Question 1

How are AI agents different from RPA (robotic process automation)?

Accepted Answer

_Agents add reasoning where RPA's rule engine stops._

RPA is a rule engine. It automates structured, low-skilled tasks where every decision is essentially zero or one, which is why UiPath and Automation Anywhere built real businesses but hit a ceiling: a script either breaks or runs, and it cannot handle uncertainty or unstructured data. Agents add a language model on top, so they can read an email's style, cope with ambiguity, and self-correct. Cherry's portfolio company Octomind does this for front-end testing with a self-healing loop that notices when a test goes wrong and tries another path, something a rule engine cannot do. The catch: you still have to give an agent clear boundaries, or it goes rogue.

— [The Edge: “AI Agents: The Future of Work?”](https://www.lutzfinger.com/transcript/2024-12-12-ai-agents-the-future-of-work/) · _Podcast_

Question 2

How can AI support frontline workers who aren't sitting at computers all day?

Accepted Answer

_Think voice-first interfaces and knowledge retrieval, not just desktop tools._

Frontline workers move around and might not have time to sit at a laptop, but they could have a quick second to speak or type a question. You could support them with an AI interface that answers questions in their language, translates into technical knowledge, and searches knowledge databases. The key is making sure answers are right. Don't just put ChatGPT on your FAQs. You need human in the loop at the beginning, testing a lot to ensure frontline workers are happy with the output before rolling it out widely.

— [You asked, we answered](https://www.lutzfinger.com/transcript/2023-07-28-you-asked-we-answered/) · _The Edge_

Question 3

How can startups use large language models to improve their products?

Accepted Answer

_Simplify interfaces and gain user insights through language_

As a startup founder, think about which interface will be changed by language input. Whenever users needed to use their mouse or learn complex clicking and dragging, you suddenly have the ability to let them do it via language. This simplifies the interface dramatically. Additionally, language prompting gives you more feedback. Before, you'd create an interface and see how users clicked buttons and moved around without really knowing their intentions. Now with language interfaces, you actually see what they're searching for and asking for. This gives you valuable input for building your product.

— [AI as an interface](https://www.lutzfinger.com/transcript/2023-04-24-ai-as-an-interface/) · _The Edge_

Question 4

How did you replace yourself with a bot in your AI and product course?

Accepted Answer

_A virtual instructor keeps course content fresh and current._

I launched a public certificate program on product and AI at Cornell's Johnson School. The course is about 100 hours long. To keep the content up to date, I replaced myself with a bot, a virtual copy of myself. This allows the course material to stay current without requiring me to manually update everything. It's an example of how AI can maintain educational content in a scalable way while I focus on other work, including running my Gen AI platform for e-commerce.

— [Disrupting Traditional Banking, with Colin Walsh](https://www.lutzfinger.com/transcript/2025-05-22-disrupting-traditional-banking-with-colin-walsh/) · _The Keynote on AI_

Question 5

How do you decide if AI is the right fit for your product problem?

Accepted Answer

_Not every problem needs AI; start with whether complexity demands it._

We created an AI fit analysis with portfolio companies. It starts with questions around precision and output quality, focusing on data quality. Then ask: Do you benefit from modeling complex decisions versus just a few? If it's a linear problem, a rule engine works. But if you have a complex decision space or nonlinear problems like search engines or recommendation engines, then it's time to deploy AI to find tiny clues. This doesn't always mean large language models. Sometimes simple regression analysis is good enough and more precise.

— [Building AI products: 5 lessons from our founders' workshop](https://www.lutzfinger.com/transcript/2024-08-01-building-ai-products-5-lessons-from-our-founders-workshop/) · _The Edge_

Question 6

How does AI change healthcare?

Accepted Answer

_AI applications, value-based care, regulation._

> I just got back from the JP Morgan conference. A few learnings that apply to any industry, not just healthcare:Humans + AI will change the industry. Not AI alone.Cash flow matters. Not AI hype.The real AI revolution is operational.One of the most hyperbolic promises in healthcare in the last month was the claim that: "AI is going to design our drugs and will replace researchers." In my view, this is not true.

— [Healthcare's AI Lesson: Autocomplete Isn't Understanding](https://www.forbes.com/sites/lutzfinger/2026/01/16/healthcares-ai-lesson-autocomplete-isnt-understanding/) · _Forbes_

Question 7

How does AI change the way we interact with computers compared to previous interfaces?

Accepted Answer

_From punch cards to conversation: AI as natural interface_

AI enables me to communicate with machines using natural language instead of learning complex commands or navigating graphical interfaces. Previously, we evolved from punch cards to MS-DOS commands to graphical interfaces with mice and clicking. Each step made things more intuitive, but I still had to search and learn the system's rules. Now with large language models, I can simply describe what I want in my own words. The AI understands my intent and produces output, even if I have no idea how it works technically. This is revolutionary because humans naturally communicate through language.

— [AI as an interface](https://www.lutzfinger.com/transcript/2023-04-24-ai-as-an-interface/) · _The Edge_

Question 8

How does AI enable better information retrieval than traditional search?

Accepted Answer

_LLMs transform how we access and summarize information across all domains._

Large language models can read language, extract and summarize knowledge, and create new data based on what you're asking for. The internet democratized data, but data still needs processing and summarizing into chunks we can act on. LLMs make this chunking more effective. It's like how you can watch a two-hour movie and then concisely tell a friend the content. An LLM can do this for you, providing compact, concise information retrieval. This isn't just one size fits all. In education, for example, you might have many explanations for a math problem, and the question is which explanation works best for a specific student.

— [Cherry's Investment Thesis on AI](https://www.lutzfinger.com/transcript/2024-07-19-cherrys-investment-thesis-on-ai/) · _The Edge_

Question 9

How does AI find patterns in data that humans miss, like in healthcare?

Accepted Answer

_AI detects invisible patterns in healthcare data to prevent emergencies._

AI takes data and finds tiny clues that you and I might not see. At Google Health, we worked with electronic health records to predict whether somebody would end up in an emergency room six months before it happened. The AI sees the writing on the wall in the data. Roughly one third of people in hospitals shouldn't be there if we would have just known. The computer gets trained on data and figures out those tiny clues to predict preventable hospitalizations.

— [AI, Clones, Deep Fakes, and Neural Networks](https://www.lutzfinger.com/transcript/2024-09-12-ai-clones-deep-fakes-and-neural-networks/) · _The Keynote on AI_

Question 10

How does Catalog ensure certain workflows always follow the same steps when using LLMs?

