App Store

Google Play

is Device Banner Block 894x1036

Stígðu inn í heim af óteljandi sögum

Hlustaðu og lestu

Cherry Ventures

Breaking down the latest developments in AI with two experts — Jasper Masemann, investment partner at Cherry Ventures, and Lutz Finger, a visiting senior lecturer at Cornell University's SC Johnson College of Business and CEO and Co-founder of R2Decide.

Lutz & Jasper powered by Cherry Ventures

This episode breaks down what agents really are, moving past the "Matrix guy with sunglasses" imagery. Lutz and Jasper explain how agents are pieces of code that change how we interact with computers, taking defined tasks and breaking them down step by step.

A key insight emerges: agents aren't entirely new. Agents build on concepts from Robotic Process Automation (RPA), but with crucial differences. Where RPA was limited to rigid rule-based decisions, agents can handle uncertainty and unstructured data.

The discussion takes an interesting turn when Lutz brings up Microsoft's Clippy - the infamous paperclip assistant. While Clippy was widely mocked, it represents an early attempt at what we're now trying to achieve with agents. The key difference? Modern agents are more sophisticated in how they can be guided and controlled.

Cherry Ventures share have insights from their investment portfolio, including:

 • Software testing companies using "self-healing" mechanisms
 • E-commerce applications that improve search and product discovery
 • Document analysis and processing systems

AI Agents: The Future of Work?

In this episode of The Edge, we sit down with Eric Siegel, a 30-year machine learning veteran and founder of Gooder AI, to discuss the critical challenges enterprises face in deploying predictive AI models.

Episode Highlights:

The Deployment Problem

 • Introduction to the "Value Translation Gap" in enterprise AI
 • Why only 15-20% of predictive models reach production
 • The four critical predictions businesses rely on: who will click, buy, lie, or die

Why Models Fail

 • The "metrics mirage" problem in AI deployment
 • Understanding the workflow-reality gap
 • Scale challenges in moving from pilot to production
 • Implementation costs (26%) and ROI translation (18%) as key barriers

BizML Framework

 • Three essential concepts for business stakeholders:
 • What's being predicted
 • How well it predicts
 • What actions those predictions drive
 • Translating technical metrics into business outcomes

The Future of AI Products

 • Evolution from consulting to product-based solutions
 • The importance of domain-specific architectures
 • How successful companies embed business logic into ML pipelines

Investment Opportunities

 • Value Translation Tools
 • Vertical Solutions
 • Deployment Frameworks
 • The shift from model development to value realization

Featured Guest: Eric Siegel, Founder of Gooder AI and machine learning veteran

The Value Translation Gap: AI's Deployment Problem

Guest: Ferdinand Terme, Lasqo AI

Key Topics

 • Why vertical AI startups are winning against tech giants
 • Building B2B AI products vs features
 • The role of human-in-the-loop in scaling AI solutions
 • Transitioning from consulting to AI entrepreneurship

Main Insights

Path to Entrepreneurship

 • Started at McKinsey Casablanca (French-speaking Africa focus)
 • Joined Alma during hypergrowth (50 to 400 employees)
 • Explored Fintech before pivoting to AI/creative space

Building Lasqo AI

 • AI designers for marketing teams in SMBs
 • Focused on brand-consistent visual asset creation
 • Accessible through Slack integration
 • Uses custom-trained models (LoRA) for each brand

Why Focus Wins

 • Large companies struggle with maintaining quality across broad offerings
 • Startups can:
 • Build deeper rather than broader
 • Maintain tight feedback loops
 • Take calculated risks
 • Focus on specific use cases

Human-in-the-Loop Strategy

 • Initial model training with brand guidelines
 • Cost-effective for B2B recurring revenue model
 • Client feedback drives continuous improvement
 • Balance between automation and quality control

AI Products: Why ‘Good Enough’ Beats Perfect

In the past year, we've seen an influx of bots—from Meta’s celebrity avatars to Character AI’s digital personalities. The rush to dominate this space has been frantic, marked by high-profile acquisitions, public missteps, and a growing realization that not all bots are created equal. In this article, Jasper Masemann, investment partner, and Lutz Finger, venture partner at Cherry Ventures, argue that for bots to truly revolutionize our digital interactions, they must operate seamlessly across different platforms and applications, enhancing both workflow and entertainment experiences.

