Live from DevDay — the OpenAI Podcast Ep. 7
In a special live episode from OpenAI Dev Day, host Andrew Mayne interviews founders from SchoolAI, jam.dev, Abridge, and Cursor. They discuss how they are using AI to transform education, web development, healthcare, and coding, sharing insights on their product strategies, technical challenges, and excitement for the next wave of agent-based AI tools.
6 min read
This special live edition of the OpenAI Podcast from Dev Day features interviews with four startups leveraging AI to innovate in their respective fields. The discussion explores how these companies are building practical applications and what recent advancements, particularly the announcement of agents and new developer tools, mean for their roadmaps and the future of AI-powered products.
SchoolAI: AI as a Classroom Co-pilot
Caleb Hicks from SchoolAI discusses their mission to place safe, managed AI into students’ hands, functioning as personalized, “classroom connected” tutors.
Key Insights & Product Strategy
- AI’s Role in Education: The journey for educators has shifted from banning AI to embracing it for productivity. The most forward-thinking schools now recognize that teaching students how to use AI is a necessity for future competitiveness. SchoolAI aims to take this further by creating a system-wide AI assistant for teachers, students, and families.
- Technical Stack: Their product consists of three main parts:
- Dot: A user-friendly AI assistant tuned for educational use cases, which enriches teacher prompts to eliminate the need for them to become prompt engineers.
- Tools: Simple, form-based features for generating lesson plans and other materials, serving as an entry point for teachers.
- AI Tutors: The core offering—customizable, guard-railed AI tutors created by teachers. A real-time dashboard allows teachers to monitor student interactions and progress.
- Founder’s Motivation: As a former teacher managing large classes, Caleb’s goal is to empower educators to support the “middle 80%” of students who are often overlooked. The tool acts as a “GPS for impact,” highlighting which students need immediate attention.
Takeaways from Dev Day
- Agent Builder: Caleb is excited about the potential of the new Agent Builder to replace or enhance their custom-built agent systems, empowering non-technical subject matter experts to create powerful tools.
- Standardization: The solidification around standards like MCP servers for AI communication will simplify integrations with partners, allowing them to bring their tools into SchoolAI’s safe, managed environment.
- Evals: Built-in evaluation tools are a significant unlock, saving development time and enabling more rigorous testing, which is critical when serving millions of students.
jam.dev: Democratizing Web Development
Danny Grant of jam.dev introduces “PleaseFix,” a new tool designed to empower non-engineers to make direct changes to websites.
Key Insights & Product Strategy
- “PleaseFix” Tool: A browser extension that allows product managers, designers, and marketers to edit website content and design directly, as if it were a Google Doc. The tool then generates a clean pull request that adheres to the site’s existing design system, removing a common bottleneck in web development for minor changes.
- Vision for “Web 4”: Danny envisions the next phase of the web as “Read, Write, Think,” where interaction is less mechanical and more a stream of consciousness, enabled by agents and integrated apps.
- The Future of Software: He predicts a bifurcation of software into two categories:
- Long-term Software: Human-built, finely-tuned applications like Zillow or Canva.
- Disposable Software: One-time use, dynamic tools whipped up by an agent to solve an immediate need, like a temporary data dashboard.
Takeaways from Dev Day
- Agent-Improved Agents: His team is most excited about the possibility of agents that can improve themselves—for instance, an agent that could automatically write an evaluation for itself based on user data and then use that eval to optimize its own prompts.
- Democratization of Building: The new tools lower the barrier to entry, enabling a “Cambrian explosion of software” where individuals with domain expertise but no coding experience can build impactful products for their communities.
Abridge: Supercharging Medical Documentation
Zach Lipton from Abridge explains how their platform is tackling the clerical burden in healthcare by automatically generating medical notes from doctor-patient conversations.
Key Insights & Product Strategy
- Core Problem: Doctors spend roughly two hours on paperwork for every one hour of patient care, leading to burnout and what’s known as “pajama time”—finishing notes after hours. Abridge saves doctors over an hour a day, improving not just efficiency but also work-life balance.
- Managing High-Stakes AI: In medicine, accuracy is paramount. Abridge has developed a bespoke approach to managing hallucinations, defining them as any information “unlicensed” by the conversation. They’ve built specialized models that achieve 97% recall in identifying and flagging potential errors in generated documentation for remediation.
- Beyond Scribing: Abridge’s vision extends far beyond simple transcription. They see the doctor-patient conversation as the central, most valuable moment in healthcare. Their goal is to build a comprehensive platform that provides real-time decision support, pre-charting assistance, and anticipates documentation needs for billing and insurance.
Takeaways from Dev Day
- Agent Developer Kit: Zach is enthusiastic about having a common platform for agent development. This will unify disparate processes like context engineering, evaluation, and monitoring, allowing his team to focus more on the core medical content and less on tooling.
- Trust in High-Stakes Domains: Trust is “earned every single day.” For developers in critical fields, it requires continuous delivery on product commitments, data security, and service, building a deep, long-term partnership with clients.
Cursor: The AI-First Code Editor
Lee Robinson from Cursor details the evolution of their AI-native code editor and the future of software engineering.
Key Insights & Product Strategy
- Dogfooding Culture: Cursor is built with Cursor. This deep internal usage allows the team to evaluate new features and models not just with hard metrics but also with a qualitative “vibe check” on the developer experience. Features gain product-market fit internally before being released externally.
- Hybrid Model Strategy: Cursor employs an “all of the above” approach. It integrates powerful foundational models from partners like OpenAI while also training its own highly specialized, smaller models for tasks like code completion. Their autocomplete model is updated every 30 minutes using online reinforcement learning.
- Evolving User Base: While built for professional engineers, Cursor’s accessibility has attracted a new wave of users like product managers and designers. This led to features like an agent-first view that is less intimidating for those new to coding.
- The Future of Engineering: Lee believes the next frontier is automating the mundane, non-coding tasks of software engineering, such as managing on-call incidents and bug reports. He envisions a future where engineers primarily review code that was automatically generated, tested, and deployed overnight.
Takeaways from Dev Day
- We Are Still Early: Despite the rapid progress, Lee’s biggest realization is that the adoption of AI coding tools is still in its infancy outside of the tech bubble. There’s a massive opportunity to educate the next generation of developers.
- From Vibe Coding to Coding: The term “vibe coding” aptly describes the new ease of prototyping. However, this is just the tip of the iceberg. As these tools become ubiquitous and integrated with the entire software development lifecycle, this will simply evolve into the new standard of what it means to “code.”