Codeium’s Vision for AI-Driven Development with Varun Mohan | E2064
23 Dec 2024 (1 month ago)
Jason and Alex kick off the show (0s)
- Venture capitalists (VCs) view entrepreneurs as crucial to achieving their goal of having a 3-5x fund, requiring a driven individual to lead a mission, which may involve recruiting other like-minded individuals (7s).
- The pursuit of this goal is likened to becoming an "Elite Jedi," an elite pursuit that is not suitable for everyone (15s).
- Not everyone can become a Jedi, as it requires unique abilities and strengths, and even the most powerful Jedi can experience setbacks and losses (28s).
- The analogy of fighting a Sith Lord is used to illustrate the challenges and risks involved in this pursuit, where even the best can get hurt (25s).
Failure examples and Star Wars character confusion (30s)
- Anakin Skywalker has a long list of lost limbs in the Star Wars universe, indicating a brutal pursuit even for lead characters like him (30s).
- There is confusion between Anakin and Luke, with Anakin being the character who eventually becomes Darth Vader (42s).
- The speaker hasn't seen the Jar Jar Binks movie since high school (48s).
- The segment is sponsored by Squarespace, which allows users to turn their ideas into new websites, with a free trial available at squarespace.com and a 10% discount using the offer code "twist" (53s).
- LinkedIn is also offering a $100 ad credit for launching a first campaign at linkedin.com (1m11s).
- PrizePicks is mentioned as the best place to get real money action while watching favorite sports, with a $50 instant reward after playing the first $5 lineup using the code "twist" (1m21s).
Introduction to the episode, guests, and Perplexity AI (1m38s)
- The episode features an interview with Varun Mohan, the CEO of Codeium, a startup building something interesting, and a demo will be shown later in the episode, followed by a live Q&A session with Jason (1m39s).
- Perplexity, an AI search startup, has been in the news due to its rapid growth, conflicts with major publishers, and impressive fundraising (2m2s).
- In 2024, Perplexity announced a $74 million Series B in January, which raised its valuation to $100 million, and later added $63 million to its Series B in April, reaching a $1 billion valuation (2m22s).
- The company continued to raise funds, securing $250 million at a $3 billion valuation in the summer, and most recently, $500 million at a $9 billion valuation, as reported by Bloomberg (2m46s).
- Perplexity's query load has significantly increased, from 500 million queries in 2023 to 250 million in July alone, and recently scaled to 100 million per week or about 400 million per month (3m5s).
- Although Perplexity is still much smaller than Google, which handles about 5.9 million searches per minute, the value of search is incredibly high, and controlling even a small percentage of the search market can generate enormous wealth (3m26s).
- With its AI-first approach, subscription business, and ad business, Perplexity's valuation may be expensive at $9 billion, but if it can capture a significant percentage of the search market, it could be considered a good investment (4m17s).
Varun Mohan and Codium's journey in AI and IDE development (4m36s)
- Varun Mohan is the founder and CEO of Codeium, a company that has developed an AI coding tool, and has raised around a quarter billion dollars from investors such as Kleiner Perkins, General Catalyst, Green Oaks, and Founders Fund (5m35s).
- Codeium initially started by building extensions to bring AI intelligence into various Integrated Development Environments (IDEs), but has now developed its own IDE (5m58s).
- Varun Mohan's entrepreneurial journey began after graduating from MIT, where he worked at Nuro, an autonomous vehicle company, before starting Codeium (6m42s).
- Codeium initially started as a GPU infrastructure company, making it easier to run deep learning applications, and managed around 10,000 GPUs, generating a couple million dollars in revenue (7m4s).
- The company's target customer base was initially autonomous vehicles and robotics companies, but the rise of Transformers and GPT models in mid-2022 led to a pivot in the company's direction (7m20s).
- Varun Mohan considers Codeium his first startup, but due to the pivot, he also considers himself a second-time founder (6m22s).
- The company's goal is to build a platform that can usher in new applications, similar to how the internet did, by leveraging the power of AI and machine learning (7m31s).
- Codeium was started after pivoting from a previous company called EXA Function, which was initially focused on infrastructure expertise, but the founders decided to vertically integrate and create a product for AI code assistance (7m45s).
