Good day and welcome to Eye on AI. On this version…Nvidia sees $1 trillion in AI chip gross sales by the top of 2027…Meta delays the debut of its newest AI mannequin (once more)…Moonshot AI develops a brand new structure for big neural networks…and why we might quickly be worrying about ‘moral crumple zones.’
For the reason that introduction of ChatGPT in November 2022, one of many professions that folks typically declare is now toast is consulting. In any case, what’s it that consultants do? They advise corporations on technique; they assist them restructure their companies to create new organizational designs and processes, typically with the assistance of expertise from third-party distributors; and so they act as suppliers of outsourced companies, or no less than conduits to outsourced companies, similar to buyer help or software program improvement. Properly, a frontier AI mannequin can supply strategic recommendation. It could possibly additionally advise on learn how to restructure a company and about which software program to purchase. AI brokers can really assist sew a few of these programs collectively too. Lastly, AI brokers also can now deal with coding and buyer help. So it’s lights out for consultants, proper?
Properly, it hasn’t turned out that means thus far. AI corporations have found that they want consultants, or “systems integrators” as they’re generally known as within the software program world, to assist them promote their AI brokers, as a narrative in final week’s Wall Road Journal highlighted. The reason being that utilizing AI brokers successfully typically requires various organizational transformation—cleansing up information, redesigning workflows, and serious about learn how to redeploy human employees—in addition to strategic serious about how AI is likely to be used to offer an actual aggressive benefit.
The AI mannequin distributors have discovered they don’t have the assets to offer this sort of recommendation at scale—OpenAI solely has about 70 so-called “forward deployed engineers” who go on website with prospects to assist them implement options primarily based on their AI fashions; Anthropic is assumed to have the same quantity. And whereas it’s attainable that AI itself might serve this perform, AI nonetheless suffers from a belief deficit—most boards would nonetheless quite put their religion in recommendation from McKinsey or BCG than ChatGPT. (A extra cynical take: CEOs nonetheless like to make use of consultants to justify their very own selections to boards, in addition to to have another person in charge if all of it goes mistaken.)
OpenAI has shaped what it calls its Frontier Alliance with McKinsey, Boston Consulting Group, Capgemini, and Accenture to assist purchasers use its Frontier platform for constructing and managing AI brokers. (You may learn my protection of that announcement right here.) Anthropic has struck comparable offers with Deloitte, Accenture, and Cognizant and is reportedly in talks with personal fairness teams, similar to Blackstone, to implement Claude-based options of their portfolio corporations.
I just lately caught up with Capgemini’s Chief Technique Officer Fernando Alvarez to speak about how his agency is viewing the way forward for consulting in an AI world.
Area experience issues
First, Alvarez says that whereas each shopper needs to make use of AI brokers, additionally they acknowledge the necessity to govern these brokers, ensure there may be ample cybersecurity round them, and guarantee they’ll work together with legacy programs and fragmented information sources. Advising purchasers on all of that stuff and infrequently serving to them construct it has been Capgemini’s bread and butter. He says purchasers nonetheless need Capgemini to offer these companies. They aren’t prepared at hand it off to AI.
The opposite large promoting level for the consulting corporations, Alvarez says, is deep trade and area experience. The frontier AI labs don’t have the experience in learn how to optimize a pharmaceutical manufacturing plant or the easiest way to run logistics for a fast-fashion retailer. Consulting corporations do. And that makes a distinction when attempting to make use of AI brokers efficiently. Alvarez says the conversations purchasers need to have will not be about what number of brokers you may spin up or the way you orchestrate them. “The conversation is, do you have the domain expertise to understand my problem?” he says.
‘People want the cake, not the recipe’
That doesn’t imply that Capgemini itself isn’t utilizing AI to assist serve purchasers. Alvarez says the large shift that Capgemini, in addition to some opponents similar to Accenture, try to make is to maneuver from promoting expertise and recommendation, to promoting outcomes. On this mannequin, the consulting agency takes on the chance of attempting to determine learn how to ship, say, higher buyer help, whether or not that’s by enterprise course of outsourcing to people in decrease wage nations, such because the Philippines or India, or by AI brokers.
“At the end, people want the cake,” he says—not a tour of the elements or the recipe. The brand new pitch boils right down to a easy proposition: “Here is the problem. Here is the risk I’m willing to take, and this is the outcome I give you.” The shopper pays for the end result: improved KPIs like profitable buyer subject resolutions and improved internet promoter scores. The distinction too is that the consultants on this mannequin cost for the end result, not by the variety of individuals deployed on a venture as some consultants have historically billed.
