SUMMARY

In an exclusive interview at OpenAI's DevDay 2025, CEO Sam Altman discusses major launches like Apps in ChatGPT and Agent Builder, viral Sora deepfakes, zero-person companies, AGI progress, and AI's impact on jobs and scientific discovery.

STATEMENTS

OpenAI announced Apps in ChatGPT, Agent Builder, Sora API, and Codex updates capable of handling day-long tasks at DevDay 2025.
ChatGPT now has 800 million weekly active users, positioning it as a major distribution platform for developers.
Developers can build apps on ChatGPT using the apps SDK, with recommendations evolving through iteration and documentation.
Agent Builder represents a significant advancement over GPT Builder from two years ago, enabled by improved model capabilities and easier deployment.
Users can now build sophisticated agents with no code via visual tools, uploading files and granting data access in minutes.
The no-code revolution in AI agents lowers barriers, allowing knowledge workers to create tools, potentially exploding software creation and idea testing.
Betting pools speculate on the first zero-person billion-dollar company run by agents, expected in years, not months.
Agents like Codex are approaching week-long autonomy, with rapid progress in task length surprising even insiders.
Technical bottlenecks for longer agent tasks include smarter models, longer context windows, and better memory systems.
Young entrepreneurs today have vast opportunities with AI tools, outpacing what 20-year-old Altman could have built.
Unique advantages for AI products emerge iteratively, such as memory in ChatGPT, rather than being predefined strategies.
On the GDPval benchmark, GPT-5 ranked second to Claude Opus, highlighting enterprise strengths without altering OpenAI's GPT-6 strategy.
AGI is defined by AI outperforming humans in most economically valuable work, with early signs in novel scientific discoveries.
AI is beginning to expand human knowledge through small breakthroughs in fields like mathematics, marking a pivotal AGI milestone.
The "workslop" study reveals low-effort outputs costing $186 per employee monthly, a human issue AI may exacerbate but the economy will correct.
Viral Sora deepfakes of Altman adapted quickly in society, similar to past AI milestones like the Turing test.
OpenAI explores revenue sharing for cameos in Sora videos to incentivize participation amid watermark removal concerns.
A billion knowledge worker jobs will transform post-AGI, creating new meaningful pursuits while challenging current work perceptions.
Global AI safety requires a framework for superintelligent models, starting with testing standards to mitigate catastrophic risks.
ChatGPT aims to be an AI super assistant, not an everything app like WeChat, keeping features separate for user experience.
Exciting agents include industry-specific tools like Codex for law or finance, enabling one-person startups via conversational building.
Voice interfaces for AI are natural for many scenarios but not universal, with smart speakers poised for revival with better AI.

IDEAS

Apps in ChatGPT could redefine distribution, with developers discovering novel mechanics like name-calling or contextual suggestions for user engagement.
Agent Builder's visual no-code interface democratizes sophisticated software creation, compressing development from weeks to minutes.
The explosion of agent-built software will accelerate idea validation, forcing builders to generate concepts faster than ever before.
Zero-person companies, run entirely by AI agents, shift speculation from one-person unicorns to fully autonomous entities soon.
AI task duration progress, from hours to weeks, feels disappointingly fast, highlighting exponential improvements in autonomy.
Envy for today's young dropouts stems from AI's vast opportunity space, enabling builds unimaginable a decade ago.
Competitive edges in AI emerge organically through features like memory, turning tactics into enduring strategies without upfront planning.
AGI's fuzziness sharpens as novel discoveries begin, with Twitter examples showing AI expanding knowledge in small but real ways.
Society's rapid adaptation to AI milestones, like passing the Turing test, suggests quick normalization of scientific breakthroughs by machines.
Workslop isn't AI-exclusive; tools historically amplify both productivity and drag, with self-correcting economies favoring effective users.
Viral deepfakes inoculate society against future unwatermarked fakes, evolving norms faster through controlled early releases.
Sora's personal, low-engagement use cases challenge monetization models, potentially shifting to per-generation payments or revenue sharing.
Post-AGI work may resemble entertainment from today's view, betting on human drives to find meaning in expansive pursuits like space exploration.