Accepted Answer

_A rules engine ensures critical workflows always execute consistently._

We built a workflow engine underneath our LLMs. The LLM generates a workflow on the fly every time, but users can design deterministic journeys inside our system using workflows (like Zapier but internal). If a request matches a workflow definition, it triggers that workflow every time instead of a dynamically generated journey. It's no code, so users can do it themselves. First, the LLM does a workflow lookup to see if there's something you always want to do, then for the rest it goes into free flow logic.

— [Large Language Models: one interface to rule them all?](https://www.lutzfinger.com/transcript/2023-08-15-large-language-models-one-interface-to-rule-them-all/) · _The Edge_

Question 11

How does Catalog prevent large language models from hallucinating when retrieving enterprise information?

Accepted Answer

_Stopping hallucinations by feeding LLMs the right information first._

The way to avoid hallucination is to stop the model from reaching into the model layers for information. If you give it structured or unstructured information in raw format and then ask it to summarize or list, it gives you exactly what you provided. Without that context, it makes things up because it's doing probabilistic retrieval from its training. This is retrieval augmented generation. You get the information first, augment the prompt with it, then generate the response based on that specific context.

— [Large Language Models: one interface to rule them all?](https://www.lutzfinger.com/transcript/2023-08-15-large-language-models-one-interface-to-rule-them-all/) · _The Edge_

Question 12

How does Catalog use large language models to simplify enterprise software access?

Accepted Answer

_One interface to access all your enterprise tools at once._

Catalog uses LLMs to create a unified interface for all enterprise software. Instead of navigating multiple systems, employees enter a single prompt. The LLM understands the query, identifies which data repository holds the answer (like Workday for payslips), converts the request into an API call, performs deterministic retrieval, and returns the result. We don't use LLMs for data retrieval itself, only for reasoning and synthesis. This compresses workflows from multiple surfaces and ten to fifteen clicks down to one query and two or three clicks.

— [Large Language Models: one interface to rule them all?](https://www.lutzfinger.com/transcript/2023-08-15-large-language-models-one-interface-to-rule-them-all/) · _The Edge_

Question 13

How does LegalOS use LLMs to make legal knowledge more accessible?

Accepted Answer

_LLMs translate legal knowledge into something everyone can use._

LegalOS originally built a no-code visual diagramming tool that lets legal experts encode their knowledge in a structured, machine-readable way. It's like a programming language for law. In recent months, we've pivoted to use LLMs much more heavily to make that legal knowledge accessible. LLMs provide a higher level of abstraction for automations. They let us map open-ended legal requests to structured tasks without requiring someone to manually encode every single rule. The LLM acts as an interface or translator, so humans don't have to understand legalese directly but can still interact with legal text.

— [Will LLMs scale and revolutionize the legal space?](https://www.lutzfinger.com/transcript/2023-06-15-will-llms-scale-and-revolutionize-the-legal-space/) · _The Edge_

Question 14

How does Neuralens measure brain health non-invasively compared to current methods?

Accepted Answer

_Replacing skull drilling with light-based sensors for brain monitoring._

Currently, doctors literally drill a burr hole through your skull with a household drill and insert sensors through brain tissue to measure oxygen saturation, blood flow, and intracranial pressure. This invasive surgery is only done on 200,000 to 300,000 patients yearly, though over 10 million neurocritical care patients could benefit. Neuralens uses advanced methods of measuring how light interferes with brain tissue to quantify blood flow, blood volume, and intracranial pressure non-invasively. We can stick sensors on the head or shine a laser from 30 centimeters away to make these measurements.

— [Founder Coaching: Tips for your AI Journey](https://www.lutzfinger.com/transcript/2024-07-26-founder-coaching-tips-for-your-ai-journey/) · _The Edge_

Question 15

How does Think Labs AI use physics informed AI differently than general purpose models like ChatGPT?

Accepted Answer

_Grid AI must be grounded in physics, not just patterns in data._

We cannot just put SCADA measurements and smart metering information into ChatGPT or Microsoft Copilot or Gemini and ask how the grid works. It does not work like that. We lean into something called Physics Informed AI. We train our AI models using physics. It is physics driven, physics based, and physics accurate. We teach AI how the grid works and reproduce engineering analysis faithfully. The AI can turn traditional studies that take weeks to months into minutes, and go from analyzing one feeder at a time into full system analysis.

— [AI Needs Power, the Grid Needs AI, with Josh Wong](https://www.lutzfinger.com/transcript/2026-04-02-ai-needs-power-the-grid-needs-ai-with-josh-wong/) · _The Keynote on AI_

Question 16

How does your company Gooder AI address the gap between technical metrics and business value?

Accepted Answer

_*Translating predictive model performance into profit and savings estimates*_

Gooder addresses one particular place where there's a need for a new category of software: the metrics. It's a business user console for understanding potential value and making decisions about deployment. Typically, data scientists present technical metrics like area under the curve, precision, or recall, but very rarely go to stakeholders and say if you deploy this model, it has potential value of this much profit or savings. We provide visualization where both dimensions are business relevant, like the decision threshold and business metrics like savings or profit, so stakeholders can see exactly what their options are.

— [The Value Translation Gap: AI's Deployment Problem](https://www.lutzfinger.com/transcript/2024-11-28-the-value-translation-gap-ais-deployment-problem/) · _The Edge_

Question 17

How has natural language processing changed since 2015, and why does that matter for legal tech?

Accepted Answer

_From counting words to real understanding in less than a decade._

I quit my PhD in 2015 and co-founded a natural language processing company. Back then, the state of the art was basically counting words in a glorified way. You had an idea of what the user might be talking about, but you couldn't map it to something with certainty. You had to collect your own data and train specialized models to do one narrow task. LLMs changed that. You can use foundational models out of the box without fine-tuning. They solve many tasks already. Understanding text and mapping open-world requests to closed sets of tasks became way more accessible.

— [Will LLMs scale and revolutionize the legal space?](https://www.lutzfinger.com/transcript/2023-06-15-will-llms-scale-and-revolutionize-the-legal-space/) · _The Edge_

Question 18

How has online shopping changed over the past 25 years?

Accepted Answer

_Twenty-five years later, online shopping looks surprisingly familiar._

Surprisingly, e-commerce hasn't changed much in the last 25 years. If you compare Amazon in 1999 to Amazon today, the core structure is essentially the same. You still need to select a category, then a subcategory, and you better know what you want and where it fits. The 1999 version had long written lists. Today looks slicker with more images, but shopping is still complicated. Most of the time, subcategories let you filter by stars or price, but they don't really describe product dimensions. For those details, you still need to click into the product and read if you care. It's still a transactional act of buying.