- The decision to pivot was made despite having raised $28 million in Series A funding from Green Oak Capital and having a couple of million in revenue, as the founders did not see a clear path to reaching a billion-dollar revenue business (8m17s).
- The pivot was made over a weekend, and the founders believe that it's essential for startups to ask themselves if they're taking the right test, and if not, to be willing to move on to something else (8m43s).
- The founders acknowledge that there can be shame and attachment to the original idea, but they believe that it's essential to define reality and reframe the situation, rather than getting stuck in the sunken cost fallacy (9m15s).
- The founders frame the pivot as a decision to "Ace the test" but realizing that it's not the right test to take, and instead, finding a new opportunity that has the potential to lead to a billion-dollar revenue business (9m59s).
- The founders believe that great leaders define reality and framing is everything, and that it's essential to be honest with investors and stakeholders about the decision to pivot (9m50s).
- Investors, such as Green Oak Capital, bet on teams and founding teams, and they appreciate when a team comes to them and says that they've found a new opportunity that has the potential to lead to a billion-dollar revenue business (10m41s).
Squarespace - Use offer code TWIST to save 10% off your first purchase of a website or domain (11m1s)
- Squarespace is an all-in-one platform that allows users to build a beautiful website and grow their business, whether they're just starting out or scaling up (11m2s).
- Squarespace's design intelligence feature combines decades of design expertise with cutting-edge AI, making it easy to craft a stunning personalized website in just minutes (11m20s).
- The platform offers seamless payment management through Squarespace Payments, with fast onboarding and support for popular methods like Apple Pay (11m30s).
- Blueprint AI is a feature that automatically builds a polished, on-brand site with curated content tailored to the user's style, after answering a few quick questions (11m45s).
- Users can check out Squarespace for a free trial at squarespace.com/twist, and get 10% off their first website or domain purchase by going to squarespace.com/twist (11m57s).
- The offer code "TWIST" can be used to save 10% off the first purchase of a website or domain (12m3s).
The challenge of moderate success and building 10x better experiences (12m29s)
- The concept of success and failure is discussed, with the idea that it's not failure that people fear, but rather the inability to move on from it, and having moderate success can be beneficial as it allows for learning and growth (12m29s).
- The importance of keeping a team lean is emphasized, as having too many people can be unnecessary and inefficient, and it's better to have a smaller team that can adapt and change quickly (13m14s).
- The ability to make strategic changes and adapt to new situations is seen as a major advantage for startups, as it allows them to stay competitive and innovate (13m37s).
- The idea of changing management philosophies while keeping the company's goals and mission the same is discussed, with the idea that this can be beneficial for growth and innovation (14m19s).
- After the pivot from X, the company built a number of extensions that worked with other IDEs to help people code, and eventually developed an AI code assistant platform with trained models that could offer autocomplete and chat capabilities for free (14m50s).
- The Codium extensions were offered entirely for free across all IDEs, and were even enabled to run entirely within the firewall of large enterprises, with personalized suggestions based on private internal data (15m11s).
- Codeium has over 800,000 users and more than 1,000 enterprises, including companies like JP Morgan and Chase, using its extensions, with thousands of developers internally, and was inducted into their Hall of Innovation two months ago (15m37s).
- The company decided to create its own IDE, called Wier, after realizing it was limited by existing IDE applications, and this decision was made two months ago (16m12s).
- Building its own model allowed Codeium to offer its extension for free, as it had a lower cost per query, which was made possible by its background in GPU infrastructure and ability to build its own compilers and run Transformers efficiently on its own servers (16m39s).
- The company's ability to build a massive free individual product enabled it to AB test features, which is critical for working with enterprises, as they don't want to be guinea pigs for new products (17m7s).
- The new IDE, Wier, was built on top of the company's existing extensions and provides a high-quality UI workflow, as well as an agentic experience called Cascade, where AI can write more and more of the software (17m51s).
- Codeium's extensions currently generate around 50% of all committed software, but with Wier, this number is expected to increase to over 80 or 90% (18m5s).
- The decision to create Wier was driven by the company's desire to take its product to the next level and provide something that users may not even know they want yet, rather than just incremental changes (18m22s).
- This approach is necessary for startups to succeed, as they need to differentiate themselves from incumbents who have more distribution, capital, and resources (18m31s).
