“Software is eating the world, but AI is going to eat software.”
– Jensen Huang, CEO and Founder of Nvidia
By Dmitry Alimov, Fabian Flatz, Frontier Ventures
Computing history is full of once-prominent companies that have disappeared because of shifts in technology. Names like Atari, Blockbuster, and Kodak serve as reminders that no company, no matter how successful, is immune to the impact of new technological developments. Now, a new wave of innovation is sweeping the world in the form of next-generation Artificial Intelligence (AI). Tech startups of all stages must now adapt to these changes, or risk facing extinction. In this post, we’ll explore the ways in which AI is transforming the startup landscape, and what founders can do to stay ahead of the curve. This post is primarily for non-technical founders, please reach out if you would like to have a more technical discussion with our team or technical implementation advice.
Only in Q1 2023, generative AI startups raised over $1.6 billion, a 100% increase year-over-year, defying the declining venture investments trend (Pitchbook). Big Tech, including Meta, Alphabet, Microsoft, and Apple, are making significant AI investments. Non-tech companies, too, are following suit, with 64-80% planning AI budget increases over the next three years (B2BaCEO). Someone will build large businesses to service this surging demand.
Frontier Ventures has been investing in AI companies since 2019, our portfolio companies PredictLeads, Gravity AI and Turing are examples. This year, we spent time with hundreds of AI companies and experts at the cutting edge of the field to understand the rapidly evolving AI landscape. Our most important conclusion is that the current revolution in AI is a make-or-break event for all tech startups. Tech founders are now facing a stark existential choice: leverage AI to dramatically accelerate their business or allow someone else to use AI to disrupt their business.
AI will also affect fundraising: investors will understandably avoid companies that do not leverage AI and thus expose themselves to AI driven challengers. Normally, founders should not choose a strategy based on what investors want but AI is a massive paradigm shift and founders neglecting it both send out a negative signal to investors and are indeed more likely to fail.
How did we get here?
AI research, beginning in the 1950s, transitioned from symbolic logic and rule-based systems to knowledge-based systems and neural networks in the 80s and 90s. Deep learning in the 2000s spurred advancements in computer vision, speech recognition, and natural language processing. The 2017 “Attention is All You Need” paper by Google introduced the transformative transformer architecture, paving the way for Large Language Models (LLMs) such as GPT 4. Increased computing power allowed training on massive datasets, enabling LLMs to predict subsequent words, produce contextually appropriate responses, and answer complex questions. In 2023, the pace of progress in AI has accelerated dramatically, including the release of GPT 4, high quality open source LLMs and related technologies. As the recently leaked Google memo illustrates – open source models have been rapidly catching up to big tech’s “closed source” models.
First Goal for Founders: Understand Capabilities of AI / LLMs
In practical terms, the number one goal for tech founders should be to understand what modern AI tech is already capable of, so that you can explore potential use cases for AI in your business, especially capabilities of LLMs such as OpenAI’s GPT 4. A list of resources for understanding AI capabilities in detail is provided in the Appendix but on a high level generative AI / LLMs capabilities can be divided into four areas: text, image, audio and video. In all four areas, use cases exist, in which high-quality LLMs are already outperforming humans. Here’s an incomplete list of current LLM capabilities (rapidly growing):
- Text: LLMs can sift through documents that are hundreds of pages long within seconds, summarize the main points for you, explain the tone of the document, suggest improvements etc. They can also classify, transform and translate text. Additionally, LLMs can also create new documents and texts from scratch, simply based on a sentence or a few bullet points you provided. These capabilities have many potential applications in processing information, communications, customer service, marketing etc.
- Images: if you describe an image, generative AI is able to create said image to a very high degree of quality, in some cases virtually indistinguishable from a real photograph or painting. A fun example of this is where AI has been used to expand old, masterpiece paintings, such as Starry Night by Van Gogh, and the results look stunning. Here’s Adobe’s take on generative AI.