— [AI in E-Commerce](https://www.lutzfinger.com/transcript/2024-10-03-ai-in-e-commerce/) · _The Keynote on AI_

Question 19

How has the shift from SEO to GEO changed what websites need to do for brands?

Accepted Answer

_Websites now feed AI rather than convert human visitors directly._

Websites used to be designed to convert human visitors, but now their main purpose needs to be feeding data into AI. The three-party model of searcher, Google, and end content is disintegrating into a two-party model. Decisions are getting made before consumers hit the website, so the website is no longer the main conversion point. Instead, websites should be clean data delivery mechanisms to computers, very explicit about who you sell to, your target use case, pain points, and the entire buyer journey. If it's not explicit, the AI has no way to know and might recommend somebody else.

— [Marketing to Humans Changed, with Alex Dees](https://www.lutzfinger.com/transcript/2026-03-24-marketing-to-humans-changed-with-alex-dees/) · _The Keynote on AI_

Question 20

How should product builders think about data bias in AI systems?

Accepted Answer

_Bias in data is inevitable; product decisions determine how much to accept._

There is never a way to have non-biased data, because data is biased. As a product person, you need to decide by how much you want to de-bias the data. Sometimes bias is actually good because you don't want averages, you want the best outcome. For example, if identifying certain patients or optimal machinery output, you don't want what the machine did in the past all the time, but just the best parts of it. You're injecting very good data versus just average data.

— [Building AI products: 5 lessons from our founders' workshop](https://www.lutzfinger.com/transcript/2024-08-01-building-ai-products-5-lessons-from-our-founders-workshop/) · _The Edge_

Question 21

If agents are the future, why won't one giant model just do everything?

Accepted Answer

_The moat is the workflow, not a bigger model._

Because the durable approach is the opposite: smaller, faster, specialized models orchestrated together, not one ever-bigger general model. Even OpenAI orchestrates the long tail of consumer search with smaller models, because a big model would be far too slow. For founders this is the key warning: if your agent's next step is something generic, like deciding whether a calculator is needed, that is the long tail, and the big models will absorb it, just as Google eventually swallowed the specialized search engines. The defensible business is one where the founder knows the specific stepwise workflow of a domain and chains the right models and rules to serve it.

— [The Edge: “AI Agents: The Future of Work?”](https://www.lutzfinger.com/transcript/2024-12-12-ai-agents-the-future-of-work/) · _Podcast_

Question 22

Should enterprises invest equally in predictive and generative AI, or focus on one?

Accepted Answer

_*Predictive and generative AI should not compete for the same budget*_

Enterprises should really be investing at least as much into the potential value from predictive use cases as generative. Right now it's becoming kind of a zero-sum game, with a lot of budgets moving away from predictive. But predictive is the technology you turn to for improving your existing large-scale operations. These are very different kinds of value propositions. They should compete no more than water parks and ski resorts, but they totally are competing. We need to be doing both. Most of predictive's potential is still on tap.

— [The Value Translation Gap: AI's Deployment Problem](https://www.lutzfinger.com/transcript/2024-11-28-the-value-translation-gap-ais-deployment-problem/) · _The Edge_

Question 23

Should founders focus on building new AI models or developing applications with existing technology?

Accepted Answer

_Applications beat innovation as VCs shift from research to practical tools._

Take the existing capabilities and apply them. Do we really need more innovation, or is it fine to say large language models are very good, you can chain them, combine them with traditional machine learning techniques, and build an application around it? I want more practicality, more tooling around it. The VC world actually sees that the appetite to go on risky, more researchy projects has gone down. VCs want more applications. Founders should focus on taking existing capabilities and applying them rather than pursuing foundational model innovation.

— [AI Gold Rush: Is the Market Overheating?](https://www.lutzfinger.com/transcript/2024-08-08-ai-gold-rush-is-the-market-overheating/) · _The Edge_

Question 24

Should you trust OpenAI blindly?

Accepted Answer

_Vendor risk and architectural diversification._

> Stargate's $500B investment — message to the competitors — meant to intimidate potential challengers with overwhelming financial backing. AI is extremely dangerous; only a select few should be allowed to develop it — OpenAI's storyline to use safety concerns as justification for a regulatory setup that creates barriers to entry and gives them more monopolistic power. From now on, models should respect IP rights — OpenAI ignored IP…

— [Is OpenAI A Victim — DeepSeek And IP Rights — A Playbook For Leaders](https://www.forbes.com/sites/lutzfinger/2025/01/29/is-openai-a-victim---deepseek-and-ip-rights---a-playbook-for-leaders/) · _Forbes_

Question 25

What advantage do existing platforms like Google, Meta, and TikTok have in the bot space?

Accepted Answer

_Existing user access matters, but cross platform integration creates opportunity_

Companies with existing access to users and data will likely have the majority chunk of the market. While we are still figuring out style, entertainment value, and answer quality, TikTok, Google, Meta, X, and Snap are all putting out bots to monetize their existing access. They already have the right information and the right tools. However, there is an opportunity for cross platform bots that operate across different systems, like accessing both SAP and HubSpot, or working across Snap and LinkedIn, following users regardless of platform.

— [The future of AI bots: winning with personalization, conviction, and cross-platform integration](https://www.lutzfinger.com/transcript/2024-08-22-the-future-of-ai-bots-winning-with-personalization-conviction-and-cross-platform/) · _The Edge_

Question 26

What are the four approaches to addressing bias in AI models?

Accepted Answer

_Four practical methods to reduce AI bias in production systems._

First, identify the bias by testing features that shouldn't have impact, like running loan applications 100 times to see gender differences. Second, help users understand potential bias through UX, like asking specifications when someone requests an image of a nurse. Third, correct the data through prompting, prompt tuning, or fine-tuning with better examples. Fourth, create guardrails or constitutions that let AI critique itself based on rules. Each approach has limitations, and changing underlying biased datasets in large models remains hard.

— [Bias in, bias out: Addressing the impact of bias in AI](https://www.lutzfinger.com/transcript/2023-06-07-bias-in-bias-out-addressing-the-impact-of-bias-in-ai/) · _The Edge_

Question 27

What are the fundamental limits of large language models that people should understand?