Meta’s celebrity bots, touted as the next big thing in digital interaction, have struggled to meet expectations. Despite significant investment and the allure of celebrity names like Snoop Dogg, the novelty wore off fast when the bot couldn’t capture the essence of the person it was modeled after. On the other side of the spectrum, Character AI tapped into a growing demand for digital companionship. Character AI found success by offering bots that not only replicate famous figures but also create new, engaging personalities. However, this success also revealed a troubling reliance on AI for emotional support, particularly among those struggling with clinical depression, raising ethical questions about AI's role in mental health.

What the bot is happening?

The challenges faced by Meta and Character AI underscore a broader issue in the AI space: the delicate balance between realism and user comfort. The closer AI comes to mimicking human interaction, the more it risks crossing into the "uncanny valley," where the experience becomes unsettling rather than engaging. This is particularly problematic for entertainment bots, where the line between fascinating and creepy can make or break user engagement.

Looking forward, the future of AI bots will likely hinge on two critical factors: personalization and conviction. Users don’t just want a tool that can perform tasks; they want a bot that understands them, anticipates their needs, and responds in a way that feels uniquely tailored to them. It might mean pulling back from hyper-realism and focusing instead on crafting experiences that are enjoyable without trying too hard to mimic human behavior. This is where bots like Meta’s Pi might have stumbled. Pi was designed to be the “friendliest AI,” but in trying too hard to be conversational and too friendly, it often failed to deliver the right answers efficiently. Users don’t just want a friendly chat—they want a bot that can make decisive, context-aware choices.

The next step for AI chatbots

The future of AI chatbots is also about breaking down silos. Jasper and Lutz believe that the bots of the future will be those that combine deep personalization and context-aware choices with the ability to operate across various platforms. Imagine a bot that can help you with your work tasks, keep you updated on your social feeds, and entertain you—all while maintaining a consistent, personalized interaction. This cross-platform capability is where the true potential of AI lies, and it’s the direction in which tech giants like Google, Meta, and Microsoft are moving.

While functional bots serve practical purposes like setting timers, entertainment bots face greater challenges in maintaining user engagement over time. For bots to succeed, Jasper and Lutz argue that they must balance functionality with entertainment while navigating issues of authenticity and user expectation. As companies race to develop the next generation of bots, the focus must shift from merely replicating human interaction to creating experiences that are genuinely useful, engaging, and, above all, authentic.

The future of AI bots: winning with personalization, conviction, and cross-platform integration

AI Gold Rush: Is the Market Overheating?

The AI industry, which has seen explosive growth and investment over recent years, is now entering a critical phase of maturation. Once dominated by a rush to build the necessary infrastructure for artificial intelligence (AI) technologies, the focus is increasingly shifting towards practical applications that deliver tangible business value. In this article, Jasper Masemann, investment partner, and Lutz Finger, venture partner at Cherry Ventures, discuss this shift and how to navigate the AI bubble. 

The tech industry is no stranger to bubbles fueled by excessive hype, such as the late 1990s internet boom and the current AI hype. Companies like Nvidia have seen massive stock price increases due to high demand for AI chips. Yet, there’s a disconnect between the market valuation and the practical applications of AI. The hype around generative AI, while significant, often outpaces the actual implementation and usability of the technology. This situation mirrors historical events like the gold rush, where heavy investments in tools were made with hopes of striking it rich, only to find that the actual availability of gold was uncertain.

