would speed up to above 10% 12 months over 12 months in 2026 and proceed that method. The corporate now forecasts $60 billion in annual gross sales by 2030, forward of analysts’ consensus view. That projection despatched Salesforce’s refill 5% on the day, and shares have continued to climb over the previous week. So what’s actually taking place right here?
Innovation out-stripping adoption
In his keynote at Dreamforce Benioff, acknowledged that there was at the moment a “bifurcation” between fast client adoption of AI chatbots, corresponding to OpenAI’s ChatGPT, and comparatively slower enterprise adoption of AI. And through a press convention afterwards, he went additional, saying, “That is the second the place this know-how innovation [is] out-stripping buyer adoption. Our job is to get these clients into adoption mode. The way in which to do it’s by displaying them clients who’re front-runners on this, so if you have a look at these clients, they’re making it occur.” Salesforce can be doing this by creating extra “forward-deployed engineers” to work immediately with its clients, serving to them construct AI brokers. The corporate appears to have realized previously 12 months that enterprises will want Salesforce to carry their hand greater than was true with its conventional SaaS merchandise.
I toured a piece of the conference corridor branded “Agentforce Metropolis” to have a look at a few of these early adopters and their AI brokers. Just a few, like Williams & Sonoma’s AI agent that provides clients recipe recommendation—and by the way in which, additionally suggests cookware they could must make that roast or pie—appeared a bit gimmicky. That recipe agent may give the corporate extra touchpoints with a buyer, nevertheless it wasn’t clear it will immediately translate into greater gross sales. Extra attention-grabbing was PepsiCo’s AI agent that enables it to offer higher customer support to the various small Mother-and-Pop shopkeepers and bodega house owners who inventory Pepsi merchandise however who don’t essentially get loads of consideration and recommendation from Pepsi’s gross sales reps. Extra promising nonetheless is Dell’s use of Salesforce’s AI brokers to automate components of its provide chain, together with onboarding new suppliers, which has reduce its common onboarding time from months to days.
However Salesforce executives universally acknowledged that it will take time for extra corporations to comply with the lead of Dell or PepsiCo. Adam Evans, Salesforce’s govt vice chairman and common supervisor of Salesforce AI, advised me that he has spent many of the previous 12 months making an attempt to indicate clients the way to transfer past experimentation with AI brokers. “What’s exhausting is to create brokers that scale, that do issues persistently, you can work into an enterprise course of to create worth,” he stated.
Slack because the interface to all of Salesforce’s software program
To assist clients with that, Salesforce unveiled a number of new Agentforce options at Dreamforce. There’s an Agent Builderthat enables a person to easily describe what it needs an agent to do; the system then robotically units it up, with a lot much less handbook tinkering that was wanted beforehand. There’s a brand new voice interface for AI brokers, powered partially by way of OpenAI’s voice fashions, that improves interplay with brokers. There’s an Agent Script instrument that lets companies set up rule-based processes for a part of a given course of and use the much less predictable, however probably highly effective, reasoning of a big language mannequin (LLM) for different components of the method. It additionally launched a brand new vibe-coding instrument referred to as Agentforce Vibes, which is geared toward builders already expert at constructing Salesforce purposes, however permits them to create these apps, together with extra subtle agentic workflows, utilizing pure language.
Maybe the most important information is that Salesforce is hoping to place Slack—which Salesforce purchased in 2020 for $27.7 billion—as the primary “conversational gateway” to all of Salesforce’s software program, together with its Agentforce choices. Denise Dresser, Slack’s CEO, advised me that the concept is that as a substitute of getting to study to configure and run processes in Salesforce’s Advertising and marketing Cloud or its Service Cloud, a Slack person might merely message an AI agent inside Slack that can run these processes for them utilizing Salesforce’s software program within the background.
Slack has additionally created “data brokers” that may floor data from a selected Slack channel and carry out sure actions—serving to them onboard a brand new rent for, occasion, or set up software program on a brand new laptop computer, immediately from Slack. Dresser additionally advised me that she thinks Slack is the best interface as a result of it might probably incorporate each person-to-person and workforce interactions on the identical channel in which you’ll have people and groups interacting with AI brokers, whereas some AI corporations are solely optimizing their merchandise for human-to-AI collaboration.
