A few years ago, generative AI was hardly discussed outside of a few research labs. The discussion has already progressed at this point. Agentic AI, or software that does tasks for you, sometimes without your request, is the new obsession that tech executives discuss at dinners and on earnings calls. This is an odd change that is occurring more quickly than most boardrooms can handle.
Sinan Aral, an MIT Sloan professor who keeps a close eye on this area, puts it plainly. He claims that the agentic age will not arrive. It is present. Agents are already in place throughout the economy, silently performing tasks that previously required a human to click through screens. If you walk into a mid-sized business today, you might see the finance team getting coffee while an AI agent reconciles invoices in a back office. It’s the kind of detail where the change is truly taking place, but it doesn’t make the news.
| Topic Snapshot | Details |
|---|---|
| Subject | Agentic AI |
| Primary Source | MIT Sloan School of Management |
| Featured Expert | Sinan Aral, Professor of Management, IT, and Marketing |
| Co-Researchers | John Horton, Kate Kellogg |
| Industry Adoption (2023) | 35% of surveyed firms |
| Planned Adoption | 44% additional firms |
| Major Vendors | Microsoft, Salesforce, Google, IBM |
| Notable Voice | Jensen Huang, Nvidia CEO |
| Projected Market Impact | “Multi-trillion-dollar opportunity” |
| Realistic Maturity Timeline | Possibly 10 years (per Andrej Karpathy) |
| Core Capabilities | Perceive, reason, act, transact |
| Risk Areas | Data quality, governance, trust, security |
Speaking with employees at big companies gives me the impression that no one really knows what they’ve committed to. According to a survey conducted in the spring of 2025 by BCG and the MIT Sloan Management Review, about one-third of participants had already used AI agents, and another 44% intended to do so shortly. The adoption curve is quite steep. However, according to Aral, society’s understanding of agentic AI is still developing and occasionally nonexistent. Businesses are deploying first and then asking the more difficult questions, which is rarely how this kind of situation turns out.
On stage at CES 2025, Jensen Huang referred to enterprise AI agents as a multi-trillion-dollar opportunity. This is the kind of statement that appeals to shareholders but causes some discomfort for regulators. Many businesses will eventually use these systems whether they intended to or not because Microsoft, Salesforce, Google, and IBM are incorporating agentic features straight into their platforms. It’s important to recognize this subtle manifestation of inevitability.
What precisely is an agentic AI, then? To be honest, there isn’t yet a clear definition. According to John Horton of MIT Sloan, it is autonomous software that can use tools, transfer money, and negotiate with other agents while perceiving, reasoning, and acting in digital environments on behalf of a human principal. These systems extend large language models by carrying out multi-step plans inside actual workflows, according to Kate Kellogg and her co-authors. More like a coworker who never logs off than a chatbot.
Nevertheless, there is a significant discrepancy between the pitch deck and reality. OpenAI cofounder Andrej Karpathy, who is notorious for being modest, recently hinted that it might take ten years before agents function well. The circular nature of the transactions supporting the AI economy—the same vendors investing in the same customers purchasing the same goods—has drawn criticism from market analysts. It has a slight rhyme with earlier tech cycles.
It’s difficult to ignore the pattern. The technology is real, the productivity gains are real, and there are real, mostly unaddressed risks related to data security, governance, and quality. Every organization needs an agentic strategy, according to Aral, but only if it is based on a thorough evaluation of the advantages and risks. You get the impression that most businesses will omit that step as you watch this play out.
Perhaps the workforce of the next ten years will be agentic AI. Perhaps its own complexity causes it to stall. In any case, the offices have already undergone a subtle transformation, and those who work there are still figuring out what that means.