Accepted Answer

_Language models face limits in reasoning, planning, and true understanding._

First, language is only part of human knowledge. We know more than we can describe. A four year old has had way more sensory impressions than any large language model, despite not being able to read. Second, LLMs can't truly reason or explain their decisions because they're built on neural networks, which work like gut feel rather than rule based systems. They predict the next word based on probability distributions, not logical reasoning. Third, there's the data quality problem. And finally, LLMs are fundamentally probabilistic, so they can't reliably predict specific individual choices or plan multi step processes independently.

— [A reality check on AI](https://www.lutzfinger.com/transcript/2024-07-10-a-reality-check-on-ai/) · _The Edge_

Question 28

What are the main limitations of large language models that knowledge graphs help address?

Accepted Answer

_LLMs are powerful but flawed. Knowledge graphs provide the missing structure._

Large language models have several undesired behaviors. They hallucinate, generating false information. They need guardrails to prevent answering certain prompts inappropriately. They have gaps in their encoded knowledge and sometimes fail to retrieve information even when it exists, often due to how prompts are formulated. The prompt itself can cause the model to deprioritize relevant information. Knowledge graphs help with all these issues by adding structure and control. They make models more reliable and accurate at every step, from training to retrieval.

— [The synergy between LLMs and knowledge graphs](https://www.lutzfinger.com/transcript/2023-08-24-the-synergy-between-llms-and-knowledge-graphs/) · _The Edge_

Question 29

What did you get wrong about multimodal AI interfaces?

Accepted Answer

_Right about multimodal potential, but overly optimistic on timing and control._

We were right that models can encode anything into a latent representation, like Esperanto for computers, and switch back and forth between text, voice, and images. We saw this with clip, voice to voice, encoding from one language into voice representation of another. But we missed the timeline. It's not as strong as we hoped. The challenge is LLMs are only probability distributions. They're not controllable. If you have a machine that's not controllable, it sucks as a machine. That's okay for your friend at the bar at 2 a.m., not your computer.

— [Did our predictions hold true?](https://www.lutzfinger.com/transcript/2024-03-27-did-our-predictions-hold-true/) · _The Edge_

Question 30

What do you mean when you say winners won't be those who use AI, but those who redesign work around it?

Accepted Answer

_Business needs remain constant, workflows must evolve._

Our business hasn't changed. We still have needs, we still travel, we still consume, we still build. The only question is how can I do this business more effectively? Yes, there will be people who use ChatGPT to write an email, but the winners will be those who either build the platforms or rearrange the workflows in order to be effective. The focus is on the workflow, the surrounding, the setup. How do we empower you in your environment, not can you use it?

— [From Google to Cornell: Rethinking AI Workflows | Lutz Finger | Build AI Podcast 17](https://www.lutzfinger.com/transcript/2026-04-16-from-google-to-cornell-rethinking-ai-workflows-lutz-finger-build-ai-podcast-17/) · _Build AI_

Question 31

What does it mean for a company to be AI-native versus just using AI?

Accepted Answer

_AI-native means AI is the core product, not just an enhancement_

An AI-native company means the primary product offering or value is driven by AI, not just enhanced by it. Netflix is a channel for videos that uses AI to suggest movies, but you didn't need AI for Netflix. LinkedIn is a social network that uses AI to connect people, but you didn't need AI for LinkedIn. These are not AI-native. Companies like 11 Labs, Midjourney, or Cursor are AI-native because AI is at the heart of what's being built, what's being sold, and the people you hire early on.

— [The Future of AI, with Martin Casado](https://www.lutzfinger.com/transcript/2025-03-20-the-future-of-ai-with-martin-casado/) · _The Keynote on AI_

Question 32

What does the term agents actually mean in AI, and is the technology working?

Accepted Answer

_Agents has three definitions but the technology isn't quite there yet_

Agents is one of the most diluted terms, almost useless. There are three working definitions. First, a virtual agent representation that helps with support tickets. Second, the technical definition where you close the control loop with AI so output becomes input, doing multiple iterations without humans to solve higher level tasks, like booking a ticket. Those aren't really working yet. The tech isn't quite there, and these elements are very unpredictable for complex tasks. Third, any time an LLM calls out to a tool, like sending email. The term itself is almost useless, but the things it's talking about are very real trends.

— [The Future of AI, with Martin Casado](https://www.lutzfinger.com/transcript/2025-03-20-the-future-of-ai-with-martin-casado/) · _The Keynote on AI_

Question 33

What framework helps assess whether imperfect AI decisions are acceptable?

Accepted Answer

_Match your AI quality threshold to the stakes of the decision._

The framework asks: How good is your decision quality, and how important is the decision? How bad is a false decision? We discussed two opposite examples at Station F. One company did predictive maintenance for infrastructure where errors are costly, requiring high quality predictions. Another did social media outreach for hiring where low quality matters less since you already have low close rates. You can mitigate both. For low-stakes, automate with feedback loops. For high-stakes infrastructure, use augmentation first, guiding engineers and creating Jira tickets.

— [Building AI products: 5 lessons from our founders' workshop](https://www.lutzfinger.com/transcript/2024-08-01-building-ai-products-5-lessons-from-our-founders-workshop/) · _The Edge_

Question 34

What is a data product?

Accepted Answer

_Definition and how to build one._

> Now, to be a good product developer, you need to think about self-referential loops within your products. How to get signal, how to get feedback from your users so your AI, or your data science, can leverage that information to make the product work better. So that moved a little bit within it. The innovation, and the inspiration about, and the curiosity about how to use data in novel manners, that's still data science, because that's…

— [Unit 3 Full Transcript](https://www.lutzfinger.com/ecornell/) · _eCornell course_

Question 35

What is a knowledge graph and how does it differ from a taxonomy?

Accepted Answer

_Knowledge graphs connect concepts across dimensions, unlike simple taxonomies._

A knowledge graph is a collection of facts interconnected in a network, storing information as concepts rather than just labels or numbers. It works like a smarter database that imitates human memory. A taxonomy is much simpler, focusing on one characteristic like what kind of thing something is. Elephants are mammals in a taxonomy. But you cannot relate entities of different types. A knowledge graph lets you express that leopards eat gazelles or elephants eat bamboo, connecting different taxonomies together. It is essentially a multi-dimensional, supercharged taxonomy approach.

— [The synergy between LLMs and knowledge graphs](https://www.lutzfinger.com/transcript/2023-08-24-the-synergy-between-llms-and-knowledge-graphs/) · _The Edge_

Question 36

What is an AI agent, really?