Dresser definitely might have some extent about chat as the brand new interface to software program. It’s a imaginative and prescient that AI corporations like OpenAI and Anthropic are additionally pursuing. And a few of these AI corporations are projecting the concept even additional, envisioning a future the place AI brokers use their coding skills spin up bespoke software program on the fly to deal with most of the duties that now require enterprise software program, like, um, Salesforce. However whether or not that imaginative and prescient will come to fruition or whether or not conventional SaaS merchandise will live on, simply with AI front-ends, stays to be seen. One factor that’s clear from Salesforce’s expertise previously 12 months because it began rolling out AI brokers is that enterprise adoption will in all probability run behind over-hyped market expectations.
With that, right here’s extra AI information.
Jeremy Kahn
jeremy.kahn@fortune.com
@jeremyakahn
If you wish to study extra about how AI will help your organization to succeed and listen to from trade leaders on the place this know-how is heading, I hope you’ll think about becoming a member of me at Fortune Brainstorm AI San Francisco on Dec. 8–9. Amongst these confirmed to seem to this point embrace Google Cloud chief Thomas Kurian, Intuit CEO Sasan Goodarzi, Databricks CEO Ali Ghodsi, Glean CEO Arvind Jain, Amazon’s Panos Panay, and lots of extra. Register now.
FORTUNE ON AI
Marketing leaders say AI is rewriting how brands reach Gen Z and millennials: ‘Forget what you know; learn this’—by Jessica Coacci
Exclusive: Early AI darling LangChain is now a unicorn with a fresh $125 million in funding—by Sharon Goldman
Sam Altman wants to ‘treat adults like adults’—but can OpenAI keep ChatGPT safe after opening the door to erotica?—by Beatrice Nolan
Empathy is the most under-hyped factor of the AI transformation era, American Express exec says—by Sydney Lake
EYE ON AI NEWS
OpenAI launches its long-awaited AI-powered internet browser. The corporate debuted ChatGPT Atlas, an internet browser with a built-in conversational assistant designed to behave as a “companion” whereas customers navigate the web. Accessible globally on macOS (with Home windows, iOS, and Android variations coming quickly), Atlas contains an agent mode for Plus and Professional subscribers that enables ChatGPT to take real-world actions like reserving reservations, enhancing paperwork, or managing emails. The discharge marks OpenAI’s most direct problem but to Google and Perplexity, each of which have already launched their AI-capable browsers. The shares of Google-parent Alphabet fell sharply on the information. You’ll be able to learn extra from Fortunehere.
Walmart companions with OpenAI on commerce. Throughout the subsequent few months, U.S. ChatGPT customers will be capable of purchase most Walmart merchandise immediately within the chatbot utilizing its new On the spot Checkout characteristic, the 2 corporations introduced. The transfer is an indication of what could also be a brand new paradigm of “conversational buying”—and it means Walmart will share buy information from ChatGPT transactions with OpenAI, a notable concession given retailers’ traditional grip on such information. You’ll be able to learn extra from the Wall Avenue Journalhere.
Examine finds AI copilots might save U.Ok. well being service 43 minutes per staffer per day. That’s the results of a examine that appeared on the deployment of Microsoft’s 365 Copilot throughout 90 completely different organizations which might be a part of Britain’s Nationwide Well being Service. That would translate to thousands and thousands of hours per 12 months if rolled out throughout your complete well being service. You’ll be able to learn extra from The Telegraphhere.
Anthropic pushes again after criticism from White Home ‘AI czar.’ Anthropic CEO Dario Amodei wrote a weblog put up rebutting social media assaults from David Sacks, the White Home’s crypto and AI czar. Sacks accused the corporate of pursuing a technique of “regulatory seize”—making an attempt to lift fears of existential danger from AI to be able to foyer for AI regulation on the state and federal stage that will favor its personal merchandise over these of rivals. Amodei argued that Anthropic aligns with the Trump administration on key AI priorities—citing reward for competitiveness-focused govt orders and a July assembly the place he spoke with President Trump—whereas nonetheless opposing a proposed 10-year state-level moratorium on new AI legal guidelines. He additionally defended Anthropic’s help for California’s just lately enacted AI legislation, which requires AI corporations constructing highly effective AI techniques to report the outcomes of their in-house safety-testing and gives extra whistleblower-protection to staff of these corporations. Amodei stated the corporate had employed each Republicans and Democrats to coverage positions and that its fashions had been much less politically-biased than some rivals’. You’ll be able to learn extra from my Fortune colleague Beatrice Nolan here.
