This is one in a series of posts with excerpts from our new fiction mystery book, The AI Analyst (click on badge on left to go to the Amazon page with description, sample download, reviews etc)
News about DeepSeek from China and two conversations this week – one on Eric Kimberling’s show and questions from his global audience and one with Dion Hinchcliffe about Futurum and Kearney’s recent survey of CEOs and their thoughts on AI projects led me to reflect on so many automation topics that characters in the book touch on.
Here are a few of them:
Ethics around AI
One of the investigators in the team investigating the disappearance of Barry Roman, the SV billionaire CEO of Polestar asks Patrick Brennan, the main analyst character who is assisting them “We hear a lot about ethics in AI. Is there a possibility that Barry suddenly caught religion? Could Barry’s disappearance be a calculated move to expose vulnerabilities in the very systems Polestar has created, forcing a reckoning with AI’s societal impact?”
Patrick responds “I am a technology analyst, not a philosopher, but like you say ethical AI is a hot topic. Many people are behaving like ethics only applies to AI. Every new technology brings out new ethical issues for society. As far back as the 1940s, Isaac Asimov, the science-fiction writer, developed an ethical framework for robots. These days, as we evolve autonomous driving, the auto industry is being forced to answer the question—if an accident is unavoidable, should the vehicle protect the driver or a baby in a stroller on the sidewalk?’ When I was part of Polestar’s executive team, our philosophy was that society broadly has to play Socrates, and our politicians then enact laws to embody that guidance. Individual companies can only go so far when it comes to defining ethical boundaries. You also have to realize that the ethical AI conversation is being stoked by academics and activists and they often talk in the abstract. You have to look at ethics as they apply to specific scenarios. We adopted a best practice from the healthcare world. Many hospitals can quickly convene a committee of nurses, local priests and rabbis, lawyers and healthcare ethicists. Such a committee is available to doctors or to family members of a patient for guidance on thorny issues from genetic testing to euthanasia. Our Chief Counsel, Maria, was responsible for convening our ethics committee, if and when needed.”
Beyond LLMs and GPUs
While there are a whole host of questions around DeepSeek, one of its most promising aspects is we could lower the current dependence on expensive LLMs and GPUs.
In the book, Fabrice Legrand who heads Polestar Ventures describes their own attempts to do so, including via acquisitions. “When we got to the due diligence stage for Amin’s software, all the references came back positive. Tina Chang, who heads our R&D, was intrigued by its architecture. I am not a technologist but she explained to me that they were using large language models for some of the processing, then branching out to small language models and to what they called narrow models for others. They said it would minimize utilization of expensive graphical processing units, or GPUs, and their energy demands. Also smaller models tend to be easier to reuse across many functions. As you may know, Nvidia is going gangbusters with its premium-priced GPUs. Let me give you a quick tutorial on the massive computing power it takes to process the prompts millions are posing to ChatGPT, Claude, and other large language models….The issues arose when we tested the software on our large test database. It required us to parse the data into smaller chunks to feed their models and it still caused performance issues. So we examined the software code and found that it was poorly written. It could never scale. Meaning, it could never process the large data volumes needed to make it valuable. We also got concerned about the data fragmentation. Customers increasingly want detailed audit trails on how AI generates results. You may have heard about machines ‘hallucinating’—when misleading results can be caused by insufficient training data, incorrect assumptions made by the model, or biases used in the training data.”
AI hype cycles
In a presentation at an Analyst Summit, after singing a line from the kid’s song: “Sally the Camel has no humps” Tina says “Now if AI were a camel, it would have at least six humps. You know Gartner has hype cycles for emerging technologies. It shows how different technologies go through expectations of success. If Gartner had a time series going back to the ’50s, it would show multiple AI hype cycles.”
She proceeds to describe several of them in the book, the turns to the present “The most exciting recent development is LLMs and GenAI. You guys, of course, keep the scorecard. ChatGPT from OpenAI reached 100 million monthly active users just two months after launch, making it the fastest-growing consumer application in history. I’ve been exposed to some of the scaling problems they hit scampering to beg, borrow, or steal GPUs. And that of optimizing the KV cache in GPU RAM and moving data at a speed of around 3TB a second. Now, even as vendors like Microsoft, Google, Workday and Salesforce are introducing digital agents, many of you analysts are saying it’s reaching the top of its hype curve. I get much more sobering feedback from geeks in labs. They talk in terms like ‘double descent.’ Look it up in the context of machine learning. They say there is so much we don’t know about these language models. As one said, ‘We are where physics and chemistry were a century ago. Plenty of witches’ brews. So many unknown ‘unknowns.’ AI the camel may just be growing yet another hump. No point telling tech marketing guys and Wall Street to tone down the hype. Or telling the media to not just focus on deep fakes or politicians’ fearmongering that all jobs will be lost to AI.”