- The goal is to offer users 10x better experiences every year, which requires cannibalizing the existing state of the product annually, setting a high expectation internally (18m42s).
- There is a free version of the product, and it's fascinating to see how it works, especially with the use of agents, which are similar to co-pilots (18m57s).
- Agents work by predicting what the user will do next, similar to how email or Grammarly suggest words or phrases to writers, as English and code are just different languages with predictable patterns (19m2s).
- This predictability allows agents to suggest code snippets, which can be a few lines or even entire paragraphs, or even code found on GitHub (19m25s).
- The product has been built to showcase its capabilities, and the agents are different from co-pilots, which have been discussed for years (19m40s).
- The product's vision is to revolutionize AI-driven development, and it's exciting to see what has been built (19m47s).
LinkedIn Ads - Get a $100 LinkedIn ad credit (19m48s)
- To effectively reach target customers, LinkedIn ads can be used to precisely target professionals most likely to care about a product, allowing users to reach people by job title, industry, company, location, and more (19m56s).
- LinkedIn ads can help build a B2B audience, drive results, and gain access to over a billion members, including 130 million decision-makers who can make purchasing decisions (20m28s).
- The LinkedIn audience also includes 10 million elite C-level executives, such as Chief Technology Officers, Chief Strategy Officers, Chief Revenue Officers, Chief Executive Officers, and Chief Financial Officers (20m43s).
- Using LinkedIn ads can result in two to five times the return on ad spend, with 79% of B2B marketers saying LinkedIn delivers the best paid media results (20m57s).
- A call to action is made to start converting B2B audiences into high-quality leads today, and a $100 credit is offered to start the next campaign by visiting linkedin.com/thisweekinstartups (21m8s).
- To claim the $100 credit, terms and conditions apply, and users should visit linkedin.com/thisweekinstartups (21m14s).
Alex Wilhelm's Cascade demo and Codium's capabilities (21m21s)
- A demo of Cascade was performed, showcasing its capabilities in building an app from scratch, with the user having no prior knowledge or experience, and the system guiding them through the process (21m23s).
- The user was asked what kind of app they wanted to build, and they chose a simple messaging service for themselves and a few friends, with Cascade quickly coming up with ideas and requiring the user's approval for each step (21m47s).
- The system handled tasks such as installing Python, running the app, and troubleshooting issues, including a denied access to Local Host, which it was able to resolve on its own (22m15s).
- The user interacted with the app, registering, logging in, and running into errors, which Cascade helped to resolve, including restarting the server (22m43s).
- The user encountered an error message stating that the error rate limit had been exceeded and was prompted to upgrade to a Pro account or try again in an hour (23m1s).
- The cost of the interaction was discussed, with the company using a combination of its own models and Anthropic models, and the cost being estimated to be in the low tens of cents, around 25-50 cents (24m42s).
- The company's product uses a combination of its own models and Anthropic models, with the latter being used for local edits, and the company has been taking down Anthropic models multiple times due to capacity issues (23m20s).
- The company has raised a lot of money but is running out of capacity, and it was suggested that they should put up a roadblock to limit the number of users and control the acquisition cost of customers (23m55s).
Codium's pricing and the future of no code development (24m49s)
- Codeium offers a free version and two paid tiers: Pro at $15 a month and Ultimate at $60 a month, with an average revenue of $25 per user per month, aiming to break even by converting a small percentage of free users to paid ones (24m55s).
- The concept of giving away something for free and then converting users to paid customers is compared to offering a sample of a milkshake and then selling the full product, with the goal of making the milkshake profitable (25m26s).
- A demonstration of Codeium's capabilities is shown, where a user can log in and build a simple Twitter-like application using natural language and without writing code (25m50s).
- The user interface is intuitive, and the system can understand and execute commands given in plain English, making it accessible to non-technical users (26m1s).
- The creator of the demonstration has no formal training in development and learned through self-study and experimentation, highlighting the potential of Codeium to democratize coding (26m27s).
- The vision for Codeium is to enable users to build applications using natural language and without writing code, making it possible for people without extensive coding experience to create functional apps (27m1s).
- The company has over 100 employees, and everyone was able to build an application using Codeium, with some team members continuing to build apps for personal and professional use (27m39s).