- Audio: AI can generate music and voices. For example, a company could clone the voice of its CEO, with his/her permission, in order to create a customized audio greeting for every new user that registers for the product. AI can also capture conversations as text transcripts.
- Video: while creating photorealistic videos from scratch is not (yet) possible, generative AI can already modify existing videos. For example, imagine you produce an expensive video ad only to discover in post-production that an intrusive pedestrian encroached into your frame, causing a jarring visual distraction. Thanks to AI you can simply remove the pedestrian from the video instead of having to reshoot the whole scene.
To decide where in the tech stack you want to build AI driven products (or improve existing products), you can use this helpful AI tech stack framework from NFX as a reference. In general, you should focus on your part of the stack and avoid building non core “horizontal” functionality already built by existing players. We believe that in most tech businesses AI can add critical value, happy to brainstorm with you how it might apply to your company.
However, even if AI is not at the core of your product, most tech startups can achieve dramatic improvements in quality and/or cost of their operations by using AI tools already available. We include a list of helpful AI resources and tools for founders below; it is not intended to be comprehensive, only to be used as a starting point for further research.
AI Strategy Advice
Our portfolio company Turing is an AI-powered deep-vetting talent platform that matches companies with engineering talent worldwide. By leveraging AI, Turing streamlines the recruitment process and helps businesses find the best candidates quickly. Recently, Turing announced several new AI-first features, such as an AI Advisory Service, consisting of an elite team of AI technology advisors from Meta, Google, Apple etc., and a first GPT Powered Software Development workflow with the goal of significantly increasing developer productivity. Happy to introduce you to Turing if you are looking to get AI advice or to hire software developers, data scientists or other tech professionals.
Low code no code tools
While deploying AI will often require sophisticated tools and infrastructure, there are also many low code no code tools available (and many more will be launched). As Andrej Karpathy, former Director of AI at Tesla and former Researcher at Open AI said, “The hottest new programming language is English.” This means that many non-technical employees in your company may be able to use AI tools without writing a line of code. Existing low code no code players such as Airtable, Zapier and Make have launched various AI / LLM integrations you can use without or with minimal code.
Tools to explore are Webflow, helping users to build beautiful websites, Glide, enabling users to build workflows on top of spreadsheet data, Bravo, which can turn Adobe and Figma designs into apps, and Bubble, a comprehensive tool for building fully-functioning apps without code. We also encourage you to explore specialized low code no code building tools for AI products, such as for example Stack AI, which streamlines the AI development and deployment process.
Some skills in writing LLM prompts are essential but learning them does not require a computer science background. We added a helpful Deeplearning.ai course for writing Open AI GPT prompts at the end of this post.
Intently is a tool for sales teams looking to automate lead identification and engage potential customers with personalized outreach that demonstrates an understanding of their needs. Using AI, Intently helps sales teams identify the best targets and delivers tailored outreach to increase conversions. Regi.AI helps Sales Development Reps write hyper-personalized cold emails 10x faster. Pilot automatically turns sales calls into detailed notes and structured data that syncs directly to your CRM.
Our portfolio company PredictLeads offers comprehensive AI collected data on tens of millions of companies globally which is used to generate high conversion b2b sales leads. Jasper and Copy.AI are AI content creation tools to create product descriptions, marketing copy, and social media posts with ease. Anyword analyzes the content you published previously on your website, ads, social, and email channels, and generates effective copy that follows a consistent tone of voice based on the data it analyzed.
Finchat is a chat-based AI trained specifically to answer finance related questions. Trullion is an AI-powered platform that automates manual work for accounting and finance teams, such as Revenue Recognition. This tool is particularly relevant for mature startups at the Series B+ level. You can use Clockwork to quickly generate forecasts for your financial model.
Beautiful.ai and MagicSlides leverage AI to create visually appealing presentations. Midjourney and Stable Diffusion can generate stunning images based on text descriptions. Runway allows you to remove objects from an image. Mubert is a platform that allows users to generate royalty-free music for apps and videos. UiZard can build UI/UX wireframes and functional prototypes based on text prompts (available in early access). Genius is an AI copilot for Figma, helping designers create designs faster.
