OpenAI Podcast: Episode 7 – Live from DevDay (October 8, 2025)
Episode Overview
Live from OpenAI’s DevDay, host Andrew Mayne brings together leaders from four companies using AI to transform their industries: education (Caleb Hicks, SchoolAI), web development (Danny Grant, Jam.dev), healthcare (Zach Lipton, Abridge), and software tooling (Lee Robinson, Cursor). The episode is a fast-paced, insightful exploration of how OpenAI’s latest features—especially the Agent SDK and new developer tools—are empowering teams to build advanced, user-focused applications, and how these innovations are changing what’s possible across sectors.
Key Discussion Points & Insights
1. Advances in AI-Driven Education — Caleb Hicks, SchoolAI
The Evolution of AI in Classrooms
-
Progression of Attitudes: Years ago, AI was broadly banned; now, educators are moving towards seeing AI as essential for productivity and student learning.
“Two and a half years ago, it was everyone under the sun was just banning AI altogether. ... Now, most people orient to ‘yeah, we have to teach this.’”
— Caleb Hicks [02:57] -
AI as an Individual Tutor: SchoolAI focuses on safe, managed AI tutors accessible for every student, with orchestration for classroom integration and real-time dashboards for teachers.
“The special part is when you start doing these one-time, guardrailed, safe managed AI tutors ... students are interacting and the teacher gets a real time dashboard of how the students are doing.”
— Caleb Hicks [05:32]
Impact of OpenAI's New Tools
- Agent SDK and Builder: Accelerates development and democratizes “expert” engineering for non-technical users. SchoolAI has built similar proprietary systems but sees OpenAI’s new agent builder as game-changing.
"Being able to drag and drop something like file search, the permission structure seemed really well thought out, particularly for the classroom."
— Host [09:00]
Product Stack & Teacher Empowerment
- Layered AI Assistants:
- Teachers and admins have AI assistants (no prompt engineering required).
- Lesson planning tools ("101, checkers level features").
- Student-targeted, safe AI tutors with social-emotional check-ins and formative quizzes.
- Real-time data surfaces which students need attention, aiding with large classroom management.
“Teachers have to make this impossible choice. … We think about what we've been able to build with OpenAI is give teachers almost a GPS for impact.”
— Caleb Hicks [06:56]
Evaluations (Evals) and Iteration
- Scaling Impact: Even tiny improvements (2–3%) matter when serving millions of students.
- Rapid Experimentation: Built-in evals let them prototype and test teacher-created AI tutors for each lesson, supporting fast feedback loops.
“Take 10 and do the evals and that'll be fun.”
— Caleb Hicks [11:53]
Advice for Developers
- Start Simple: Use GPT Builder, expand with Agent Builder and other OpenAI tools as needs grow.
“Just start in the GPT builder and then expand ... Agent Builder that we saw today is going to be another fun one to start connecting the dots.”
— Caleb Hicks [13:28]
2. Rethinking the Web & Software Usability — Danny Grant, Jam.dev
New Paradigms in Web Interaction
-
Please Fix Tool: Lets anyone—non-coders included—fix website issues via a browser extension, editing sites like a Google Doc or Figma and auto-generating clean pull requests.
“It helps anyone fix what's broken instantly without writing code ... it lets you edit your site right there like it's a Google Doc.”
— Danny Grant [14:25] -
OpenAI’s Impact:
- The new Agent SDK and in-ChatGPT apps represent a “Web Four” moment—“read, write, think” interactions.
“I think we just saw a whole new way of experiencing the web ... a lot less mechanical and a lot more stream of consciousness.”
— Danny Grant [15:41]
Empowerment & Democratization
- Non-Engineers Building Impactful Tools:
- Examples: Firefighters and church members with no software experience, enabled by AI to build useful community tools.
“We are about to see the Cambrian Explosion of software ... it's one of the best things for humanity.”
— Danny Grant [18:06]
Iteration, Design, and Product Quality
-
Continuous User-Focused Improvement:
- Every user who signs up for Jam is personally contacted by the team.
- Prioritize emotional “wow” moments for users:
“The difference between a fine designed product and a well designed product is that the well designed product changes the world.”
— Danny Grant [16:39] -
Internal Tools Becoming Products:
- Internal tooling frequently becomes the main product because it solves real, daily problems.
