Sam Altman thinks the chatbot era is already over. The OpenAI CEO has laid out a vision for what comes next — and it centers on persistent AI agents that don’t just answer questions but actually show up to work, manage complex tasks independently, and keep context even when no human is watching.
Summary
Key takeaways
- Sam Altman describes a “third wave” of AI built around autonomous, persistent agents that function as digital colleagues alongside human teams.
- The three waves: predictive AI, generative AI (the ChatGPT era), and now agentic AI that manages multi-step workflows without continuous human input.
- Salesforce CEO Marc Benioff targeted deploying one billion Agentforce agents by the end of 2025, signaling major enterprise commitment to agentic AI.
- OpenAI and Paradigm released EVMbench in February 2026 to benchmark AI capability in securing crypto smart contracts.
- For the crypto market, AI agents handling wallets, trades, and DeFi interactions will require robust blockchain-based identity and audit infrastructure.
Sam Altman’s Vision of Persistent AI Agents
Altman frames this as a moment of genuine architectural shift — not an incremental product update. Where previous AI breakthroughs changed what the technology could produce, this one changes what it can do on its own.
Defining the Three Waves of AI
The progression is surprisingly clean. The first wave of AI was predictive — recommendation engines, spam filters, systems that pattern-matched on data to anticipate outcomes. The second wave was generative: the ChatGPT era of producing text, images, and code on demand, responding to whatever prompt a user typed.
The third wave is different in kind, not just degree. Agentic AI doesn’t wait for a prompt. It takes initiative, executes multi-step processes, and manages workflows without constant human direction. That shift from reactive to proactive is what makes this wave structurally distinct from what came before.
Autonomous Persistent AI as Digital Colleagues
The word “persistent” carries real weight here. Altman isn’t describing an AI that helps you draft an email and then goes idle. He’s describing systems that retain context across sessions — they don’t reset when you close your laptop. They pick up where they left off, independently, across complex multi-step tasks that might unfold over hours or days.
The framing Altman uses is deliberately human: staff, co-workers, colleagues. That’s not just marketing language. It implies accountability structures, division of labor, and a relationship between AI agents and human teams that looks much more like collaboration than tool use.
Enterprise Deployment and Market Impact
The gap between vision and reality in AI has historically been wide. But the enterprise commitments forming around agentic AI suggest this wave is already hitting the ground.
Salesforce’s Ambitious AI Agent Rollout
Salesforce CEO Marc Benioff set a target of deploying one billion Agentforce agents by the end of 2025. That number is striking not just for its scale, but for what it signals about how enterprise software companies are repositioning themselves. Salesforce committed to integrating persistent agents into its product roadmap at significant scale.
That commitment also creates competitive pressure. OpenAI’s push into enterprise workflow automation, including tools that let companies set guardrails and policies around how AI agents access company data, has already rattled software stocks. According to reporting by Business Insider, shares of Workday, Atlassian, HubSpot, Salesforce, and Okta all fell after OpenAI’s recent enterprise product announcements, with TD Cowen analysts citing OpenAI’s moves as a “major reason” for a 3% sell-off in the IGV software index.
Productivity Gains as Key Market Signal
There’s an important distinction between companies announcing AI agent deployments and companies actually reporting productivity improvements as a result. The real confirmation that the third wave has arrived won’t come from press releases — it will show up in earnings calls, when finance teams start attributing measurable output gains to agent-driven automation. Until that reporting emerges at scale, the enterprise adoption story remains more thesis than fact.
That makes enterprise adoption metrics the most honest signal for anyone tracking where this technology is actually landing versus where it is being announced.
Crypto Security and Infrastructure Developments
Launch of EVMbench for AI-Powered Smart Contract Security
OpenAI partnered with Paradigm to release EVMbench in February 2026 — a benchmark specifically designed to assess how well AI performs at securing crypto smart contracts. Smart contract vulnerabilities have been one of crypto’s most consistent and costly problems, responsible for billions of dollars in losses from exploits and bugs over the years.
EVMbench is a direct response to that exposure. By creating a standardized way to evaluate AI’s effectiveness in this domain, OpenAI and Paradigm are establishing a baseline for what “good” looks like before agents get deployed at scale in high-stakes financial environments. The crypto angle here is notably practical — it’s about fixing known infrastructure problems, not launching speculative tokens.
Blockchain-Based Identity and Audit for AI Agents
The prospect of AI agents autonomously managing crypto wallets, executing trades, or interacting with DeFi protocols raises a set of questions that don’t have clean answers yet. Who authorized this transaction? Which agent executed it? Can the action be audited?
Blockchain-based identity and accountability mechanisms could provide exactly the kind of verifiable trail that autonomous agents would require in financial settings. The technical infrastructure for this doesn’t fully exist at scale yet, but the demand for it will grow in direct proportion to how broadly persistent agents get deployed in crypto contexts.
Investment Implications and Strategic Opportunities
Shift in Company Approaches to Productivity and Workforce
Altman’s framing of AI agents as colleagues rather than tools implies a quiet but significant restructuring of how companies think about headcount and operational leverage. If an agent can handle a sustained, complex workflow independently — not just assist with a task but own it — the calculus around staffing decisions starts to change.
That shift won’t happen uniformly or overnight. But companies that integrate agentic systems effectively will likely generate measurable output advantages over those that don’t, which is exactly why enterprise adoption timelines are worth watching closely.
Infrastructure, Security, and Compliance as Durable Investment Themes
The more analytically interesting question for investors isn’t which AI provider builds the best agent — it’s what every agent deployment will need regardless of who wins that race. Identity verification, authorization frameworks, audit trails, and security tooling are table stakes for any enterprise deploying agents at scale. Those picks-and-shovels layers — the infrastructure underneath the agents rather than the agents themselves — represent a more durable thesis precisely because they’re agnostic to which model or platform dominates.
EVMbench is one early example of that kind of foundational work. It won’t generate headlines the way a new model launch does, but it addresses a structural gap that has to be filled before agentic AI in crypto can function responsibly at any meaningful scale.
FAQ
What are persistent AI agents according to Sam Altman?
Persistent AI agents are autonomous AI systems that act as digital colleagues, managing workflows independently without losing context after a user disengages. Unlike a chatbot that resets between sessions, they retain memory and continue tasks autonomously over time.
How does the third wave of AI differ from the first two waves?
The first wave was predictive AI — recommendation algorithms and spam filters. The second was generative AI, the era of producing text, images, and code on demand. The third wave is agentic AI that takes initiative, executes multi-step processes, and manages workflows without continuous human oversight.
What was the significance of Salesforce’s goal to deploy one billion AI agents?
Salesforce CEO Marc Benioff targeted deploying one billion Agentforce agents by the end of 2025, demonstrating major enterprise commitment to integrating persistent AI agents at scale, aimed at boosting productivity and operational leverage across the company’s customer base.
What role does EVMbench play in crypto security?
EVMbench, released by OpenAI and Paradigm in February 2026, benchmarks AI’s ability to secure crypto smart contracts. It establishes a standardized baseline for evaluating AI performance against one of crypto’s most persistent vulnerabilities — the smart contract exploits and bugs responsible for billions in annual losses.
Article produced with the assistance of artificial intelligence and reviewed by the editorial team.