GPT 4 has been trained on a large body of code. It can competently generate code, explain existing code, and help debug code. Try GPT 4 with browsing to access API documentation. Google’s Bard is another “Coder Co-Pilot” AI tool that can be useful for engineers looking to improve their workflows. Bard is connected to recent knowledge and can provide footnotes, rewrite code into a different language, and explain blocks of code with ease. GitHub Copilot is another popular Co-Pilot tool that uses OpenAI Codex to suggest code directly from a developer’s editor. Phind is a search engine specifically designed to help developers. Replit is an online integrated development environment (IDE) that allows the user to collaborate in real-time and offers Ghostwriter, an AI-powered pair programmer. Mintlify allows developers to auto generate documentation based on their code using AI. Codacy provides visibility into the quality of your developers code
Our portfolio company Turing is an AI-powered deep-vetting talent platform that matches companies with engineering talent worldwide. By leveraging AI, Turing streamlines the recruitment process and helps businesses find the best candidates quickly. Happy to introduce you to Turing if you are looking to get AI advice or to hire software developers, data scientists or other tech professionals.
OnLoop helps companies with a hybrid workforce to gain visibility into the performance of their team members and can automatically generate performance reports. Eightfold provides talent management and recruitment solutions using an AI-powered platform that analyzes large volumes of data to help companies find the right candidates for their job openings and upskill their employees.
Casetext offers CoCounsel, an AI legal assistant which can assist in document review, legal research memos, deposition preparation, and contract analysis. Hadrius is another useful tool that automates several key SEC compliance processes for regulated firms, particularly relevant for fintech companies.
Mem is an AI-powered notetaking and knowledge management tool that automatically captures and organizes information from various sources to help individuals and teams collaborate and share knowledge efficiently. AutoGPT, which started as an open-source application available on GitHub, creates an autonomously acting agent to whom you can assign a goal. The agent then sets out to perform a series of tasks in order to achieve that goal without human intervention.
Many different variants of autonomous agents have been launched. A particularly interesting one is the Langchain implementation. It is designed specifically for autonomous agents and natural language generation use cases, and it provides pre-built models for various applications, such as chatbots and email automation. This is different from the more general-purpose use cases of Auto-GPT and, while more limited in flexibility, it comes with a shorter time-to-market. Note that the current interfaces of autonomous agents are often not user-friendly and not meant for non-software developers. However, versions with an improved user experience are expected in the near future.
Zoe is a natural language chatbot that provides complex data analysis for Business Intelligence. ProbeAI is the copilot for data analytics, making it far easier and more convenient to perform data analysis. Fireflies.ai is an AI meeting assistant that records meeting from Zoom / Google Meet / MS Teams, and generates transcripts and summaries
To find additional AI service providers, check out this helpful Generative AI Market Map with over 550 AI vendors open sourced by our friends at NFX.
Nobody knows what the full long-term implications of Artificial Intelligence will be. Predictions range from a utopian future of prosperity to a dystopian future led by a destructive AI. However, what is clear is that AI is no longer an experimental technology, but a very powerful transformative force that is reshaping the technology and other industries today. This year, the pace of change accelerated dramatically and the moment of truth has now arrived for all tech founders. Those who embrace and leverage AI will thrive, while those who do not are almost certain to perish. Inaction is not an option. It is the responsibility of founders as leaders to first understand this transformative technology and its many applications and then take action and embrace the transformative potential of AI without delay. We are here to help. Let us know if you would like to set up a call with the Frontier Ventures team for further discussion.
Additional AI Resources
Follow Frontier Ventures for new AI content
- ChatGPT Prompt Engineering for Developers – Short course on Deeplearning.AI (link). While this course is intended for developers, non-technical users will find it very useful – just ignore python code instructions and focus on understanding the prompt engineering – with this advice any non-technical person can improve their results significantly without needing to write a single line of code
- Coursera Class in Deep Learning taught by Stanford Prof. Andrew Ng (link).