INSIGHTS

Rapid AI tooling lowers creation barriers so profoundly that human ideation becomes the new bottleneck, reshaping entrepreneurship.
Iterative emergence of advantages, like ChatGPT's memory, underscores that strategies form from action, not abstraction.
Early AGI signals lie in knowledge expansion, not just benchmarks, marking a subtle but profound shift in human-AI symbiosis.
Societal adaptation to disruptions like deepfakes or Turing tests happens swiftly, turning initial weirdness into normalized utility.
Economic self-correction via tools favors productivity amplifiers, predicting AI will widen gaps between effective and inefficient users.
Global safety frameworks must prioritize testing for superintelligent edges, balancing innovation with catastrophe prevention.
Separating AI assistants from social features preserves personal trust, avoiding the discord of everything-app overload.
Industry-specific agents, mirroring Codex's coding prowess, could spawn solo-founder empires through conversational orchestration.
Voice as an interface revives overlooked potentials in smart devices, blending natural language with seamless, unobtrusive computation.
Future jobs may devalue current "real work" perceptions, urging bets on innate human quests for purpose beyond labor.
Revenue models for generative media like Sora must adapt to private novelty uses, blending subscriptions with micro-payments.

QUOTES

"The amount of software that's going to get written in the world is clearly going to drastically increase."
"We're finally in the moment where it's starting to happen is when I can do novel discovery, when I can expand the total human knowledge base."
"It's only weird once. It's only weird for like three minutes, and then you get used to it."
"People and companies that use tools to get more done will have more ability to influence the future than people who use it to slow organizations down."
"I think great video will be important to that for a bunch of reasons. Spatial reasoning the things we can learn from world models hopefully someday real progress with robotics will be important."
"Let tactics become a strategy. You can start just doing things that work and surprisingly often in the process of that will emerge something that develops into a strategy."
"I kind of hope everything just goes in every direction. And we go. We go do all the things like space does seem very cool to me."

HABITS

Iteratively build and release features, allowing unexpected advantages like memory to emerge naturally over time.
Experiment boldly with new technologies, such as allowing cameos in Sora despite initial strangeness, to drive societal adaptation.
Reflect on past capabilities to appreciate progress, like comparing model improvements from two years ago to now.
Engage directly with user feedback and demos to inspire internal improvements, as seen in reactions to enterprise benchmarks.
Prioritize co-evolution of tech and society by releasing tools early with guardrails, fostering quick learning curves.
Bet on human adaptability by observing rapid normalization of AI milestones, maintaining optimism amid transitions.

FACTS

ChatGPT reached 800 million weekly active users, establishing it as a premier distribution platform.
The GDPval benchmark showed GPT-5 second to Claude Opus in economically valuable tasks across knowledge jobs.
Stanford's "workslop" study found 41% of desk workers encountered low-effort outputs monthly, costing $186 per employee.
AI agents like Codex now handle day-long tasks, with week-long autonomy expected soon via model upgrades.
Sora's cameo feature led to viral memes of Altman, viewed millions of times shortly after launch.
OpenAI's first DevDay two years ago launched GPT Builder; today's Agent Builder enables no-code agent deployment in minutes.

REFERENCES

Apps in ChatGPT SDK for building and distribution.
Agent Builder and Agent Kit for no-code agent creation.
GPT Builder from the first DevDay.
Codex for extended task handling and coding.
Sora API for video generation and cameos.
GDPval benchmark for economic AI performance.
Claude Opus model from Anthropic.
WeChat as an everything app example.
Waymo for self-driving adaptation analogy.
Turing test as a historical AI milestone.

HOW TO APPLY

Start with visual no-code tools in Agent Builder: Upload files, define goals, and grant data access to prototype agents quickly without programming.
Iterate on distribution mechanics for ChatGPT apps: Test name-based calls or contextual suggestions, using released docs to optimize recommendations.
Build industry-specific agents: Mirror Codex by creating tools for law or finance, starting with conversational prompts to handle workflows autonomously.
Experiment with revenue sharing in generative apps: Allow cameos in videos like Sora, tracking engagement to implement per-generation payments.
Foster AI intuition through onboarding: Create workflow content emphasizing productive use, monitoring adoption like Codex's days-to-weeks integration.