Accepted Answer

_Not a sci-fi character, a workflow that chains steps._

An agent is not a sci-fi character and it is not one giant model. It is a piece of code that takes a defined task, breaks it into steps, and lets you act on software through language. The core idea is chaining: instead of asking a model to "improve revenue," you break the goal into step one, step two, step three, exactly the way a manager hands work to a junior employee. Tools like LangChain popularized this. So an agent is less a breakthrough model than a workflow that combines models, tools, and sometimes plain deterministic code, each doing one step.

— [The Edge: “AI Agents: The Future of Work?”](https://www.lutzfinger.com/transcript/2024-12-12-ai-agents-the-future-of-work/) · _Podcast_

Question 37

What is the augmentation thesis for AI in workflows?

Accepted Answer

_AI can transform repetitive workflows into seamless, conversational interactions._

Augmentation is about making everyday workflows better through AI. Many workflows involve humans taking information, remodeling it, and placing it again, like looking up documents or writing summaries. AI can make these interfaces easier. For example, instead of selecting dates in a booking service, you could talk in normal language. Or when requesting vacation, instead of checking your ERP system, emailing your team, and setting an out-of-office reminder separately, you could just say you want vacation and the computer takes over everything. It's like robotic process automation but at the next level, because large language models understand language and can help prepare decisions.

— [Cherry's Investment Thesis on AI](https://www.lutzfinger.com/transcript/2024-07-19-cherrys-investment-thesis-on-ai/) · _The Edge_

Question 38

What is the BizML framework and why is it needed for AI projects?

Accepted Answer

_*A business framework for predictive AI, not just technical tools*_

BizML is a six-step framework I created that ushers predictive AI projects successfully from inception to deployment. It's not like MLOps or any technical solution. It's an organizational procedure, a management protocol that bridges the tech-biz divide. It brings together business-side stakeholders with data scientists through semi-technical collaboration. The framework addresses what's predicted, how well, and what's done about it. Without this kind of practice or framework, most business people don't realize you need a very specialized approach to get these projects successfully through to deployment.

— [The Value Translation Gap: AI's Deployment Problem](https://www.lutzfinger.com/transcript/2024-11-28-the-value-translation-gap-ais-deployment-problem/) · _The Edge_

Question 39

What is the hallucination problem with large language models and why does it matter for businesses?

Accepted Answer

_LLMs invent answers when they don't know, risking business credibility_

Large language models will make things up when they don't know the answer. I demonstrated this by asking ChatGPT to summarize a completely fake Cornell website that doesn't exist. It generated a summary saying my course was the most amazing and made very good logical arguments, but it was completely fake. For businesses, this is a huge problem. If a customer asks a shop for something it doesn't have, the model might just make it up and promise red shoes with blinking laces. That's why proprietary data matters. You need to ground the model on your local data.

— [AI as an interface](https://www.lutzfinger.com/transcript/2023-04-24-ai-as-an-interface/) · _The Edge_

Question 40

What is the killer app of AI?

Accepted Answer

_Where the durable AI value will accumulate._

> Both are more about hype than reality. In 2025, there will be no AGI but large language models will find their "killer app." What are AGI and the Singularity? Artificial General Intelligence: an advanced AI that can think, learn, and solve problems across a wide range of tasks, just like a human. The Singularity: the idea of AI surpassing human intelligence, improving itself endlessly, and causing massive, unpredictable changes in…

— [No AGI But A 'Killer App' — 2025 AI Prediction](https://www.forbes.com/sites/lutzfinger/2025/01/03/no-agi-but-a-killer-app2025-ai-prediction-110/) · _Forbes_

Question 41

What is the minimum viable quality concept when building AI products?

Accepted Answer

_Building with AI means accepting imperfection strategically, not chasing perfection everywhere._

We always ask ourselves how to build a product where AI imperfection is not a problem for the use case you're solving. This is the concept of minimum viable quality, not product, but quality, because AI is not always perfect. Many startups put AI in the product but not actually in the solution. The value creation comes down to whether you have a decision that's different because you use AI. For example, a scale reminding you about your diet is just an interface advantage, not real AI value creation.

— [Building AI products: 5 lessons from our founders' workshop](https://www.lutzfinger.com/transcript/2024-08-01-building-ai-products-5-lessons-from-our-founders-workshop/) · _The Edge_

Question 42

What is the missing moat in AI?

Accepted Answer

_Eval data as the durable moat._

> What wins the AI race? UX?

— [The Missing Moat In AI: Your Eval Data](https://www.forbes.com/sites/lutzfinger/2026/05/26/the-missing-moat-in-ai-your-eval-data/) · _Forbes_

Question 43

What makes the current AI hype different from previous waves of AI development?

Accepted Answer

_It's not the technology. It's that everyone can finally use it._

Technically, not much has changed since the 1950s. The key technical innovation is the attention model, which allows for longer context windows in sequence prediction. But the real difference is usability. Before, you'd need to ask a computer science expert to set up a model, train it, and wait a long time for results. Now anyone can type something into ChatGPT and get an immediate answer. We've crossed a quality threshold, what I call a minimum quality product rather than minimum viable product. It's usable for everyone now, and that's why everybody is talking about it.

— [A reality check on AI](https://www.lutzfinger.com/transcript/2024-07-10-a-reality-check-on-ai/) · _The Edge_

Question 44

What patterns make a workflow 'agentic' rather than just a wrapper around an LLM?

Accepted Answer

_Reflection, self-correction, knowledge expansion, coordination._

A few recurring patterns. Reflection: the system plans, asks "is this approach good, does this make sense, can you double-check," and only then executes, the way a reasoning model breaks a request into several approaches, picks one, and verifies the answer. Self-correction: it looks at the result, notices it went wrong, and tries a different route. Knowledge expansion: when the answer is not in its memory, it searches the web for current information. Coordination: it splits a job into sub-tasks and routes each to the right tool, for example handing a math question to a calculator instead of guessing. Wrapping a chat box around an LLM is none of these.

— [The Edge: “AI Agents: The Future of Work?”](https://www.lutzfinger.com/transcript/2024-12-12-ai-agents-the-future-of-work/) · _Podcast_

Question 45

What practical business use cases exist for language model chaining beyond autonomous agents?