AI/Automation and job losses
Sharon, the HR executive at Polestar shares her excitement about the wide range of technologies the company offers – not just software “My job is to focus on employee satisfaction and their productivity, and so it was super exciting to hear about the world of humanoid robots, drones, satellites, machine vision, different kinds of sensors—all turbocharged with AI, which can make all of us super workers in every one of the nearly 900 occupations the Bureau of Labor Statistics tracks.”
Tina adds “The coming convergence of Agentic AI and Humanoid Robots positions us very well. We don’t want to try to be everything to everyone. But if we can keep adding new professions and automating specific tasks for each, our growth will be endless. Other vendors talk about five or six verticals. Well, to us every one of the occupations we address is a unique vertical.
In another setting, Jonathan, one of the new analysts, engages with Patrick : “Henry mentioned that you’ve studied the history of automation in detail, and believe that humans always overestimate its likely impact on jobs. Is that correct?” Patrick responds, “Yes, I can send you a link to a presentation where I talked about a century of automation—UPC scanners in the grocery sector, ATMs in the banking sector, autonomous cars in the auto sector, and numerous others. In many cases, automation actually increased jobs in the sector. It typically targets individual tasks, not complete jobs. It tends to reduce dull, dirty, and dangerous activities. I expect AI will do similar. You have likely heard the expression—‘Those that fail to learn from history are doomed to repeat it.’ But try telling that to journalists and politicians who would rather just take advantage of people’s fear of automation.”
“Who put OUR oil under THEIR sand?”
The DeepSeek news has also led to questions about how OpenAI has collected the large body of data that ChatGPT relies on, and of course, how DeepSeek got its own data and what it will continue to mine from users around the world. In the book, there is plenty of discussion around proprietary data and the many intrusions companies face as they try to protect that data.
Patrick tells an analyst offsite audience “Enterprises will prioritize unique products and market-insight projects, like drug discovery, mineral insights, product design advantages, trading patterns, etc. Smart customers will protect that data for themselves, train their own large language models, or LLMs, and commercialize that data asset. Vendors will continue to generate proposals, job descriptions, demand forecasts, etc. with their AI—clearly useful stuff, but not deserving significant premium pricing. We need to be associated with the first group of customers, the smart ones.”
Another character in the book describes the data piracy we are seeing “Did you know I was in the Service? Spent a fair amount of time in the Middle East. I used to hear the expression often, ‘Who put OUR oil under THEIR sand?’ Sounded funny most of the time. Idle talk, but sometimes it sounded ominously hostile. We are going through something similar in the technology world: ‘Who put OUR data in THEIR data centers?’ There is zero respect for customer data. There’s a land grab happening. Sure you have heard The New York Times sued OpenAI and Microsoft for the unlicensed use of Times articles to train GPT large language models.”
The plot in the book has a China connection and elaborate data piracies and lots of “who put our data in their data center?” twists and turns. Read the book – I don’t want to spoil it.
Automation in Analyst and Law enforcement world
Oxford Research is a next-gen analyst firm which is re-inventing how many occupations and additionally how their own analysts work
“Patrick next invited Henry Novak, the head of Oxford’s labs, to show off the AI copilot named Curmudgeon that Oxford analysts were busy developing. It would access tools like OpenAI and Google Gemini to gather news, reports, and press releases about vendors. It would also tap into Wall Street and other proprietary databases for more vendor analysis. In addition, it would access Oxford’s own client query database to see how corporations were using vendor capabilities. Finally, analysts would enter details from vendor briefings, vendor events, their own observations on each vendor, and their competition. The human expertise would validate and enrich the machine learning. The tool would help generate first drafts of Oxford’s twice-a-year Golden Circle (similar to Gartner's Magic Quadrants) report for each market category.”
Oxford Labs has also developed a digital agent called Sherlock which plays a critical role in the book “Patrick described the tool again and how Henry’s team had developed the core features of the digital tool while working with Ramon (a FBI agent) and that Polestar had licensed and was enhancing for commercial release. Several law enforcement agencies were trying it on pilot projects.”
The SWAT teams in the book use many different bots. Here’s one “a wall-climbing bot packed with heat, vibration, and sound sensors. As it crawled the outside of the building, it transmitted exception data which was painted on a 3D model of the building on the team’s computer. The thick walls made the job difficult but the sensitive sensors managed to capture a location where it appeared to hear someone’s frequent hacking. It also picked up vibrations of a water pipe—presumably the same person cleaning up after a coughing fit.”