ONE-SENTENCE TAKEAWAY

AI's rapid evolution empowers zero-person companies and scientific breakthroughs, urging builders to iterate boldly for enduring advantages.

RECOMMENDATIONS

Embrace no-code agent tools to prototype ideas rapidly, accelerating validation and outpacing traditional development.
Release AI innovations early with safeguards to inoculate society against misuse, like unwatermarked deepfakes.
Focus on emergent features in products, letting user interactions reveal durable edges like memory in assistants.
Prioritize novel discovery capabilities in AI development, targeting small knowledge expansions as AGI harbingers.
Advocate for global testing frameworks on superintelligent models to balance progress with safety.
Build industry-tailored agents for niches like law or finance, enabling solo entrepreneurs to orchestrate startups conversationally.
Adapt monetization to personal AI uses, exploring per-generation fees alongside subscriptions for tools like Sora.
Bet on human adaptability during job transitions, preparing for meaning in post-AGI pursuits like space exploration.

MEMO

At OpenAI's DevDay 2025, CEO Sam Altman unveiled a suite of transformative tools, including Apps in ChatGPT and the Agent Builder, amid ChatGPT's milestone of 800 million weekly users. These launches position the platform as a burgeoning ecosystem for developers, where apps can integrate seamlessly, potentially revolutionizing distribution through innovative mechanics like contextual suggestions. Altman, reflecting on the no-code revolution, marveled at how visual interfaces now allow even non-coders to deploy sophisticated agents in minutes—uploading files, setting instructions, and connecting data sources. This shift, he noted, could exponentially increase global software creation, compressing idea-testing timelines and challenging builders to ideate faster than ever.

The conversation delved into the tantalizing prospect of zero-person companies, with informal betting pools speculating years until AI agents helm billion-dollar ventures autonomously. Altman highlighted Codex's leap to day-long tasks, predicting week-long operations soon through smarter models and enhanced memory, a pace that feels "disappointingly fast" even to him. Looking back, he expressed envy for today's young dropouts, armed with AI's vast opportunity space—far beyond what a 20-year-old version of himself could have built post-Stanford. Unique advantages, he advised, emerge iteratively: Tactics like ChatGPT's memory feature evolve into strategies, underscoring the wisdom in starting small and adapting.

AGI's horizon sharpened in Altman's view, defined not just by economic outperformance but by novel discoveries now flickering on platforms like Twitter—small breakthroughs in mathematics and physics signaling knowledge expansion. On benchmarks like GDPval, where GPT-5 trailed Claude Opus, he praised competitors' enterprise finesse, vowing it inspires rather than derails OpenAI's path. Yet societal adaptation remains key: Just as the Turing test whooshed by with minimal fuss, AI-driven science will normalize quickly, much like self-driving cars felt "weird only once." The "workslop" phenomenon—polished but rework-heavy outputs costing firms $186 monthly per employee—predates AI, Altman observed, but the economy favors those wielding tools productively.

Viral Sora deepfakes of Altman himself tested these waters, from GPU heist memes to global scrolls that lost their strangeness in minutes. Releasing such tech early with guardrails, he argued, inoculates against inevitable open-source fakes, though watermark removers pose branding risks. Exploring revenue sharing for cameos could sustain personal uses, like meme-sharing in group chats, potentially blending pay-per-generation with ads if Sora evolves into a feed. Jobs, too, face upheaval: A billion knowledge roles may morph first, echoing how farmers once dismissed desk work as unreal, yet human drives promise meaning in uncharted realms like space or brain-computer interfaces.

Altman's vision eschews an "everything app" like WeChat, favoring ChatGPT as a trusted AI super-assistant with separate features to avoid discord. Exciting agents loom in sectors like law and finance, Codex-style, enabling one-person startups via talk. Voice interfaces, he teased, could revive smart speakers for seamless interaction, though not universally. Patience is urged for hardware like io, aiming for a high-quality, scale-redefining device. Ultimately, a global safety framework—starting with rigorous testing—looms essential for superintelligent edges, ensuring AI's canvas of possibility unfolds in every direction without catastrophe.