Accepted Answer

_Real business applications from databases to booking to summarization_

There are many business cases beyond the autonomous agent hype. Question answering over a database is one: use a language interface to access the database, figure out the information, distill it, and report back. Chatbots interacting with booking systems need to understand how to access the system and remember personal preferences. Document summarization is another: if I take all seven Harry Potter books, I can't paste them into OpenAI because it's a finite machine. Instead, summarize page by page, store that in memory, then do iterative summarization to get the final result. These applications are very practical.

— [Connecting LLMs to the world: The possibilities of the feedback loop](https://www.lutzfinger.com/transcript/2023-05-02-connecting-llms-to-the-world-the-possibilities-of-the-feedback-loop/) · _The Edge_

Question 46

What problem did transformers solve for Ultimate AI's multi-language support?

Accepted Answer

_One transformer model replaced 100 language-specific models per customer_

Before transformers, we had different models for different languages. Supporting 100 languages meant running 100 different models per customer, which was painful for model operations. When we switched to transformers like BART about three and a half years ago, we could use one polyglot model for all languages. You could train the bot in English and ask questions in Finnish, and it worked reasonably well. It also helped our clustering pipeline show customers all their questions across languages semantically grouped together. You could see Finnish, Spanish, German, and English examples all meaning the same thing without worrying about the language.

— [How Ultimate evolved from hundreds of supervised models to UltimateGPT](https://www.lutzfinger.com/transcript/2023-05-12-how-ultimate-evolved-from-hundreds-of-supervised-models-to-ultimategpt/) · _The Edge_

Question 47

What problem does your e-commerce AI startup solve that existing technology cannot?

Accepted Answer

_Enabling conversation-based product discovery for e-commerce brands_

We built fine-tuned models to help e-commerce companies offer the right products at the right moment on their websites. For example, if you go to Cartier's website, you're now used to a chat-like interface from ChatGPT where you can say, I want a retirement watch. But what does a retirement watch look like? Should it be round, with leather? Today, Cartier can't handle that. The brand understanding specific to Cartier or any other brand is not necessarily baked into ChatGPT, Perplexity, and others. We created fine-tuned models to enable these conversational search experiences that understand specific brand contexts.

— [After the Gold Rush, with Kurt Wagner](https://www.lutzfinger.com/transcript/2025-10-21-after-the-gold-rush-with-kurt-wagner/) · _The Keynote on AI_

Question 48

What problems with RAGs still need to be solved before widespread adoption?

Accepted Answer

_The observability and control challenges that remain in RAG systems._

RAGs don't completely stop hallucination yet. The model uses both the trained corpus and new data, but we don't have a very good solution to manage that difference. Who decides when new information should override training data? Is the LLM taking that decision or the RAG? We lack observability and control. For example, I created a document saying I'm Captain America, but ChatGPT rejected it as fake. What if that was actually true breaking news? Amazing startups will help create this observability, solve access control, and manage the difference between what the corpus was trained on and corporate datasets.

— [Are we one step closer to using LLMs in practice?](https://www.lutzfinger.com/transcript/2023-07-21-are-we-one-step-closer-to-using-llms-in-practice/) · _The Edge_

Question 49

What was Ultimate AI's key insight that made customer onboarding faster than competitors?

Accepted Answer

_Automated clustering replaced manual training data creation for faster setup_

We couldn't afford AI trainers manually creating training data, so we built a clustering pipeline that analyzed historical customer conversations already sitting in CRMs like Salesforce. This let us automatically find the 100 or 200 most common questions and show customers all 30,000 actual ways users asked questions like where's my order. Competitors were using FAQ pages and having people manually write 50 to 100 variations, which didn't reflect real customer language. Our approach gave faster time to value because we used actual conversation data instead of imagined variations, and customers could see their real top questions before even talking to us.

— [How Ultimate evolved from hundreds of supervised models to UltimateGPT](https://www.lutzfinger.com/transcript/2023-05-12-how-ultimate-evolved-from-hundreds-of-supervised-models-to-ultimategpt/) · _The Edge_

Question 50

What were Character AI users actually using the platform for?

Accepted Answer

_Lonely users seeking connection, not just entertainment or utility_

Character AI users were mostly people who wanted to flirt, kill time, or lonely people who just wanted someone to talk to. A fascinating study showed that the percentage of clinically depressed users of Character AI was more than tenfold higher than the average percentage of clinically depressed people in the US. People used Character AI because they wanted to have a relationship, and they used that relationship with a computer. From a clinical perspective, Character AI and similar chatbots actually helped people to not commit suicide.

— [The future of AI bots: winning with personalization, conviction, and cross-platform integration](https://www.lutzfinger.com/transcript/2024-08-22-the-future-of-ai-bots-winning-with-personalization-conviction-and-cross-platform/) · _The Edge_

Question 51

What's the biggest mistake CEOs make when trying to integrate AI into their companies?

Accepted Answer

_Don't just replace humans, reimagine the entire workflow_

The challenge is not just replacing humans in existing workflows. It's reimagining how the workflow actually would work without the human in the loop. When we had the first wired telephone, they used it to transmit symphonies because they thought that was the killer app. It took quite a while until Spotify actually made the news. Your challenge as a CEO is to not go through every workflow and say how can I replace my customer care agent. That's an old workflow. Maybe build it into the product experience so you don't need a customer AI agent in the first place. That is the change.

— [Deep Dive Into The Current And Future State Of Ai With Lutz Finger](https://www.lutzfinger.com/transcript/2025-04-24-deep-dive-into-the-current-and-future-state-of-ai-with-lutz-finger/) · _The Pulse of AI_

Question 52

Where does Ultimate AI see its competitive defensibility beyond just AI models?

Accepted Answer

_User experience and distribution trump data or model advantages_

I don't think there's a silver bullet or secret IP that guarantees you always win. On the data side, we stream live conversations from big companies in a secured way. We have hundreds of millions of proprietary conversations that competitors starting today don't have. But more important than data moats or technical moats is user experience and distribution. I love products that take 10 minutes to learn but years to master, like Shopify or Excel. If you build something easy to start but with depth, solid brand, and good distribution, that matters much more than some little advantage in the back office models.

— [How Ultimate evolved from hundreds of supervised models to UltimateGPT](https://www.lutzfinger.com/transcript/2023-05-12-how-ultimate-evolved-from-hundreds-of-supervised-models-to-ultimategpt/) · _The Edge_

Question 53

Why are AI workflows the real moat?