- A class specifically on Marketing in Artificial Intelligence by University of Virginia Darden School of Business Prof. Rajkumar Venkatesan (link)
- The Complete Beginners Guide to Autonomous Agents – Youtube video series (link)
- LLM Bootcamp – Spring 2023 – Course for beginners and intermediate learners who want to learn how to build and deploy applications powered by large language models (link)
AI Capabilities and Business Use Cases
- McKinsey’s Overview of ML Models And Business Use Cases
- Ibm’s Guide To Implementing AI Applications
- Dataiku’s Enterprise Guide To LLMs Uses Cases
- The Future Of Generative AI Is Niche, Not Generalized – this MIT Technology Review article makes a strong case for making use of the emerging ecosystem of tools beyond OpenAI, such as open source models, to power your AI application
- Overview Of AI Tools From The Rundown And A Generative AI Market Map From NFX – these maps can serve as a source for tech stack and vendor selection but you can also use them to understand current AI use cases
- First Economic Assessment of LLMs Capabilities – TLDR: When incorporating tools built on top of LLMs an est. 47-56% of all worker tasks can be performed significantly faster at the same level of quality
- The AI Startup Litmus Test – must read strategy guide on finding a defensible use case within the AI landscape
- The Large Language Model Landscape – helpful overview of the current Large Language Model ecosystem
Blogs and GitHub repos
- Ben’s Bites — Daily, easy to understand newsletter about recent developments in AI
- Dair.AI — Blog explaining academic AI papers to a non-academic audience
- TheAivalley.com – Weekly AI newsletter
- B2BaCEO – Newsletter for technical founders by Foundation Capital’s Ashu Garg
- Practical Guides for Large Language Models – GitHub repo with curated list
- The Comprehensive Artificial Intelligence Index 2023 Report from Stanford
- Scale’s 2023 AI Readiness Report
- Databrick’s 2023 report on the State of Data + AI
- Semianalys – Blog sitting at the intersection of semiconductors and business
- HuggingFace’s blog
- The Sequence of AI Knowledge – A newsletter curated by industry insiders
- Machine Learning Mastery – A hub of practical AI resources aimed at developers
- KDnuggets – Website that provides news, and tutorials related to data science and AI
- CS 324 – Excellent AI resources from Stanford
- Building LLM applications for production – Practical essay from Chip Huyen
- Knowledge Retrieval Architecture for LLMs
- Github repo with a curated selection of AI courses and resources
- LangChain, Haystack, and LlamaIndex (GPT Index) are open source tools that help you build model chains, pipelines, connect your data to the models and do many LLM related tasks. This article is a good starting point
- OpenAI Blog
- GoogleAI Blog
Interesting AI related Twitter accounts to follow
|@ashugarg||Ashu Garg, Partner @ Foundation Capital|
|@geoffreyhinton||Geoffrey Hinton, the “Godfather of deep learning”|
|@drfeifei||CS Professor @ Stanford and co-founder of of the non-profit AI4AL|
|@Thom_Wolf||Thomas Wolf, co-founder of HuggingFace|
|@AndrewYNg||Prof. @ Stanford, founder of Google Brain|
|@demishassabis||Demis Hassabis, founder of DeepMind|
|@hwchase17||Harrison Chase, founder of LangChain|
|@lilianweng||Lilian Weng, leads applied AI @ OpenAI|
|@karpathy||Andrej Karpathy, Former Director of AI @ Tesla|
|@ylecun||Yann LeCun, Prof. @ NYU and Chief AI Scientist @ Meta|
|@heyBarsee||editor @ TheAiValley, a weekly AI newsletter|
|@osanseviero||Omar Sanseviero, Machine Learning Engineer Lead @huggingface|
|@gdibner||Gil Dibner, investor @ Angular Ventures|
|@mustafasuleymn||Mustafa Suleyman, Co-founder of InflectionAI and Partner @ Greylock|