Less than a year ago, I posted this on Threads:

"I see agents as bots that are constantly working in the background 24/7 without an app open or human intervention. Are there any use cases of this yet? There seems to be zero. Everything still relies on a human trigger or a human in the loop or the app has to be open."

That was May 2025. I was genuinely asking, and the honest answer at the time was no, not really. Everything marketed as an "agent" still needed you to open an app, type a prompt, or click a button. The closest thing to autonomy was a calendar tool that could suggest meeting times. A few months earlier, in March 2025, I had asked a slightly different version of the same question: "There are agents everywhere now, but how much of your workflow is actually being automated by agents today?" The answer then was also basically nothing. Now it's 2026, and the landscape looks completely different.

From talking about agents to using them

2025 was the year agents went from being a concept people debated on social media to something that actually shipped. As TechXplore put it, "systems once confined to research labs and prototypes began to appear as everyday tools." The term "agent" had been floating around AI research for decades, but 2025 was the first time it meant something concrete to developers and consumers. A big part of what accelerated this was the Model Context Protocol (MCP). In early 2025, we were talking more about MCP and agents than actually building with them. MCP promised a universal way for AI models to connect to external tools and data sources. By the end of 2025, that promise had largely been delivered, with major AI companies shipping production-ready agent SDKs built around it. The shift from "interesting demo" to "thing I actually use" happened faster than most people expected.

The numbers tell the story

The AI agent market was valued at roughly $8 billion in 2025. Analysts project it will reach $52 billion by 2030, growing at a compounding rate of over 46% per year. That's not steady, incremental growth. That's an explosion. Gartner predicts that 40% of enterprise applications will embed task-specific AI agents by the end of 2026, up from under 5% in 2025. Goldman Sachs called 2025 the year of "the biggest changes" their CIO had seen in 40 years of technology, and predicted 2026 would be even bigger. But the numbers also reveal a gap. According to Deloitte's 2026 State of AI in the Enterprise report, 75% of companies plan to invest in agentic AI, but only 11% have agents running in production. The ambition is there. The execution is still catching up.

What changed, practically

A year ago, I defined agents as bots working autonomously in the background, 24/7, without a human trigger. That bar felt impossibly high at the time. Today, that's just... a feature. Notion has agents that trigger when a page is created and draft content automatically. GitHub has agents that review pull requests and suggest fixes. Scheduling tools now do more than suggest times, they negotiate across calendars, reschedule conflicts, and send follow-ups without you lifting a finger. The pattern that emerged wasn't one giant leap. It was a steady accumulation of small capabilities:

Each of these on its own was incremental. Together, they crossed a threshold.

The gap between hype and production

Not everything is rosy. The "only 11% in production" stat from Deloitte is telling. Many organizations spun up agent pilots that never scaled. The reasons are familiar: unclear ownership, no integration with existing workflows, and a tendency to build impressive demos that fall apart at the edges. Kore.ai's analysis put it well: "most agent initiatives were never designed to scale." The technology works. The problem is organizational, not technical. Companies that succeeded treated agents as infrastructure, not experiments. They embedded them into existing tools and processes rather than building standalone "AI products" that nobody asked for.

What this means going forward

The speed of this transition is what surprises me most. In March 2025, agents were a talking point. By March 2026, they are a line item in enterprise budgets and a feature in consumer products. If there's a takeaway, it's this: the window between "we're talking about this technology" and "this technology is reshaping how we work" has compressed dramatically. It used to take years for enterprise software trends to move from conference talks to production deployments. With agents, it took about twelve months. The next question isn't whether agents will be widely used. They already are. The question is which workflows will be the last to give up human control, and whether the 89% of companies still in the pilot phase will catch up before the early movers pull too far ahead.

References