OpenAI: AI Agents to Dominate Internet Actions

Tibo Sottiaux of OpenAI asserts that the majority of internet actions will soon be taken by AI agents, necessitating that product developers build for this new level of agent scale.

The shift represents a move from simple task automation to AI agents becoming the primary drivers of interaction on the internet, transforming how products need to function and integrate.

Sottiaux describes a cyclical development process where he initially builds larger teams of agents when pushing technical frontiers and exploring new capabilities.

However, significant model breakthroughs frequently allow a single, more capable agent to perform tasks previously handled by complex multi-agent setups.

This dynamic leads to periods of shrinking agent teams as new models consolidate capabilities, only to expand again with the next set of challenges.

This future state would see agents operating as background utilities, capable of tasks like attending meetings or managing notes without requiring constant device interaction from the user.

The ultimate goal is to liberate users from being tethered to physical hardware, allowing technology to actively work on their behalf rather than as a reactive tool.

Tibo Sottiaux anticipates that the nature of work will continue to undergo radical transformation through AI integration, emphasizing the ongoing evolution of human-AI collaboration.

While current voice and multimodal inputs are becoming more integrated, they still feel cumbersome, but the goal is to achieve interactions that are as natural as human conversation.

The future envisions shared digital whiteboards and other surfaces that facilitate seamless human-agent collaboration, moving beyond the limitations of current chat-based AI to a more intuitive experience.

The aim is to ship all the advanced capabilities of 'Dots' into ChatGPT, making them available to 1.2 billion users through a simplified, zero-configuration model.

This consolidation seeks to reduce product complexity and provide a more unified, efficient user experience without requiring intricate setups.

This vision underscores the drive to create intuitive, conversational AI systems that integrate effortlessly into daily life.

Initially, OpenAI's 'Dot' product focuses on a single primary agent that learns user preferences and behaviors.

Future updates will enable users to create and manage multiple 'Dots,' each assigned specific roles such as monitoring social media platforms like Twitter.

This expansion will allow users to build a virtual team of AI agents, empowering them to manage complex workflows and tasks that require dedicated, specialized attention.

This initiative aims to provide developers with the opportunity to ship their products to a vast user base of 1.2 billion ChatGPT users.

A revenue-sharing model will be implemented for popular plugins that demonstrate high usage and consistent user retention.

Developers are encouraged to focus on building high-quality plugins that effectively engage and retain users to qualify for the revenue-sharing program.

ChatGPT's memory feature, a result of this research, enables it to recall specific personal details and past events, enhancing its ability to provide personalized assistance.

Dot was launched with a strong emphasis on providing 24/7 productivity and ensuring user safety.

The project utilizes the Astra model, which the development team identifies as their safest and most aligned model to date.

This approach seeks to empower users, fostering creativity and personal expression through AI, rather than replacing human ingenuity.

While traditional roles, such as coding, may evolve, Sottiaux believes the demand for builders and innovators will only increase in the AI era.

He also posits that human interest in peer-to-peer connections will remain a constant, unaffected by advancements in AI technology.

The AI industry must avoid the pitfall of merely increasing the number of agents or tasks, instead focusing on AI's potential to improve human experience.

Sam Altman, CEO of OpenAI, has emphasized that AI's core promise lies in reducing noise and allowing individuals to allocate their attention more intentionally.

The ultimate goal is to optimize human attention, enabling better rest and increased productivity by filtering out distractions and streamlining workflows.

Conversely, skills such as having good taste, empathy for the user, and the ability to connect with the target audience are becoming increasingly crucial.

OpenAI highly values employees with prior startup founder experience, currently employing over 120 individuals who fit this profile, recognizing their ability to define and build successful products.

Product Managers are thriving in this environment due to their inherent ability to iterate rapidly and articulate what constitutes a 'good' product experience.

Sottiaux points to Ahmed Ibrahim, an engineer who started as a new graduate, as an exemplary model of a successful developer at OpenAI.

Key traits for career progression and growth include being highly collaborative and demonstrating a selfless approach to problem-solving, prioritizing collective goals over individual recognition.

The 'sponge' mentality, characterized by an accelerated pace of learning and eagerness to absorb new information, is identified as a critical factor for success in the rapidly evolving AI landscape.

OpenAI operates with an 'incredible bottoms-up energy,' where development is driven by grassroots collaboration rather than rigid top-down structures.

Projects like the Decisions API often originate from small, informal Slack channels and weekend hacking sessions, reflecting the company's entrepreneurial spirit.

While fostering this innovative environment, OpenAI maintains a high quality bar, sometimes delaying launches if projects require further refinement or 'baking.'

The organizational culture empowers individuals to initiate ideas, take ownership, and drive projects from conception to the shipping stage.

AI models are experiencing rapid advancements, becoming significantly faster and more cost-effective.

This progress will enable seamless integration of all modalities, fundamentally altering how users interact with technology.

Companies should anticipate that AI capabilities will improve tenfold within a year and design products accordingly, assuming a future dominated by AI agents.

Notion's implementation of multi-channel processing (MCP) led to massive traffic spikes, illustrating the economic strain AI agents can place on existing systems.

The increased traffic generated by AI agents creates new challenges for product infrastructure and demands reconsideration of economic models.

A critical product design challenge emerges in building interfaces that effectively satisfy both human users and the unique interaction patterns of AI agents.

Significant compute resources are specifically allocated to safety measures and guardrails, highlighting a commitment beyond mere capability expansion.

In certain instances, OpenAI has chosen not to release advanced models to the public, prioritizing safety and hardening its systems to prevent potential misuse or unintended consequences.

Tibo Sottiaux acknowledges that the current process of selecting AI models is overly complex, leading to user fatigue and requiring an almost academic understanding of various options.

He personally expresses frustration with the existing 'model picker' workflow, which often demands significant reasoning efforts to choose between multi-agent or ultra models.

OpenAI's ultimate goal is to simplify the application experience to such an extent that the interface effectively disappears, making AI interaction intuitive and effortless for all users.