“Three out of four ... their startup started as an internal tool at their company that they needed.”
— Danny Grant [21:20]
Aspirations for AI Tooling
- Evals and Autonomy:
- Desire for AIs that can automatically generate and optimize their own evals for further self-improvement.
“What if the agents could improve themselves? If we had that at Jam, we could move a lot faster.”
— Danny Grant [23:40]
Founder Advice
- Build What You Love:
- “If your startup works out to your wildest dreams, you're going to work on it for like 10 years... you better just love the problem.”
— Danny Grant [25:31]
- “If your startup works out to your wildest dreams, you're going to work on it for like 10 years... you better just love the problem.”
3. Transforming Healthcare Workflows — Zach Lipton, Abridge
The Clinical Documentation Crisis
- Pre-AI Pain Point:
- Doctors spent 2 hours on paperwork per 1 hour of patient care, often logging in at home (“pajama time”).
- Abridge’s ambient AI listens to doctor-patient conversations, auto-drafts notes, and prepares documentation artifacts.
“It was a situation where technology was pulling doctors away from patients rather than bringing them closer.”
— Zach Lipton [26:38]
Quantifiable & Emotional Impact
- Time Saved:
- As much as an hour or more per day, with direct stories of AI “saving marriages” by giving doctors back family time.
“We see doctors saving as much as an hour or more a day ... I actually got to have dinner with my family every night this week for the first time in 10 years ... Abridge saved my marriage.”
— Zach Lipton [27:28]
OpenAI Tools for High-Stakes Domains
-
Agent Kit & Developer Tools:
- Having orchestration, evaluation, and production monitoring in one place means Abridge can stop reinventing the wheel and focus on core product.
“Seeing OpenAI take a strong position and put a kind of comprehensive offering that brings together a lot of these things... allows us to focus more on the content.”
— Zach Lipton [29:34] -
AI in Developer Tooling:
- Codex and similar tools are dramatically accelerating code-related work internally.
Hallucination, Safety, and Trust
-
Defining Error Acceptability:
- Medical AI must be precise—“hallucinations” mean something domain-specific.
“In the context of medical note taking... Even if some of the information might be correct or plausible, that's not within ours.”
— Zach Lipton [31:56] -
Custom Eval Pipelines:
- Achieved ~97% recall in detecting documentation errors, using models for validation and custom pipelines for remediation.
“We create our own special purpose models that are able to... process for each one: does it contain an error of an unacceptable variety, like of what kind? We can do that with about 97% recall at this point.”
— Zach Lipton [33:21] -
Advice:
- “It all starts with getting really crisp about what you really mean” when defining hallucinations and evals.
— Zach Lipton [34:18]
- “It all starts with getting really crisp about what you really mean” when defining hallucinations and evals.
Expanding the AI-Enhanced Moment
- Beyond Scribing:
- The most important moment in healthcare is the doctor-patient conversation; Abridge’s vision is to support the whole workflow from pre-visit to post-visit, including real-time decisions, insurance support, and more.
“The central thesis wasn't just about scribing. Central thesis was about medical conversations ... it's the most important moment in the entire experience of healthcare.”
— Zach Lipton [35:10]
Trust Building in High Stakes
- Long-Term Relationships:
- Trust in medical AI is earned “every single day," through delivery, partnership, and ongoing support.
“Trust is, trust is earned. Every single day, trust is earned.”
— Zach Lipton [42:55]
4. Coding with and for AI — Lee Robinson, Cursor
From Code Completion to Autonomous Agents
- Cursor’s Evolution:
- Transition from simple autocomplete—using Codex/GPT3—to building and dogfooding advanced, agent-driven workflows that can self-correct and integrate external data.
“Turns out there's a lot that goes into it, especially from maybe simple text autocomplete, but to where we're at now with fully autonomous coding agents who can self correct.”
— Lee Robinson [45:20]
Internal Product Development Dynamics
- Features Need Internal Product-Market Fit:
- All team members can push features; adoption and enthusiasm internally drive what gets to users.
- Reliance on metrics and “vibes”—qualitative feelings from frequent use—guide decisions.
Evals and Continuous Model Improvement
- Dogfooding and Custom Models:
- Cursor internally tests new models, integrates multiple backends, and uses RL online training for tab completions, with model updates every 30 minutes.
