Jul 28, 2026 · 1h 10m · latent-space
OpenAI’s Vision for the AI Super App — Akshay Nathan, OpenAI
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
OpenAI Core Product Engineering Lead Akshay Nathan joins Latent Space to discuss the architectural vision behind ChatGPT Work, the transition to interactive artifacts, and how AI shifts the bottleneck of software creation from technical execution to human taste.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. How this is scored →
The hosts as informed peer 5.0 Guest teaching 4.1 Guest disagreement 1.1 The hosts pushing back 1.7
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 58:39 Challenging premise on memory depth differences
Akshay directly rejects Vibu's framing that chat session memory is fundamentally shallower or less valuable than project-level work memory.
Hardest push from the hosts ▶ 51:00 Pushing on lack of persistent semantic data layersSwyx challenges the adequacy of just-in-time runtime data fetching via MCPs, arguing that heavy data workflows still require caching and semantic warehouse layers.
Biggest teaching moment ▶ 16:39 Explaining harness convergence and sandbox executionAkshay educates the hosts on why OpenAI decided to base knowledge work tools on Codex's sandbox computer environment rather than traditional conversational chat harnesses.
The host holds their own ▶ 27:31 Demonstrating 1.7B token auto-research web artifactSwyx showcases an advanced 3D simulator and hyperparameter tuning lab he built using multi-agent self-play and 1.7 billion tokens, demonstrating deep practitioner expertise.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Akshay Nathan's Journey From No-Code to OpenAI | 4 | 3 | 1 | 1 | Swyx introduces Akshay's background spanning fintech, Walrus, and Airtable, drawing parallels to ChatGPT Work as the ultimate no-code super app. Akshay elaborates on how LLMs fulfilled the thesis of democratizing code abstractions. | |
| Enterprise AI Realities and Meeting Users Where They Are | 4 | 4 | 1 | 2 | Swyx probes into enterprise learnings and distinguishes product-led onboarding from forward-deployed engineering. Akshay explains the necessity of product-level guidance over custom deployment motions. | |
| The Genesis of ChatGPT Work and the Super App Vision | 4 | 5 | 1 | 1 | Akshay recounts the internal realization when non-developers at OpenAI started aggressively adopting Codex for finance and marketing, sparking the ChatGPT Work super app merge. The hosts ask clarifying questions about user positioning. | |
| Comparing Agent Harnesses, UX Affordances, and Team Structures | 5 | 4 | 1 | 2 | Swyx asks technical questions regarding whether the underlying harness between Codex and ChatGPT Work is shared or differentiated. Akshay clarifies that the harness and knowledge work plugins are unified while the UX abstractions and Git defaults diverge. | |
| Product Philosophy: Merging Interfaces vs. Maintaining Silos | 5 | 5 | 1 | 2 | Swyx explores the oral history of harness engineering from classic O1 chat to Codex. Akshay articulates the cyclical divergence and convergence cycle between latency-optimized conversational chat and flexible sandbox environments. | |
| Model Selection, Reasoning Tiers, and the New Slider UI | 6 | 4 | 1 | 2 | Swyx and Vibu analyze model selection trade-offs, mentioning Sol Light with Goal versus Ultra configurations. Akshay details how multi-agent ultra modes are optimized for parallelizable explorations while advising defaults for everyday users. | |
| Live Demo: Agentic Excel Artifacts and Future Collaboration | 5 | 4 | 1 | 2 | The hosts inspect live spreadsheet artifact outputs generated by ChatGPT Work. Akshay explains how artifact fidelity builds user iteration trust and addresses the future of team-wide context sharing. | |
| Interactive Sites as the Next-Generation Knowledge Artifact | 7 | 3 | 1 | 1 | Swyx presents an extensive personal case study building an interactive 3D board game simulator using 1.7B tokens and auto-research sites. Akshay agrees that HTML sites are replacing static slide decks as high-bandwidth knowledge artifacts. | |
| Product Scoping: Expanding From Developers to the General Public | 5 | 4 | 1 | 2 | The conversation examines how products expand from developer audiences to general knowledge workers. Akshay outlines OpenAI's phased sequencing strategy from software engineers to broad productivity tools. | |
| Power User Strategies, Agentic Search, and Performance Reviews | 5 | 4 | 1 | 2 | Akshay shares how agentic search across Slack and codebases assists in gathering performance review context. Swyx questions etiquette and edge cases, while Vibu contributes his own Markdown-based memory setups. | |
| Scaling ChatGPT Work to 10 Million Users and Brand Strategy | 4 | 4 | 1 | 1 | Swyx discusses the 10M user launch metrics and questions brand segregation between Codex and ChatGPT Work. Akshay affirms that Codex remains a dedicated developer brand while Work targets general knowledge tasks. | |
| OpenClaw Influences, Personal Workflows, and Extensibility | 6 | 4 | 1 | 2 | Swyx connects ChatGPT Work's persistent computer environment to OpenClaw's local OS paradigm and pushes on data warehousing requirements. Akshay details personal automation workflows and plugin extensibility. | |
| Sub-Agent Architecture, Ultra Mode, and Interface Design Trade-Offs | 6 | 4 | 1 | 2 | Vibu and Swyx explore sub-agent prompting, model steering, and UI visibility. Akshay outlines the intentional balance between showing multi-agent reasoning and preventing interface clutter. | |
| Memory Architecture Evolution and Chronicle Computer Context | 5 | 5 | 2 | 2 | Akshay explicitly challenges Vibu's premise that conversational memory is fundamentally shallower than project memory. The group then discusses Chronicle's passive background computer tracking. | |
| The Evolution of Product Development, T-Shaped Roles, and Taste | 5 | 4 | 1 | 2 | Swyx brings up tech job shifts with a humorous quote, while Akshay presents the model of T-shaped generalists where AI augments secondary skills and human taste guides idea generation. | |
| Defining Productivity: Prioritizing Progress Over Motion | 4 | 4 | 1 | 1 | Akshay reflects on defining productivity as progress over mere motion and emphasizes fast feedback cycles over proxy metrics like story points or token volume. |
Statements from this episode (22)
Nathan: Open-ended AI chat interfaces confuse enterprise users without guided use cases
“Using these models and these products, you have this box and you can say anything to it, which is the magic, but it's on the flip side. It also means that, like, you don't know what to do with it, and in enterprise, I think a big part of that is, like, actuall…”
Nathan: Untapped enterprise agent market is 10x to 100x larger
“So now like we're seeing with agents, like there is Probably a contingent of like early adopters still who, you know, truly get it. We're like, you know, you can do anything. You just have to make sure the right context is there. It's connected to the right to…”
Nathan: OpenAI saw an inflection in Codex adoption among non-developers
“When we release codex or even internally add codex. Like it was really surprising to us. I think we recently put out some stats on this, but there was this like real inflection of like adoption among non-developers at OpenAI.”