Don’t worry - it is a fast paced read with plenty of SV glamor and settings, not a geeky book. However, it covers with plenty of humor and intrigue, the growing China role, the rapid evolution of AI and related infrastructure. And it has plenty of sidebars like California wildfires, Florida hurricanes, Indian weddings, glamorous SV events which required plenty of research and interviews with FBI agents and others.
The next-gen technology analyst and AR
This is one in a series of posts with excerpts from our new fiction mystery book, The AI Analyst (click on badge on left to go to the Amazon page with description, sample download, reviews etc.)
I previously excerpted about next-gen vendors like Polestar and next-gen buyers like Sheldon Freres from the book. Lots of ripple effects from these changes are also causing technology analysts and AR to evolve.
Oxford Research is headquartered in Cambridge, MA but has recently opened a new executive briefing and research center on the waterfront in St. Petersburg, FL. It is hosting an analyst offsite there and lots is discussed about the past, present and future of the analyst world and how they are now using LLMs and copilots and have their own labs for product testing.
A fireside chat between Tucker Newberry, the CEO; Martha Weingarten, the Head of Research and Patrick Brennan, the Chief Analyst sets the stage
“After the golf, the beach, and the Dali, the guests got changed in their rooms at the Vinoy and proceeded to dinner. The group included all the analysts and research staff, as well as six client executives—three from vendors (including Polestar) and three from user organizations. These executives each sat at the head of a table and chatted with Oxford folks about their IT projects and market intelligence needs. There was lots of chatter about Generative AI in particular though most shortened it to GenAI
After dinner, Tucker kicked off the proceedings. He had founded Oxford Research almost 30 years ago and had watched it grow to its present dominating position of helping IT professionals making technology decisions…. “Boy, did we have executive access!” Tucker exclaimed. “I spent many weekends working with CEOs of the largest corporations in the world—often at their beach houses.”…
Martha: “I have mentored a number of analysts throughout my career. Good analysts have two qualities—they are both curious and skeptical...One of her favorite expressions, even today, was “Sacred cows make the best hamburgers.” She described how she once tore apart a vendor presentation: “You are allowed to be stupid or lazy, but not both.”
Tucker and Martha discussed the Gartner IPO in 1993 that literally made hundreds of analysts, overnight millionaires. It led to a glowing New York Times article, from which Martha read a quote: “Gartner may well be the richest publishing house in the world—a ‘mini-Microsoft’ in its field.”
Patrick had been worried that the audience would be bored with this walk down memory lane. Then he saw that even the youngest analysts were listening closely. Few of them knew much about the history and evolution of the technology analyst profession.
Martha said, “But that was decades ago. If you put today’s enterprise applications on a grid of industries and countries, Gartner today barely covers 25 percent. And they have nowhere near the access they once enjoyed to the technology buyer. They make vendors fill out long surveys for their Magic Quadrants (their equivalent of Oxford’s Golden Circle). Vendors, in turn, use a cottage industry of ‘analyst relations’ advisers who coach them how to game their responses. It’s become formulaic—and analysts still cling to application categories which have been around for decades.”
Tucker summarized, “So we need to catch up to the velocity of change in business, not just technology. Clients don’t want to merely read our research and talk to us on Zoom calls. They want customized advice. They still want it in bite-sized chunks, not long projects. But they want us to present it coherently. Nothing annoys them more than being handed off from one analyst to another. They have complex problems and they want us to respond accordingly. We also need to recognize that there are other critical markets we should be analyzing. The Russian invasion of Ukraine showed us we will be dependent on hydrocarbons for a long time. How do we use fossil fuels while neutering their emissions? We should be able to talk authoritatively about carbon capture and storage, and about the total cost of ownership of electric vehicles. Honestly, if I was starting my career today, I would join an energy research firm. Or a healthcare research firm. In the US, nearly a quarter of our GDP is spent on our health and yet our outcomes are miserable. Or look at how the world is changing. So many emerging countries are becoming the ‘new world.’ They’re growing much quicker than the US, EU, China, and Japan. We have an opportunity to help multinationals rebalance their global portfolios and help customers in those fast-growing economies. I don’t want to steal his thunder, but Patrick will talk more about these new horizons tomorrow. Don’t get me wrong. IT, especially AI, will keep us busy for a long time. But it was such an invigorating time that Martha and I experienced in the ’80s and ’90s. There’s no reason we cannot recreate that excitement again, in a variety of new directions. The technology world today feels like it did back then.”