OpenAI researcher retracts announcement of math breakthrough. OpenAI researchers, together with Sebastien Bubeck, claimed that their mannequin GPT‑5 had solved 10 tough math issues that had been initially proposed by mathematician Paul Erdős (who died in 1996) and had remained unsolved thus far, and that it had made vital progress on 11 others. However Thomas Bloom, who maintains the net listing of “Erdős issues” shortly debunked this, displaying that GPT-5 had merely discovered obscure printed options that had not beforehand been delivered to his consideration, relatively than producing new proofs. The incident sparked sharp trade criticism—Google DeepMind CEO Demis Hassabis referred to as the scenario “embarrassing”—and highlighted issues about whether or not the mathematical and scientific reasoning capabilities of many AI fashions is overhyped. You’ll be able to learn extra from TechCrunch here.
AI researcher Andrej Karpathy says AGI remains to be at the very least a decade away. Andrej Karpathy, a outstanding AI researcher who was a part of the founding group at OpenAI and headed AI at Tesla for a variety of years, appeared on the Dwarkesh Podcast and stated that synthetic common intelligence (AGI) is probably going nonetheless a few decade away. His place contradicts claims by a variety of AI researchers at corporations corresponding to OpenAI and Anthropic that human-level AI is imminent. Karpathy argued that at this time’s AI brokers “simply don’t work,” describing them as unreliable, unintelligent, and incapable of dealing with advanced, steady duties, and stated AGI will emerge steadily relatively than by way of a sudden breakthrough. Karpathy added that progress ought to be considered by way of the lens of regular financial and technological progress relatively than hype about machines changing people anytime quickly. You’ll be able to hear and watch the podcast episode here.
EYE ON AI RESEARCH
Google researchers use AI to assist spot genetic drivers of most cancers. Researchers on the tech large created DeepSomatic, a brand new open-source AI mannequin that helps scientists analyze most cancers genomes extra shortly and precisely. The instrument—which relies on a convolutional neural community, an older type of AI structure that’s significantly good at analyzing visible information—is ready to distinguish between genetic mutations an individual is born with and people who develop in most cancers cells. In early exams, it outperformed present strategies of detecting these cancer-related genetic modifications, making it particularly helpful for learning hard-to-analyze cancers like childhood leukemia and mind tumors. Google is open-sourcing each the AI mannequin and the coaching information set it used to create it. You’ll be able to learn Google’s weblog put up on the analysis here.
AI CALENDAR
Oct. 21-22: TedAI San Francisco
Nov. 10-13: Internet Summit, Lisbon
Nov. 26-27: World AI Congress, London
Dec. 2-7: NeurIPS, San Diego
Dec. 8-9: Fortune Brainstorm AI San Francisco. Apply to attend here.
BRAIN FOOD
The backlash in opposition to the AI-driven information middle increase is rising worldwide. The New York Occasions examined how native communities from Chile to Eire are more and more against information middle development of their backyards as a result of unfavorable environmental impacts and vitality calls for these warehouses filled with pc chips carry. The comprehensively reported story is unquestionably price studying. Practically 60% of the world’s largest information facilities at the moment are situated outdoors the U.S., typically in locations the place electrical energy and water techniques are already fragile, the paper reported. Residents in affected areas report worsening blackouts and water shortages, whereas governments—longing for funding and AI infrastructure—have supplied tax breaks and low-cost land, typically with little regulation or transparency. Tech corporations say the initiatives carry jobs and funding and declare they’re minimizing environmental affect, however critics argue they’re depleting important assets and hiding their true footprint by way of subsidiaries and nondisclosure agreements.
The information middle build-out is sparking a rising backlash, led by environmental activists. Whether or not this backlash will gradual the info middle increase, blacken the status of AI corporations and their merchandise with customers, or hasten the roll-out of AI regulation globally stays to be seen. However it’s positively a development to observe. It’s unclear if the backlash will spur AI researchers to search out different AI strategies which might be much less environmentally-costly—or if AI itself will result in clear vitality breakthroughs, in say fusion energy, which may compensate or negate the environmental harm the know-how is inflicting at the moment. However one can hope.