Nathan: Codex and ChatGPT Work share the same underlying agent harness
“So the harness is the same. The harness is shared. On, In both of the products, we made improvements to the harness to make it good for knowledge work, especially as it relates to plugins or computer use or artifacts. You get that power regardless of what your…”
Nathan: AI models are experiencing another step-function capability jump
“The models were getting infinitely more capable. That's happening again. I think it's like another step function jump now.”
Swyx: More users should switch to Terra as Sol hits capacity limits
“More people should use Terra. One, because Sol keeps running out of capacity.”
Nathan: Artifact quality improved dramatically over GPT-5.4 and GPT-5.5
“One of the big, like, pushes that we made for this launch was, like, artifacts, right? Like, both on the model side, like, I think if you compare this with 5.5 and 5.4 before that, you'll see that there's been pretty dramatic improvements in the quality of the…”
Nathan: OpenAI built its model slider fully within a site artifact
“Even the model slider that you guys were referencing earlier, like that was developed almost fully in a site. Like, you know, the collaboration between design and engineering and product on that was like on a site where we play with, you know, the affordance a…”
Nathan: OpenAI Finance Team Replaced Slide Decks and Spreadsheets with Sites
“I was actually talking to someone the other day who was on like our corporate finance team and like they're mentioning how like now when they have these reports that they're working on as a team month to month, historically those things were in, in slide decks…”
Swyx: Generated Playable 3D Board Game in ChatGPT Using 1.7B Tokens
“So over the weekend, I took like, 30 photos and just threw it into ChatGPT. 1.7 billion tokens later, out comes this site with a fully playable thing with three D block placement and everything because it requires physical blocks and I needed friends to train …”
Nathan: OpenAI sequencing agents from developers to knowledge workers to everyone
“The vision is like bring useful agents to everyone. We started with like developers Historically are like early adopters that are willing to put up with more friction, set things up, et cetera. Like that's where, you know, Codex started. I think the next oppor…”
Nathan: I would never write a performance review solely via AI
“I would never write something via, like, solely via AI and, like, present it as, like, a review for someone. What I was talking about is more, like, gathering context.”
Swyx: ChatGPT Work is OpenAI's most successful launch, surpassing GPT-5
“I have pretty much said this is the most successful launch in a long time. I think even more successful personally than five point O”
Nathan: OpenAI plans to keep developing Codex specifically for developers
“Like, I think we fully intend to like, you know, treat developer, like developers have been, you know, a core market for us for so long. And like, there's so much more that we can do to make Codex great specifically for software development, and we'll continue…”
Nathan: ChatGPT Work includes persistent computer environments across sessions
“In ChatGPT work in web and mobile, like, you get access to those, like, persistent computer environment where, you know, you can store files, and those files stay around between sessions.”
Nathan: ChatGPT Work will not completely replace open-source OpenClaw
“I don't think so. I think that there's going to be, you know, there's always a need for, like, this, like, incredible, like, open source technology that, that team has built”
Nathan: Sub-agents and tool calling significantly raise MCP execution ceiling
“It's very possible that there's a, the ceiling on what can be done, you know, with MCPs and like calling out to these server-friendly services has been raised substantially.”
Nathan: OpenAI limits sub-agent UI visibility to prevent user overwhelm
“There's another, you know, iteration of this where like you can see exactly what they're doing and things like that, which I think is like, you know, could verge on like overwhelming with information. And so this is like the deliberate trade-off that we've mad…”
Akshay Nathan: AI will make tech workers into T-shaped generalists
“My suspicion is that there's everything, everyone will be, like, T-shaped in a way, and that, like, AI will enable everyone to become a generalist. Like, you know, things that, like, I never would be able to, like, come up with a design before, and, like, even…”
Akshay Nathan: Software creation bottleneck has shifted to ideas and taste
“I think the bottleneck some becomes like sort of like ideas and taste, I guess. I think because anyone can build now, I think it really is the era of like bottoms up ambition. And because there's so much to be built, like you're always going to be bottlenecked…”
Nathan: Traditional software productivity proxies are falling apart in AI era
“I think with AI now, those proxies starting to fall apart, like, you know, and the number of tokens you use or the number of pull requests you make are like no longer like maybe as hyper correlated with that. Is your team able to hit the goal or are they on tr…”
Nathan: AI tooling makes motion easier, risking conflation with true progress
“I think maybe the trap is like conflating motion and progress. I think motion is much easier now than ever before because of the tooling that we have, but progress requires you to be like very prescriptive and deliberate about like what you're actually trying …”