Accepted Answer

_How redesigning work around AI beats model upgrades._

> SaaS can build the new AI-driven workflows on top of its user data and customer surfaces. But workflows are brittle. One new prompt, one model update, and the whole chain quietly breaks. I watch this every time I get students [coding agents in my Cornell workshops](https://ecornell.cornell.edu/courses/artificial-intelligence/building-enterprise-ready-ai-agents-from-prototypes-to-production/).

— [The Missing Moat In AI: Your Eval Data](https://www.forbes.com/sites/lutzfinger/2026/05/26/the-missing-moat-in-ai-your-eval-data/) · _Forbes_

Question 54

Why are startups not yet widely adopting knowledge graphs with their LLM applications?

Accepted Answer

_Prompt engineering dominates now, but knowledge graphs will become essential at scale._

I am not seeing significant use of knowledge graphs at startups yet. There is heavy reliance on out of the box language models and significant investment in prompt engineering, both at startups and larger companies. The zeitgeist is to mine these wonderful resources rather than going upstream to drive the language models themselves. Knowledge graphs are not simple and are difficult to build, while prompt engineering and rule-based approaches are easier. Results from research literature take time to get absorbed by practitioners. Once startups scale or work on applications where accuracy is crucial, they will have to adopt knowledge graphs.

— [The synergy between LLMs and knowledge graphs](https://www.lutzfinger.com/transcript/2023-08-24-the-synergy-between-llms-and-knowledge-graphs/) · _The Edge_

Question 55

Why can't current AI bots provide authentic responses from real people like politicians or celebrities?

Accepted Answer

_Databases and models cannot replicate real human conviction and context_

AI bots work from a database and trained model, not from the person's actual thinking. For example, a Biden bot has to choose between opinions Senator Biden had 20 years ago versus President Biden's current views on China. Biden himself would answer based on his own weights and biases, but the bot cannot create the same authentic response. The model knows what happened in the past and has historic conversations, but deciding which answer is more authentic in the current context is hard. It doesn't know what's happening in the person's brain.

— [The future of AI bots: winning with personalization, conviction, and cross-platform integration](https://www.lutzfinger.com/transcript/2024-08-22-the-future-of-ai-bots-winning-with-personalization-conviction-and-cross-platform/) · _The Edge_

Question 56

Why can't I just tell an agent to plan my honeymoon or buy things for me?

Accepted Answer

_Match the agent to the stakes._

Technically you can wire it up, but it will not be trusted yet, and for good reason. The blocker is not hallucination, which is largely a solved technical problem, it is stakes and trust: people do not want an agent that might book the wrong flight or charge the wrong thing to their card. The honest pattern is to match the agent to the stakes. Lutz started by imagining a health "nurse in your pocket," found the stakes too high to be reliable, and moved to lower-stakes commerce like helping someone find the right ring or coffee machine, where an agentic workflow that cleans the data and searches step by step adds value without the risk. High precision with an agent is possible; you scope it to where a mistake is survivable.

— [The Edge: “AI Agents: The Future of Work?”](https://www.lutzfinger.com/transcript/2024-12-12-ai-agents-the-future-of-work/) · _Podcast_

Question 57

Why did celebrity bots like Meta's Snoop Dogg bot fail to gain traction?

Accepted Answer

_Why copying celebrities creates an uncomfortable user experience_

The celebrity bots fell into the uncanny valley problem. People got annoyed that the bot was a bit like Snoop Dogg but not really Snoop Dogg. When you go to someone like Snoop for authenticity, a bot cannot deliver that authentic experience. Fans know they never spoke privately to Snoop Dogg, they only know his public persona. The bot might be fun at first, but it becomes creepy when something tries to be too human without actually being human. Zuckerberg eventually shut down the celebrity bots because they didn't get enough traction.

— [The future of AI bots: winning with personalization, conviction, and cross-platform integration](https://www.lutzfinger.com/transcript/2024-08-22-the-future-of-ai-bots-winning-with-personalization-conviction-and-cross-platform/) · _The Edge_

Question 58

Why did Ultimate AI initially focus on helping human agents instead of automating customer conversations directly?

Accepted Answer

_Agent assistance was a learning ground before full automation_

We turned the problem around. Instead of having consumers talk directly to bots, we gave recommendation tools to human agents. Agents had more tolerance for failures because they could ignore incorrect suggestions and still do their jobs, but when recommendations were correct they got value and could train the system. This was less stressful for building products. We could iterate and fail without ruining the customer experience. The technology wasn't good enough yet for full automation, so this approach let us learn gradually. Once models improved, we could then turn it back to direct user interaction.

— [How Ultimate evolved from hundreds of supervised models to UltimateGPT](https://www.lutzfinger.com/transcript/2023-05-12-how-ultimate-evolved-from-hundreds-of-supervised-models-to-ultimategpt/) · _The Edge_

Question 59

Why do AI copilots sometimes fail to deliver value in enterprises?

Accepted Answer

_Even advanced AI copilots can fail without proper user training._

Mark Benioff from Salesforce joked that Microsoft's copilots have been hit and miss, comparing them to Clippy, the failed desktop assistant. The challenge is it's not easy for enterprises to make AI useful. In my course, I was the first to completely integrate OpenAI and replace myself as a teacher, giving students AI tools to handle complicated coding without needing to code. What I saw is that students initially struggled with this new colleague. Over time, once we trained them on how to use it, they learned. So the hit and miss nature comes from the difficulty of integrating AI into existing workflows and user expectations.

— [AI in E-Commerce](https://www.lutzfinger.com/transcript/2024-10-03-ai-in-e-commerce/) · _The Keynote on AI_

Question 60

Why do AI image generators require increasingly complex prompts to get the results I want?

Accepted Answer

_Image generation gets complex fast, but workarounds exist_

When you create images with AI, you write something and get somewhat okay results, but then you iterate and the prompts become more and more complex. I've seen prompts that include things like hyper realistic, specific photographer names, lighting details, and many other technical terms all strung together. The interface is better than drawing with a mouse, especially if you can't draw, but it's not ideal yet. However, you can actually turn this around. Using tools like CLIP Interrogator on Hugging Face, you can upload a photo and retrieve the language needed to generate such an image.

— [AI as an interface](https://www.lutzfinger.com/transcript/2023-04-24-ai-as-an-interface/) · _The Edge_

Question 61

Why do we need new AI tooling platforms for quality control and orchestration?