- The use of Codeium has led to the creation of useful tools for the company, such as a quoting tool, and has enabled non-technical team members to contribute to the development process (27m59s).
- Codeium's vision involves using AI-driven development tools to revolutionize the coding process, potentially making it possible to build a better version of a quoting tool for a fraction of the cost of traditional methods (28m4s).
- The concept of "agents" is introduced, where a large language model (LLM) can work on itself, using its output to inform the next step in the process, mimicking human reasoning (28m49s).
- Agents can be used in various ways, including deep retrieval, where the AI searches through an entire code base to find the best location to make changes, and executing code, which can automate tasks such as installing requirements (29m24s).
- For experienced developers, agents can be helpful in modifying large code bases, while for new developers, agents can assist in building apps from scratch (29m40s).
- The use of agents can raise the bar for everyone, enabling non-developers to build apps and large companies to modify them quickly (30m34s).
- A feature request is made for a $10-20 per month option to spin up a server and use a custom URL, allowing users to take their projects off their local machines (30m46s).
- Codeium aims to make its product accessible to every developer in the world, not just those in large enterprises, because it is a developer product that requires widespread use within a company (31m32s).
- The company acknowledges that its product can be useful for individual developers and small-scale projects, such as personal web apps, and is open to feedback on how to cater to these users (31m37s).
- Codeium recognizes that its product is different from others, like databases, which are typically only used by a handful of people within a company, whereas Codeium's product is designed for use by everyone in the company (31m29s).
- The idea of making the product simple and affordable for individual developers, with minimal hosting costs and high gross margins, is seen as a valuable piece of feedback that the company is willing to consider (31m11s).
PrizePicks - Run your game. Download the app today and use code TWIST to get $50 instantly after you play your first $5 lineup! (31m45s)
- PrizePicks is a platform that allows users to get real money action while watching their favorite game by picking more or less on at least two players' stats, with the option to win up to a thousand times their cash (31m46s).
- The platform offers various sports and allows users to make selections, lock them in, and win based on their picks, with the option to cash out even if one of their picks doesn't hit through Flex Pay (32m10s).
- Flex Pay allows users to still cash out and get a smaller percentage of their money back if one of their picks doesn't hit, providing a safety net (32m22s).
- PrizePicks offers MasterCard for quick and easy deposits into users' accounts, and also supports PayPal for easy transactions (32m44s).
- The platform is giving away two free picks in December and $30 plus million in rewards, and new users can get $50 instantly after playing their first $5 lineup by using the code "TWIST" (32m51s).
- The $50 bonus is guaranteed and does not require users to win, making it a promotional offer for new users to try out the platform (33m0s).
Satya Nadella on SaaS's future and AI replacing agents (33m31s)
- Business applications will likely collapse into the agent era, as they are essentially databases with business logic that will be handled by multi-repo CRUD agents, which will update multiple databases and have logic in the AI tier (33m31s).
- Once the AI tier becomes the place where all the logic is, people will start replacing the back ends, and this is already being seen with high rates of wins on Dynamics backends and agent use (34m23s).
- The central premise is that the SaaS layer is going away and will be replaced by agents hitting a database in the back end, making it easy for companies to build simplistic applications themselves (34m38s).
- Many simplistic applications that are basically database access and some logic on top of it will be extremely easy for companies to build themselves and have an agent work on it (34m48s).
- In the past, companies didn't build their own applications because they couldn't maintain them easily, but now with agents and tools, maintenance costs will be lower, and companies will be able to improve their applications over time (35m3s).
- A lot of R&D will go into improving use cases that are not specific to a particular company, and companies will only need to maintain the features they need, reducing maintenance costs (35m14s).
- Simplistic apps will get replaced, but more complex apps like Salesforce will take longer to replace, as they are complex workflow software that fundamentally affects the way businesses run (35m39s).
- Satya Nadella's statement that many business applications are just database access and some logic on top of it is correct, and this will make it easy for companies to build their own applications (34m52s).
AI's impact on developer jobs, salaries, and remote work (36m0s)
- The future of being a developer is changing with the introduction of AI-driven tools, which may make it possible for a larger percentage of people to become developers, potentially increasing the supply of developers and affecting salaries (36m2s).