Alvarez says that AI can also be enabling Capgemini and different consulting corporations to maneuver into market segments, similar to midmarket corporations, that it couldn’t service beforehand as a result of the economics didn’t make sense. The engagements typically required extra workers and value than the shopper was keen to pay for. However now AI has lowered these staffing and value necessities, that means that Capgemini can supply an answer at a worth level that’s enticing to midmarket corporations whereas sustaining a good sufficient revenue margin.
Maybe the most important problem for consulting corporations, although, is retraining their very own individuals to work alongside AI brokers. “Some people will make it, some people will not,” Alvarez says.
For all of the disruption, Alvarez is unmistakably energized. He calls this second “probably the best opportunity I’ve seen in the history of technology.” The query now’s whether or not Capgemini and different consultants can rewire themselves as quick because the expertise calls for—which is, in fact, precisely what they’re advising their purchasers to do.
FORTUNE ON AI
‘The Karpathy Loop’: Former OpenAI researcher’s autonomous brokers ran 700 experiments in 2 days—and gave a glimpse of the place AI is heading—by Jeremy Kahn
Nvidia simply forecast $1 trillion in AI demand. So why isn’t Jensen Huang a goal of AI backlash?—Sharon Goldman
Elon Musk admits xAI ‘wasn’t constructed proper’ as solely 2 co-founders stay and its largest AI wager stalls out—by Marco Quiroz-Gutierrez
AI is reviving tech sectors that VCs had all however forgotten—by Lily Mae Lazarus
AI IN THE NEWS
EYE ON AI RESEARCH
Moonshot AI says it has discovered a greater method to configure and prepare massive neural networks. The Chinese language AI firm, which is finest recognized for its Kimi open-source agentic fashions, has been getting lots of consideration amongst Chinese language AI watchers this week for debuting a brand new tweak on constructing massive, transformer-based neural networks. Known as “Attention Residuals,” the concept is that quite than apply mounted weights to each layer of a skilled community—which is what commonplace neural community architectures do—the community learns to use totally different weights to totally different lower-level layers relying on the duty the mannequin is engaged on. To keep away from the computational problem of doing this for a whole lot of layers in massive neural networks, the Moonshot researchers additionally developed what they name “block attention residuals,” by which a number of layers are organized into blocks, with their outputs summarized as a single unit, and the residual consideration mechanism is then utilized to those blocks as a substitute of the person layers themselves. The tactic improves the velocity and stability of coaching a big transformer-based neural community, which is what all of at present’s main AI fashions are. The Moonshot researchers stated the structure delivered a 1.25 occasions enchancment in computing effectivity in comparison with the usual structure. You may learn the paper right here on arxiv.org.
AI CALENDAR
March 16-19: Nvidia GTC, San Jose, Calif.
April 6-9: HumanX 2026, San Francisco.
June 8-10: Fortune Brainstorm Tech, Aspen, Colorado. Apply to attend right here.
June 17-20: Viva Tech, Paris.
July 7-10: AI for Good Summit, Geneva, Switzerland.
BRAIN FOOD
O’Connor mentions two neologisms that caught my consideration. Two Wharton enterprise faculty professors apparently have coined the time period “cognitive surrender” to discuss with conditions by which individuals merely assume the AI is aware of finest and give up all judgment to the AI system. It’s an excessive type of the phenomenon referred to as automation bias. (O’Connor notes some research by which persons are put in self-driving vehicles and informed they’ll intervene to stop the automotive from hitting one thing, however nonetheless enable the automotive to roll immediately into an apparent impediment.) She additionally mentions the time period “moral crumple zone” (coined by educational Madeline Clare Elish) which refers to advanced programs by which the function of people is lowered to being there to soak up the blame when one thing goes mistaken regardless that the system’s complexity and velocity renders significant human management over the system unattainable.
Count on to begin seeing these phrases crop up an increasing number of in our AI conversations.
Fortune AIQ: One Technique, Actual AI Outcomes
AI transformation may be overwhelming—there are numerous instruments, methods, and approaches competing for consideration, typically with none confirmed outcomes. However for some corporations, AI’s largest affect may be attributed to a single instrument, technique, or strategy. Discover all of Fortune AIQ, and skim the most recent playbook under:
–The way to determine if gen AI is the proper path on your product
–Inside Financial institution of America’s ‘build once’ AI technique
–Why Pinterest goes all in on open-source AI
–How slicing out product administration enabled Kilo to compete within the hyper-fast AI coding market
–How Seismic’s AI incubation staff turned its final AI technique
In 2001, Fortune first convened “The Smartest People We Know,” bringing collectively CEOs and founders, builders and traders, thinkers and doers. Since then, Fortune Brainstorm Tech has been the place the place daring concepts collide. From June 8–10, we are going to return to Aspen—the place all of it started—to mark 25 years of Brainstorm. Register now.