Patrick discussed several new markets
“COVID, the Ukraine war, climate change, and massive digital transformations have made many vertical edge applications viable—telemedicine and personalized medicine in healthcare, EV battery management and billing in utilities, intelligent returns and reverse logistics around eCommerce, direct-to-consumer and related last-mile, small-lot logistics in consumer sectors, CPQ for industrials to handle complex outcome-based pricing which bundles products’ spare parts, all kinds of monitoring and maintenance services . . . the list is virtually endless. There’s also a growing number of application areas aimed at rapidly growing economies around the globe—they must factor in unique business practices, local languages, scripts, currencies, taxes, customs, payroll, and other nuances. Beyond these new vertical and geographic applications, we’re seeing a new generation of AI and data-enabled applications. The vertical and global data sets of most enterprise vendors are skin deep. Given how expensive GPUs and good AI talent is likely to be for the next few years, enterprises will prioritize unique products and market-insight projects, like drug discovery, mineral insights, product design advantages, trading patterns, etc. Smart customers will protect that data for themselves, train their own large language models, or LLMs, and commercialize that data asset. Vendors will continue to generate proposals, job descriptions, demand forecasts, etc. with their AI—clearly useful stuff, but not deserving significant premium pricing. We need to be associated with the first group of customers, the smart ones.”
Patrick next invited Henry Novak, the head of Oxford’s labs, to show off the AI copilot named Curmudgeon that Oxford analysts were busy developing. It would access tools like OpenAI and Google Gemini to gather news, reports, and press releases about vendors. It would also tap into Wall Street and other proprietary databases for more vendor analysis. In addition, it would access Oxford’s own client query database to see how corporations were using vendor capabilities. Finally, analysts would enter details from vendor briefings, vendor events, their own observations on each vendor, and their competition. The human expertise would validate and enrich the machine learning. The tool would help generate first drafts of Oxford’s twice-a-year Golden Circle report foreach market category. The analyst group clapped loudly when Henry finished describing the multifunctional tool. Several analysts offered to be guinea pigs for the project.
Next up was Irene Kaplan, an “AR consultant.” Raised in the public relations world, she now helped vendors make themselves more coherent to analyst firms like Oxford. She had the audience in titters as she shared anecdotes and “inside baseball” stories of what vendors thought of individual analysts. She highlighted “Bill Lou”—the analyst who spoke with a pronounced Southern drawl like the senator from Louisiana, but with devastating impact. And Megan Lewis—Ms. Multitasker, who was physically at one event while tweeting about another on the other coast. And Jean DePasquale, who asked softball questions but made sure he announced his name and company, so it would become part of the event transcript. And “Vinnie Vertical,” who managed to squeeze in a healthcare, banking, or automotive sector question at every event. All super sharp, all quirky.
Talking about quirky, she shared with them that she, too, had built an AI tool, called Gideon—presumably out of respect to the late founder of Gartner—to keep track of analysts with details on their spouses and hobbies, links to their latest research, and more. “It’s fascinating how many analysts are good musicians. One of you has amassed a giant collection of mobile devices, covering the last couple of decades. Another has a collection of soda cans from around the world. A couple of your peers have been to over 100 countries. It is amazing how many of you know the byzantine rules of the game of cricket.”
Patrick made a note to talk to Irene and compare Gideon, Curmudgeon, and Sherlock, a law enforcement digital agent that Henry’s team had developed and which Polestar was now enhancing and commercializing.
Irene then changed tone. “You analysts are under the delusion vendors want your intelligence. They may pretend to, just to stroke your ego. The vast majority of them want you to take their slides and just include them in your research reports. They seem to forget if they can convince you to do that, so can every one of their competitors. Honestly, they should use you more for intelligence. It should bug the heck out of them their customers have 100, 400, or 900 application vendors. Each of them is just a small piece in the customer’s jigsaw puzzle.”
A vendor executive in the audience raised her hand and said, “Irene, not a question, more of a suggestion: Don’t spill our secrets to all these analysts.” Everyone laughed."
There are plenty more angles in the book about analysts and vendor AR.
Don’t worry - it is a fast paced read with plenty of SV glamor and settings, not a geeky book. But read it to see why a buyer like Sheldon Freres or a vendor like Polestar is no longer fiction. And why we need next gen analysts like Oxford and next gen AR to keep up with rapidly changing technology markets.
January 13, 2025 in Agentic AI, Humanoid Robots, Industry Commentary, The AI Analyst - a fiction thriller, Vertical Applications | Permalink | Comments (0)