“We do online reinforcement learning where we can actually roll updates every 30 minutes to the model ... whether they're accepting or rejecting changes that the autocomplete suggests.”
— Lee Robinson [49:11]
User Demographics and Accessibility
- Broader User Onramp:
- Originally for professional engineers, but making it easier to welcome PMs, designers, support staff, and complete beginners.
“A lot more product managers, a lot more designers are supporting support team, uses Cursor quite a bit. … So, we've seen this really resonate with that type of persona.”
— Lee Robinson [50:12–51:07]
Training and Learning Inside Coding Tools
- Teaching as You Code:
- Cursor is adding explainers for beginners, helping students and new coders learn context like “what is a context window?” and practical computer science.
“I really want to build this into cursor. ... To be able to use the tool and learn as they're building.”
— Lee Robinson [57:58]
The Future of Coding
- Where Are We Headed?
- The "vibe coding" era: prototyping ideas quickly is now cheap and easy, but production software is still complex ("the iceberg meme"). AI will increasingly handle mundane engineering tasks so devs can focus on value creation and creativity.
“There’s still so much of software engineering ... a lot of mundane, repetitive tasks that engineers are not excited by. ... I imagine a world where you wake up in the morning and you’re able to review code that’s already been tested and generated.”
— Lee Robinson [54:18]
Education's Blind Spot
- Gap in Academia:
- Many CS programs still don’t teach agentic coding or AI-powered coding, highlighting the need for widespread upskilling.
“I asked the students in the CS program, what are they teaching you about agentic coding? ... Nothing. Not even a one day class on it.”
— Host & Lee Robinson [56:29]
Best Practices for Adopting AI Coding Tools
- For Pros:
- Start by using tab completions, then delegate bigger tasks to AI agents.
- For Newcomers:
- Use agent view and natural language to ease into code, progressively deepening technical understanding.
Memorable Quotes & Moments
-
On AI's Impact for Educators:
“If you're graduating from high school and you're competing for colleges or jobs and you don't know how to use AI yourself, you're at a severe disadvantage."
— Caleb Hicks [03:13] -
On the New Web Experience:
"Web one is read, Web two is read-write... maybe this is like, web four: read, write, think. I think we just saw a whole new way of experiencing the web."
— Danny Grant [15:41] -
On Trust in High-Stakes AI:
“Trust is earned every single day.”
— Zach Lipton [42:55] -
Describing Vibe Coding:
"Vibe coding is this idea of it's never been easier to create prototypes. ... You can just try out ideas ... That doesn't necessarily mean that you have to ship that code ... There's actually a lot more to building, delivering complicated software."
— Lee Robinson [58:55] -
On Non-Coders Creating Value:
"Some of the stories we hear are awesome ... building software for firefighters, for churches, with no software experience ... We're about to see the Cambrian Explosion of software."
— Danny Grant [18:06]
Notable Timestamps
- [02:57] – Caleb Hicks on the evolution from AI bans to AI as a classroom necessity.
- [05:32] – Caleb Hicks explains safe, managed classroom tutors and real-time dashboards.
- [11:53] – Fast evals and rapid iterative prototyping in education.
- [14:25] – Danny Grant describes “Please Fix” browser extension.
- [15:41] – “Web four” and new browser paradigms.
- [18:06] – Stories of non-coders empowered by new tools.
- [27:28] – Zach Lipton: tangible impact on doctors’ time and lives.
- [33:21] – 97% recall in detecting documentation errors for medical compliance.
- [42:55] – Building trust in medical AI is a persistent, daily process.
- [45:20] – Lee Robinson on the move from autocomplete to autonomous coding agents.
- [57:58] – Cursor’s plans to teach coding alongside development work.
Conclusion
Across education, web development, healthcare, and software tooling, OpenAI DevDay’s tools and SDKs are accelerating product cycles, lowering technical barriers, and empowering users of all backgrounds. Guests emphasized the importance of deep domain understanding, fast iteration, robust evaluation, and, above all, trust—especially in high-stakes contexts. The future, as seen here, is one where AI becomes an intuitive partner for creation, problem-solving, and learning.
For more on the guests:
Social handles:
- SchoolAI: @GetSchoolAI (X/Twitter), Instagram
- Jam.dev: jam.dev/jam.devPleaseFix
- Cursor: cursor.so
Note: Timestamps reference the MM:SS position in the original episode transcript.