Key Moments
The State of AI in production — with David Hsu of Retool
Latent Space PodcastKey Moments
Retool CEO David Hsu discusses AI in production, internal tooling, open source vs. OpenAI, and building an efficient company.
Key Insights
Retool's success is rooted in prioritizing customer value over hype and prioritizing building a solid product over premature scaling.
Hiring former founders early on instilled an outcome-oriented culture with minimal politics, vital for maintaining agility.
Raising less money at lower valuations, even during peak hype, proved beneficial for long-term company health, employee equity, and morale.
Retool, built for developers, aims to empower them by facilitating the rapid creation of internal applications, not by replacing coding.
The significant value of AI lies in automating entire processes and enabling AI-infused applications, rather than solely assisting with coding.
Open source is expected to win in AI tooling due to rapid innovation and community-driven development, making partnerships with projects like Pinecone strategic.
While AI adoption is growing internally, the leap to external customer-facing applications faces higher hurdles due to accuracy and privacy concerns.
AI's impact on jobs is perceived as highest in operations, with engineering driving adoption through experimentation and tooling integration.
The future of AI development may involve more automation and workflow integration rather than direct human-AI conversational interfaces for complex tasks.
Hiring processes are adapting to AI's influence, focusing on fundamental problem-solving skills over rote coding abilities.
FOUNDING PHILOSOPHY AND EARLY GROWTH
David Hsu of Retool shares his unconventional founding story, including deliberately not presenting at YC demo day due to an under-developed product. This decision stemmed from a core philosophy of under-promising and over-delivering, prioritizing genuine customer value and product substance over superficial validation. Retool focused on building a robust business before seeking significant external validation, a strategy that proved effective in the long run. This approach also influenced their hiring, emphasizing former founders to foster an outcome-oriented and agile culture.
THE STRATEGIC ADVANTAGE OF CAPITAL EFFICIENCY
Hsu advocates for a contrarian approach to fundraising, notably raising less money at lower valuations, even during periods of extreme market hype. He argues that maximizing fundraising numbers can be detrimental to employees through dilution and can create unrealistic expectations for future growth. This capital efficiency allows for a more sustainable path, focusing on building a high-quality company and maintaining team morale, which has served Retool well amidst market fluctuations.
RETOOL'S POSITIONING AS A DEVELOPER-FIRST TOOL
Retool positions itself not as a 'low-code' solution, but as a 'developer-first' platform, emphasizing its appeal to professional developers. The core value proposition is enabling developers to build essential internal tools and applications rapidly, addressing the common industry pain point of developers disliking the grunt work often associated with internal tooling. Retool provides flexible building blocks, allowing developers to construct sophisticated applications without being constrained by low-code limitations.
EVOLUTION OF AI PRODUCT STRATEGY AT RETOOL
Retool's AI strategy evolved from an early, almost joking inclusion on roadmaps to a significant product focus. Initially, the company explored AI's potential in speeding up code generation, but found that while helpful, it wasn't a disruptive leap. The primary focus has shifted to enabling developers to build AI-infused applications faster, a move driven by the belief that nearly all internal applications will incorporate AI within the next few years, enhancing productivity through automation and intelligent features.
INTEGRATING AI: AUTOMATION OVER ASSISTANCE
Hsu highlights the difference between AI as a coding assistant and AI as an automation engine. He believes the true economic productivity gains will come from automating entire processes, rather than just speeding up individual tasks like coding. Retool's 'Workflows' product is seen as a key enabler for this automation, allowing for the chaining of AI steps to create end-to-end automated processes, which is considered more impactful than conversational AI for one-off tasks like those typically handled by ChatGPT.
EMBRACING OPEN SOURCE IN THE AI LANDSCAPE
Retool's strategy involves partnering with and leveraging open-source AI technologies, particularly for supporting infrastructure like vector databases. Hsu believes the open-source movement will ultimately win in AI tooling due to rapid innovation and community contributions. Betting on a single commercial vendor is seen as riskier than investing in the broader open-source community, allowing for greater transparency and the ability to contribute fixes and improvements directly.
AI ADOPTION: INTERNAL USE CASES LEAD THE WAY
The survey data indicates that AI adoption is primarily driven by internal use cases, with a lower percentage of AI tools currently in production for external customers. This is attributed to the higher acceptable tolerance for errors and hallucinations in internal applications, where human oversight is present. As AI moves towards customer-facing roles, the bar for accuracy, reliability, and data privacy becomes significantly higher, posing a greater challenge for widespread consumer-level adoption.
THE FUTURE OF WORK AND AI'S IMPACT ON JOBS
The perception is that AI will significantly impact jobs, particularly in operations, while engineers are seen as the primary drivers of AI adoption through experimentation. There's a nuanced view on hiring, with a need to test fundamental skills while acknowledging AI's role as a productivity tool. The focus is shifting towards assessing critical thinking and the ability to effectively leverage AI, rather than solely penalizing its use.