- The idea that developing isn't limited to the top 2% of humans who have a propensity towards it, but can be opened up to 10-20% of people who understand basic logic and math, and can use these tools to become "good enough" (36m18s).
- The salary of a developer has been artificially high due to demand and supply, but with the increase in supply, salaries may decrease, and developers may face competition from people in other countries (36m47s).
- The increase in supply of developers may lead to a deflationary effect on salaries, with developers competing against people in countries like India, Ukraine, and Brazil (36m52s).
- However, it's also possible that as software development becomes easier, more people will become developers, and salaries will increase, as has been the case in the past when programming languages became easier to use (37m42s).
- The ease of software development may lead to companies investing more in technology, and getting more return on investment, which could lead to an increase in demand for developers (38m18s).
- Knowing more and having more "taste" in design decisions and trade-offs may make developers more valuable, even if the number of developers increases (38m48s).
- There may be a class of people who have less "taste" and are less valuable as developers, but those who can make better high-level trade-offs and understand the business and technology will be more valuable (38m52s).
- AI-driven tools may not be able to replace the high-level trade-offs and design decisions that human developers can make, at least not yet (39m7s).
- There is a belief that there are 10x people compared to others in the development space, and those who get deep in the weeds will always have great opportunities, regardless of salary distribution (39m32s).
- The idea that developer salaries will be equally distributed is unlikely, and instead, there will be a long tail of salaries with the elite developers earning more (39m29s).
- The shift to remote work has changed the dynamics of the industry, and companies like Matt Mullen's have benefited from having a global workforce and paying lower salaries (40m1s).
- The ability to hire developers from all over the world has led to a decrease in salaries for some developers, and this trend is expected to continue (40m22s).
- The elite developers will always be in high demand and will be able to extract more value from the market, while average developers may see their salaries stay the same or decrease from the peak (40m43s).
- The market will likely "dissolve" crummy developers, and only the top developers will be able to thrive and earn high salaries (41m4s).
- The use of AI-driven development tools will further exacerbate the difference between top developers and average developers, making it even more challenging for average developers to compete (41m17s).
Elite developers, hiring strategies, and in-person vs. remote work (41m27s)
- The current state of the job market and the tech industry is compared to Hollywood, where top talent like Robert Downey Jr. can still command high salaries, but average individuals may be seen as commodities, with many doing TV work instead of movies (41m38s).
- The approach to hiring in the tech industry has changed, with companies now being more selective and focusing on professional development to make existing employees 10% better every month, rather than rapidly expanding team sizes (42m26s).
- The expectations of what can be accomplished in a given day have increased, and companies are building tools to fundamentally change the speed at which technology can be built (42m30s).
- The company has a strict hiring policy, only bringing in new people when existing employees are overwhelmed, and prioritizes hard work, intensity, and long hours to build something great (42m51s).
- The company is in-person only, with no remote work, as it is believed that building something material to society requires face-to-face interaction and collaboration (43m8s).
- Young people entering the workforce are excited to work in an in-person environment, surrounded by high-caliber people, and are eager to grow quickly and learn from others (43m59s).
- The new generation of young people, who had to go through college during the COVID-19 pandemic, are particularly hungry for in-person interaction and collaboration (44m21s).
- The COVID-19 pandemic took away significant life experiences from the "COVID generation," including high school graduation and college experiences, which may have resulted in young people craving social interactions and in-person experiences (44m35s).
- Despite the shift to remote work, people still value in-person interactions and social gatherings, as evident in the continued popularity of live concerts, which have not been significantly impacted by virtual reality technology (45m20s).
- The experience of working or learning remotely during the pandemic may have given young people a unique perspective on the benefits of in-person work and social interactions, potentially making them strong advocates for in-person work in the future (45m36s).
- Personal experiences and career development are often tied to in-person interactions and networking, as illustrated by Varun Mohan's own career path, which was built through in-person connections in San Francisco (45m56s).
- Some individuals, including Varun Mohan, may prefer in-person work and social interactions, even if they have the option to work remotely, highlighting the importance of personal preference and individual differences in work styles (45m51s).
Twist 500, upcoming events, and what VCs seek in startups (46m3s)
- Varun Mohan has been added to the Twist 500, a list of the 500 best private companies, due to his work on a project considered super important (46m10s).