The rush to build AI infrastructure has led to a saturation point where additional investments in this layer are unlikely to generate significant value. The market now recognizes that AI's true value lies in practical business applications, where it can drive efficiency, innovation, and better decision-making. As investment in AI infrastructure slows, the focus rightly shifts to integrating AI into enterprise operations for transformative impact. This is where the next wave of investment should be directed.

Navigating the AI Bubble

The venture capital world is also adjusting. There’s been heavy investment in foundational models and AI research, but without clear revenue models, follow-on funding becomes challenging. As the AI market approaches bubble territory, it is crucial for startup founders to navigate this landscape wisely. Here are some strategies to consider:

Focus on practical applications and value creation 1. Now is the time for a shift from research to practical applications. Identify areas within your business where AI can solve real problems and create value. This involves understanding the specific needs of your clients and developing tailored and user-friendly AI solutions to address those needs.
Develop a prudent business case 2. There is a debate on whether further innovation is necessary or if the industry should focus on refining and applying the current capabilities of AI, such as integrating large language models into user-friendly applications. Avoid the hype surrounding AI infrastructure and concentrate on applications with clear, demonstrable benefits. Focus on applications that present a strong business case and a clear path to profitability. Your goal should be to make AI accessible and useful to the average user. 
Integrate AI into everyday workflows 3. Embracing AI-native workflows is key to creating effective and seamless solutions. This requires a deep understanding of both AI technologies and the specific business domain. You should integrate AI into core processes to enhance efficiency and decision-making. For broader adoption, AI tools must be intuitive and fit seamlessly into existing workflows. For instance, artists and creators may be hesitant to adopt AI tools due to fears of losing control or compromising quality. Designing AI solutions that enhance rather than disrupt existing practices will be crucial for gaining user acceptance.

AI Gold Rush: Is the Market Overheating?

Welcome back to another episode of The Edge by Cherry Ventures where we discuss new, edgy topics about the future and AI. Live from London, today’s episode is a recap from the previous workshop at Station F, where they discussed building products with AI, and their most common challenges and mistakes.

Building AI products requires a different approach than traditional software development. Unlike deterministic software, AI projects necessitate experimentation to determine the relevance and effectiveness of the AI used. Listeners are cautioned against the superficial integration of AI, such as adding generative AI to products where it doesn't add value.

AI fit analysis is a tool used to evaluate whether AI is suitable for a specific product or problem. The analysis focuses on the quality and precision of data, and highlighting the inherent bias of data. Deciding the extent to which data should be de-biased is important, since sometimes biased data can be useful for achieving specific outcomes. When it comes to AI-based decision making, it is crucial to understand how and if AI can enhance the processes. For instance, in customer care, simple rule-based systems may suffice for straightforward queries, while complex, non-linear problems, like legal tech applications, benefit from AI's ability to handle numerous variables and provide more sophisticated solutions.

Next, the conversation highlights real world use cases of AI. It is particularly valuable in things like handling complex decision-making processes with many variables. Some legal tech companies are using AI to analyze contracts, compare cases, and guide users through complex decisions. AI's ability to process and analyze large amounts of data quickly can significantly enhance such applications. In cybersecurity, AI can support infrastructure decisions by recognizing patterns and guiding users through changing scenarios. Industries such as social media monitoring benefit from continuously evolving AI models

Scaling AI involves considering infrastructure costs, latency, and the overall benefits of scaling the system. The costs associated with using AI, such as query charges, must be justified by the benefits it provides, such as in healthcare applications where AI can enhance a doctor's efficiency and effectiveness. Open-source models like Lama reduce costs to hosting fees, making AI more accessible. However, achieving exponential scaling can serve as a competitive advantage.

Building AI products: 5 lessons from our founders' workshop

For the first time ever, Lutz Finger, venture partner at Cherry Ventures, hosted a live founder coaching session for three pioneering startups looking to revolutionise neurological care with artificial intelligence (AI). Taking place at Cornell Tech, a graduate school and research center based in New York City, the session brought together three pre-seed stage startups for a unique coaching experience focusing on AI, data analytics, and entrepreneurship. 