THE ROLE OF MULTIMODALITY AND UNCONVENTIONAL USE CASES
The conversation touches upon the growing importance of multimodality in AI, including image, audio, and video processing. While Silicon Valley focuses on novel AI use cases, Hsu points out that many of the most impactful applications are emerging from traditional industries, such as AI-generated clothing patterns for a fashion manufacturer. This highlights the need to look beyond obvious AI applications to find real-world business problems that AI can solve.
STRATEGIC PARTNERING VERSUS IN-HOUSE DEVELOPMENT
Retool's product strategy hinges on building core competencies in-house, such as their workflow automation engine, and partnering for technologies where a vibrant ecosystem already exists, like vector databases. This approach ensures they focus on areas where they can offer a unique or superior take for developers, while leveraging the strengths of external open-source communities for supporting technologies. The goal is to provide developers with reusable building blocks rather than finished solutions.
THE CHALLENGE OF ACHIEVING UBIQUITY
While Retool aims for broad adoption among millions of developers globally, achieving true ubiquity outside of the Silicon Valley tech bubble presents a significant challenge. The strategy has evolved from a sales-led approach to prioritizing bottom-up, product-led growth. The focus is on developers discovering and loving the product organically, which necessitates making Retool accessible and valuable to a vast number of engineers across diverse industries, not just software companies.
PHILOSOPHICAL MUSINGS ON AI AND INTENTIONALITY
Beyond the practical applications, Hsu reflects on the philosophical underpinnings of AI, particularly concerning intentionality and the emergence of AGI. He contrasts human programming for survival and reproduction with the current state of AI development, questioning what criteria would truly signify artificial general intelligence. This philosophical exploration underscores the profound questions surrounding AI's future and its potential divergence from human-like consciousness.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Books
●Concepts
●People Referenced
Common Questions
Retool is a platform that allows developers to build internal tools quickly using pre-built components, akin to using 'Legos for code'. It is specifically designed for developers, not low-code users, and aims to reduce the grunt work involved in creating internal applications.
Topics
Mentioned in this video
Mentioned as a competitor in the internal tooling space, but Retool differentiates itself by being developer-first.
Mentioned as an example of a large enterprise that might be hesitant to use public AI models like ChatGPT for sensitive tasks due to privacy concerns.
A startup accelerator program that Retool participated in (Winter '17), and its demo days are discussed in the context of early-stage startup presentations.
Mentioned as an example of a vector database provider that Retool partners with, highlighting a preference for open-source or community-driven solutions.
A major customer of Retool, with 11 business units using the platform, highlighting the challenge of reaching the broader employee base.
Joseph Nelson, founder of RoboFlow, is mentioned as a source of a question regarding Retool's enterprise vs. ubiquity sales strategy.
Jeffrey Wang, co-founder and Chief Architect of Amplitude, is mentioned for his perspective on using vector databases.
Mentioned as a competitor in the workflow automation space that Retool competes with and differentiates from.
A platform mentioned in relation to open-source multimodal AI models like Phi-X.
Mentioned as a potential future competitor in the broader productivity and internal tooling space.
Mentioned as an AI model with multimodal capabilities, relevant to the discussion on the future of AI modalities.
Mentioned as an open-source model that currently lags behind GPT-4 in performance, leading customers to prefer hosted models.
An open-source multimodal AI model released by Hugging Face, discussed in the context of multimodal AI adoption.
Mentioned as a coding assistant that saves engineers time (10-20%) but doesn't represent a massive leap in productivity due to low-level frameworks.
A software company founded by David Hsu that provides a platform for developers to build internal tools.
Discussed as a key component for building AI-enabled applications, with Retool integrating them directly.
Discussed as a significant driver of AI utility for one-off tasks and as a factor in reduced Stack Overflow usage.
Discussed as a leading model for AI applications, with a higher NPS than GPT-3.5, and its use in enterprise is growing despite privacy concerns.
Author whose work on 'The Everlasting Man' and the concept of ants in a colony as a sign of intelligence is referenced in a discussion about AI abstraction levels.
Mentioned in the context of Y Combinator; David Hsu recalls him being ambitious and helpful during their batch.
Co-founder and CEO of Retool, interviewed about the state of AI in production and Retool's development.
More from Latent Space
View all 186 summaries
86 minNVIDIA's AI Engineers: Brev, Dynamo and Agent Inference at Planetary Scale and "Speed of Light"
72 minCursor's Third Era: Cloud Agents — ft. Sam Whitmore, Jonas Nelle, Cursor
77 minWhy Every Agent Needs a Box — Aaron Levie, Box
42 min⚡️ Polsia: Solo Founder Tiny Team from 0 to 1m ARR in 1 month & the future of Self-Running Companies
Found this useful? Build your knowledge library
Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.
Try Summify free