- The Twist 500 will have an event for its founders in Austin, featuring a party at a Korean vegetarian place, and Varun is invited to attend (46m52s).
- Varun is welcome to return to the podcast in the future to provide updates on his business and share insights (47m11s).
- The podcast is geared towards founders, providing educational content on building large, sustainable startups that can change the world (47m47s).
- When evaluating a seed-stage company for investment, venture capitalists consider factors such as team, product, and customer, in addition to user numbers and revenue (48m22s).
- A startup's success can be measured by its ability to assemble a team, build a product that delights or services a market, and attract customers (48m30s).
- As a business matures, metrics such as customer base, customer spending, and profitability become more important (48m41s).
- The podcast host, Jason, invites questions from the audience, including one about what venture capitalists look for when vetting a company for investment (48m0s).
- At a seed company, the ideal founding team size is considered to be three members, as this allows for a balance of skills and reduces the risk of one person quitting or being fired, with the remaining members able to continue running the company (49m24s).
- Venture capitalists (VCs) often look for a team of two to four co-founders when investing in early-stage startups, as a solo entrepreneur may be seen as too weak or unconvincing to attract top talent (49m54s).
- VCs may view a solo entrepreneur as someone who is unable to inspire confidence in others, making it difficult to build a strong team and achieve success (50m25s).
- Many successful startups, such as Google, Facebook, and YouTube, were founded by teams of two or more people, with co-founders playing a crucial role in the company's early success (50m50s).
- After the founding team, VCs may look at the product and product roadmap, as well as talk to pilot customers, to assess the startup's potential (51m22s).
- While being a solo entrepreneur may be seen as a weakness, it's possible for individuals to join an existing team and contribute to its success, even if they're not the founder (51m49s).
- VCs are looking for exceptional teams and leaders who can drive growth and achieve a high return on investment, often referred to as a "lunatic" who can inspire and motivate others to join their mission (52m6s).
- The pursuit of building a successful startup is considered an elite endeavor, requiring a high level of skill, dedication, and perseverance, similar to the challenges faced by Jedi in the Star Wars universe (52m23s).
Hyper competition in AI services and niche market strategies (53m2s)
- The concept of hyper competition in the AI agent services market is discussed, with 4,712 companies expected to deliver AI agent services by 2025, and the question is raised of how to stand out in such a crowded market (53m41s).
- One strategy for success in a crowded market is to focus on finding a customer base that loves the product or service, and to delight them, rather than trying to compete with everyone else (54m32s).
- The importance of identifying an ideal customer profile is emphasized, citing examples of companies such as SpaceX, Tesla, and Grammarly, which have successfully targeted specific customer bases (54m39s).
- SpaceX's early success was driven by securing NASA as a customer, while Tesla targeted rich geeks who wanted a high-tech, electric sports car (54m45s).
- Grammarly, on the other hand, targeted students who needed help with grammar and writing, and was able to expand its customer base as those students entered the workforce (55m31s).
- The key to success in a crowded market is to focus on a specific customer base, keep expenses low, and delight those customers, rather than trying to compete with everyone else (56m25s).
- To select a niche approach, it's essential to identify a specific customer subset with a particular pain point, such as people who need to write business memos, and then find them, for example, by targeting management consultants whose output is a lot of paid words in documents like reports and memos (56m37s).
- Management consultants, such as those from McKinsey, are potential candidates because their output is a lot of words in documents that are paid for, and errors in these documents can have significant consequences (57m21s).
- Other potential candidates include people who write for magazines, books, and newspapers, as errors in these publications can also have serious repercussions (57m28s).
- Identifying a customer subset with a specific pain point, such as writing business memos, requires understanding their output and the consequences of errors in that output (57m7s).
- Understanding the customer and surviving in a crowded space are two key strategies for success, especially in a niche approach (58m10s).
Audience Q&A and closing remarks (58m26s)
- Viewers can ask questions to Jason by attending the news live streams, which are usually held on Mondays, Wednesdays, and Fridays, and time is made for audience Q&A whenever possible (58m27s).
- The host encourages viewers to bring hard questions and business problems to the show, as there have been a couple of generic questions lately (58m39s).
- The host, Alex, announces that the show, Twist, will be back on Monday with lots more to come this year and next year (58m49s).