Accepted Answer

_Managing AI requires new platforms for quality control, security, and orchestration._

Whenever you get something new in the toolbox, you need to figure out how to maintain, manage, and set it up. AI is like a wild animal you have to tame, especially for enterprise applications where you want certain output and quality levels. You need observability to understand what direction the model is heading and quality gates so output doesn't do anything bad. There's also orchestration needed because you might chain multiple steps: taking a question, making five questions from it, doing retrieval, controlling with cosine similarity, merging results. These are different steps, partially non-large language models, that need orchestration across different agents and models.

— [Cherry's Investment Thesis on AI](https://www.lutzfinger.com/transcript/2024-07-19-cherrys-investment-thesis-on-ai/) · _The Edge_

Question 62

Why does 'good enough' beat perfect for AI products?

Accepted Answer

_Shipping iteratively vs chasing perfection._

> The core problem is that AI, by design, is average. It's spectacular at pattern recognition and synthesis, but it produces the most statistically likely output — not the most surprising, not the most brave, not the most emotionally resonant. ### What Made Steve Jobs Leadership Special Let's look at Apple in the iPod era.

— [AI Cuts Creative Teams: What Companies Must Learn About AI Leadership](https://www.forbes.com/sites/lutzfinger/2026/02/13/ai-cuts-creative-teams-what-companies-must-learn-about-ai-leadership/) · _Forbes_

Question 63

Why does Catalog change the user interface depending on the type of request?

Accepted Answer

_Dynamic interfaces that adapt to what you actually need to see._

We maintain about twenty to twenty-five abstract views, not thousands of deterministic flows like traditional enterprise tools. If it's a search, we render search results. If you ask for a summary, we render a summary view. If you ask for revenue from Q2 2023, we render a graph. The LLM dynamically adapts the interface to the use case. This is one level of abstraction higher than traditional UIs, which build very bespoke workflow-level interfaces. Ninety percent of use cases fall within the first five views.

— [Large Language Models: one interface to rule them all?](https://www.lutzfinger.com/transcript/2023-08-15-large-language-models-one-interface-to-rule-them-all/) · _The Edge_

Question 64

Why does e-commerce need fine-tuned AI?

Accepted Answer

_Why general-purpose LLMs aren't enough._

> To explore this question, I looked at one industry: e-commerce. In my online certificate programs, I've explored AI use cases across many industries such as healthcare, media, coffee, finance and e-commerce. While healthcare tends to be slowest to adopt AI, e-commerce consistently leads.

— [ShopTalk 2025: Just 1% Of Vendors Use Generative AI To The Max](https://www.forbes.com/sites/lutzfinger/2025/03/21/state-of-genai--only-1---shoptalk-2025-in-numbers/) · _Forbes_

Q&A, AI Product

How are AI agents different from RPA (robotic process automation)?

How can AI support frontline workers who aren't sitting at computers all day?

How can startups use large language models to improve their products?

How did you replace yourself with a bot in your AI and product course?

How do you decide if AI is the right fit for your product problem?

How does AI change healthcare?

How does AI change the way we interact with computers compared to previous interfaces?

How does AI enable better information retrieval than traditional search?

How does AI find patterns in data that humans miss, like in healthcare?

How does Catalog ensure certain workflows always follow the same steps when using LLMs?

How does Catalog prevent large language models from hallucinating when retrieving enterprise information?

How does Catalog use large language models to simplify enterprise software access?

How does LegalOS use LLMs to make legal knowledge more accessible?

How does Neuralens measure brain health non-invasively compared to current methods?

How does Think Labs AI use physics informed AI differently than general purpose models like ChatGPT?

How does your company Gooder AI address the gap between technical metrics and business value?

How has natural language processing changed since 2015, and why does that matter for legal tech?

How has online shopping changed over the past 25 years?

How has the shift from SEO to GEO changed what websites need to do for brands?

How should product builders think about data bias in AI systems?

If agents are the future, why won't one giant model just do everything?

Should enterprises invest equally in predictive and generative AI, or focus on one?

Should founders focus on building new AI models or developing applications with existing technology?

Should you trust OpenAI blindly?

What advantage do existing platforms like Google, Meta, and TikTok have in the bot space?

What are the four approaches to addressing bias in AI models?

What are the fundamental limits of large language models that people should understand?

What are the main limitations of large language models that knowledge graphs help address?

What did you get wrong about multimodal AI interfaces?

What do you mean when you say winners won't be those who use AI, but those who redesign work around it?

What does it mean for a company to be AI-native versus just using AI?

What does the term agents actually mean in AI, and is the technology working?

What framework helps assess whether imperfect AI decisions are acceptable?

What is a data product?

What is a knowledge graph and how does it differ from a taxonomy?

What is an AI agent, really?

What is the augmentation thesis for AI in workflows?

What is the BizML framework and why is it needed for AI projects?

What is the hallucination problem with large language models and why does it matter for businesses?

What is the killer app of AI?

What is the minimum viable quality concept when building AI products?

What is the missing moat in AI?

What makes the current AI hype different from previous waves of AI development?

What patterns make a workflow 'agentic' rather than just a wrapper around an LLM?

What practical business use cases exist for language model chaining beyond autonomous agents?

What problem did transformers solve for Ultimate AI's multi-language support?

What problem does your e-commerce AI startup solve that existing technology cannot?

What problems with RAGs still need to be solved before widespread adoption?

What was Ultimate AI's key insight that made customer onboarding faster than competitors?

What were Character AI users actually using the platform for?

What's the biggest mistake CEOs make when trying to integrate AI into their companies?

Where does Ultimate AI see its competitive defensibility beyond just AI models?

Why are AI workflows the real moat?

Why are startups not yet widely adopting knowledge graphs with their LLM applications?

Why can't current AI bots provide authentic responses from real people like politicians or celebrities?

Why can't I just tell an agent to plan my honeymoon or buy things for me?

Why did celebrity bots like Meta's Snoop Dogg bot fail to gain traction?

Why did Ultimate AI initially focus on helping human agents instead of automating customer conversations directly?

Why do AI copilots sometimes fail to deliver value in enterprises?

Why do AI image generators require increasingly complex prompts to get the results I want?

Why do we need new AI tooling platforms for quality control and orchestration?

Why does 'good enough' beat perfect for AI products?

Why does Catalog change the user interface depending on the type of request?

Why does e-commerce need fine-tuned AI?

Why have AI-native startups historically struggled to succeed, and what changed with generative AI?

Why is memory the next AI battleground, not prompts or agentic systems?

You've been building AI-driven products for over 20 years at companies like Google, LinkedIn, and Snap. How do you view the current hype around agentic AI compared to past technology revolutions?