OpenAI's annual developer conference, DevDay, has become a bellwether for where the AI industry is heading. This year, CEO Sam Altman used his time on stage not just to unveil new tools but to offer candid, sometimes counterintuitive, perspectives on the technology he helped popularize. From calling a core metric "terrible" to sharing how he personally uses AI agents, Altman's remarks painted a picture of a leader grappling with the real-world consequences of his company's creations.
Below are the five most surprising takeaways from his appearance, each with context on why it matters for developers, businesses, and anyone trying to make sense of the AI boom.
1. Altman Called Tokens a "Terrible" Metric
In an industry obsessed with benchmarks, Altman's dismissal of tokens as a measure of progress stood out. "Tokens are a terrible metric," he said, arguing that counting the units of text an AI processes says little about whether the model is actually useful. It's a surprising admission from the CEO of the company that made tokens a household term among developers.
Why does this matter? For years, AI vendors have touted token limits and processing speeds as key differentiators. But Altman's comment suggests a shift toward evaluating AI by outcomes: did it solve the problem? Did it save time? Did it delight the user? For businesses building on OpenAI's APIs, this is a nudge to focus on end results rather than technical specs. Outcome-based evaluation could become the new standard, pushing developers to think less about how many tokens a model consumes and more about the value it delivers.
It also hints at a maturing market. As AI models become commoditized, the differentiator won't be raw scale but how well they integrate into workflows and deliver tangible benefits. Altman's comment is a signal that the era of token-count bragging rights may be ending.
2. He Shared How He Uses Dots Agents Personally
Altman is known for being an avid user of his own products, and he offered a rare glimpse into his daily routine with Dots, OpenAI's agent framework. He described using Dots agents to handle repetitive tasks, from scheduling to information gathering, and emphasized that these agents are not just for techies. "I use them to free up mental space," he said, noting that even small automations can compound into significant time savings.
This is surprising because executives rarely detail their personal workflows, and it underscores a broader trend: AI agents are moving from experimental to essential. For knowledge workers, Altman's example is a practical nudge to explore agent-based tools. Start small, perhaps by automating email sorting or meeting notes, and gradually expand. The takeaway is that personal AI agents are not a distant future; they are already here, and even the busiest CEOs rely on them.
Moreover, Altman's openness about his usage patterns humanizes the technology. It suggests that the barrier to entry is lower than many assume. You don't need a PhD in machine learning to benefit from agents; you just need a willingness to experiment.
3. His Take on the Job Market Was Surprisingly Nuanced
When asked about AI's impact on employment, Altman avoided the usual platitudes. He acknowledged that some jobs will be displaced but argued that the net effect could be positive if society manages the transition well. He stressed the importance of reskilling and education, and he hinted that new categories of work will emerge that we can't yet imagine.
This is a departure from the binary narratives that often dominate the debate. Altman didn't promise a utopia, nor did he predict mass unemployment. Instead, he framed it as a challenge of adaptation. For business leaders, this means investing in workforce development now. For policymakers, it's a call to modernize education systems. And for individuals, it's a reminder to cultivate skills that complement AI, such as critical thinking, creativity, and emotional intelligence.
The surprising part? Altman's tone was more cautious than bullish. He didn't claim to have all the answers, which is rare for a tech CEO. That humility might be the most valuable takeaway for anyone planning a career in an AI-shaped economy.
4. He Hinted at a Shift from Models to Platforms
While Altman didn't announce a specific product, his comments about the future of OpenAI suggested a strategic pivot. He talked less about the next GPT model and more about building an ecosystem where developers can create and deploy AI solutions seamlessly. This includes better tools for fine-tuning, improved safety features, and more robust APIs.
For developers, this is a big deal. It means OpenAI is positioning itself not just as a model provider but as a platform. The implication is that the company's long-term value lies in enabling others to build on its technology, much like how Apple's App Store created a thriving economy. If Altman's vision pans out, the most successful AI applications may not come from OpenAI itself but from the developers who leverage its platform.
This shift also has competitive implications. By focusing on platform capabilities, OpenAI is differentiating itself from rivals who compete primarily on model performance. It's a bet that ecosystem lock-in will matter more than benchmark scores. For businesses choosing an AI vendor, that's a factor worth weighing.
5. He Admitted OpenAI Doesn't Have All the Answers on Safety
Safety is always a hot topic, and Altman's remarks were notably candid. He acknowledged that OpenAI is still learning and that some risks are not fully understood. He emphasized the need for collaboration across industry, academia, and government, and he warned against complacency.
This is surprising because tech leaders often project confidence, especially about their own safety efforts. Altman's admission that the field is still in its early days is a reality check. It suggests that businesses deploying AI should not assume that safety is a solved problem. Instead, they should implement their own guardrails, conduct regular audits, and stay informed about best practices.
For developers, this means building with safety in mind from the start, not as an afterthought. For users, it's a reminder to be discerning about the AI tools they adopt. And for regulators, it's a signal that industry is open to sensible oversight, which could shape future legislation.
Frequently Asked Questions
What did Sam Altman mean when he said tokens are a terrible metric?
Altman was criticizing the industry's tendency to measure AI progress by token counts, which indicate how much text a model can process. He believes this metric doesn't reflect real-world usefulness. Instead, he advocates for evaluating AI based on outcomes, such as whether it solves a problem or improves productivity.
How does Sam Altman use Dots agents in his daily work?
Altman shared that he uses Dots agents to automate repetitive tasks like scheduling and information gathering. He emphasized that these agents help free up mental space and that even small automations can lead to significant time savings. His example encourages others to experiment with agent-based tools for personal productivity.
What did Altman say about AI's impact on jobs at DevDay 2026?
He offered a nuanced view, acknowledging that some jobs will be displaced but arguing that new roles will emerge. He stressed the importance of reskilling and education, and he called for a collaborative approach to managing the transition. His tone was cautious rather than utopian or alarmist.
Is OpenAI shifting its focus from models to platforms?
Yes, Altman's comments suggested a strategic move toward building a platform ecosystem. He highlighted tools for developers, such as better fine-tuning and safety features, indicating that OpenAI's future value lies in enabling others to build on its technology, similar to how app stores foster innovation.
What did Altman say about AI safety?
He admitted that OpenAI doesn't have all the answers and that some risks are still not fully understood. He emphasized the need for cross-sector collaboration and warned against complacency. His candor suggests that businesses should implement their own safety measures rather than relying solely on vendors.