Zenith, led by Shang, is developing an AI-first operating system to optimize treatment protocols and reduce costs, particularly in expensive therapeutic settings. Vince Hartman, co-founder and CEO of Abstractive Health, discussed solving physician burnout by creating real-time summary of patients’ medical records using AI. Neuralenz, presented by its founder Oybeck, focuses on non-invasive methods to measure brain health, potentially revolutionizing neurocritical care by avoiding invasive procedures.

The founder coaching featured discussions on how advanced AI and data analytics are being integrated into new healthcare solutions, aiming to improve patient care and reduce costs. In this article, we outline three key tips for startups on their AI journey, using the experiences of these startups as actionable examples.

Founder Coaching: Tips for your AI Journey

Podcast: The Edge by Cherry Ventures Episode: Season 2 Episode 3

Welcome back to another episode of The Edge by Cherry Ventures, a new seasonal podcast rebranding ourselves as we discuss new, edgy topics about the future and AI. We’re joined today by Lutz Finger and Jasper for a live and uncut conversation.

To begin, Lutz and Jasper outline their approach to understanding AI, breaking it down into six key areas. They categorize AI's impact into two main types: evolutionary and revolutionary. Evolutionary AI improves existing workflows and everyday processes, while revolutionary AI introduces entirely new ways of doing things. Evolution begins with the concept of augmentation, where AI enhances daily tasks by simplifying repetitive workflows. Examples include using AI to automate booking processes, manage vacation requests, and streamline various administrative tasks. Though AI will replace some jobs, it will also create new opportunities and more efficient workflows. This shift necessitates adaptation and retraining for those affected. The value created by AI often benefits company owners and investors, highlighting the need for a fair distribution of these benefits.

AI (specifically large language models) has the ability to make everyone smarter by improving information retrieval. While AI enhances information access, human oversight remains essential to ensure the accuracy and relevance of the retrieved data. The conversation then turns to potential business models that could emerge from AI evolution. One significant area is the replacement of traditional search engines like Google. Professions like auditing, tax advising, and engineering, where professionals occasionally need to reference large volumes of documents, can benefit greatly from AI's ability to quickly and accurately retrieve relevant information. Lutz and Jasper also highlight the complexities of ensuring data security and quality, particularly for large enterprises.

Then, the conversation shifts to discuss the revolutionary potential of AI. First, future AI bots in providing personalized assistance and advice while acting as real sales consultants. AI could also streamline clunky, complicated interfaces by showing only the buttons and features relevant to the user’s current task, based on historical usage patterns and user personas. Ai-generated content is the next step for social media, too. Finally, new processing power could open up an entirely new suite of applications to improve decision making.

Links: Learn more about Cherry Ventures: www.cherry.vc

Cherry's Investment Thesis on AI

Artificial Intelligence (AI) has been a buzzword for decades, yet we find ourselves in a new wave of AI hype, one that holds a genuine promise to reshape how we live and work. What sets this era apart? While it might seem like nothing much has changed, there’s both a technical evolution and a shift in accessibility that are driving the current shift.

A reality check on AI

We are back! In this episode, we revisit our 2023 predictions. What happened? What did not happen? Dive into topics like the state of MLOps or the hardware evolution.

Tune in and we are very happy to welcome you back for the 2024 season of Lutz & Jasper!

Did our predictions hold true?

LLMs + vector search = RAG.In this episode, we introduce you to Retrieval-Augmented Generation (RAG) and how it might allow us to better and more easily implement LLMs in our day-to-day work. We believe current LLMs are best used as an interface, and if we want to make enterprise search useful, RAGs can be the solution.

Are we one step closer to using LLMs in practice?

Hlustaðu og lestu

Other podcasts you might like ...