The Age of the AI Agent Has Arrived

Because we're a Google partner, I flew out last month for Google Cloud Next '26. I've been to my fair share of tech conferences, but this one genuinely startled me — 30,000 people crammed into what was basically a stadium-sized venue. Just getting to the keynotes felt like being on a trip to Japan, racking up 20,000 steps a day; I told my colleagues I might as well have been hiking. The big tech expos we usually talk about, like CES, are typically massive assemblages of dozens of vendors piled together. This was a single vendor pulling off that scale on its own — genuinely eye-opening.

The Agent Era Officially Arrives — Engineers Become Agent Managers

Google Cloud has long been the market's laggard, with Amazon's AWS holding the top spot for years. But over the past year or so, thanks to Gemini's string of wins, Google's power as a fast-rising challenger has been genuinely startling — and not just because of the Gemini model itself, but because everything gets bundled together with the cloud. Put simply: buy Google's cloud, and you get a whole integrated package of best-in-class AI vendors thrown in. So the single most central line from this conference was CEO Thomas Kurian's direct declaration: "The era of the agentic enterprise has begun." The AI agent everyone was talking about a year ago has now entered large-scale enterprise deployment.

The potential of agentic AI was probably what struck me most at this conference. There were a huge number of use cases on display, but the interesting part wasn't the technology — it was the displacement effect. A year or so ago, everyone was talking about the "copilot" development logic (engineer writes the code, AI rides shotgun) — and that already feels like a previous era. The dominant framework now is agent-first: the engineer's role has shifted to "manager of AI agents." I think this shift is bigger than it looks on the surface, because the copilot logic still assumes tools executed by a human, whereas the agent logic has AI executing directly on its own. It's basically forcing you to relearn, this year, the exact workflow you only just upgraded to last year (though thankfully that's not really a problem, because the big players will make sure you learn it whether you try to or not).

But Google has made a choice here that I think differs meaningfully from its rivals. Anthropic, this past year, has given off the sense that it wants to strip humans out of the process wherever it's technically possible; Google Enterprise's agent philosophy, by contrast, is to keep humans in the loop and outsource just the tedious parts of the process to AI. To be fair, neither company puts it quite this explicitly, but you can feel the difference clearly once you look at the direction each one is building toward.

So I understand why the developer community loves Anthropic so much — that "one person plus one AI does the work of ten" energy really does have impact.

But when it comes time for enterprises to actually open their wallets, they may well lean toward Google's human-machine collaboration version, because the reality of enterprise settings involves compliance requirements, plenty of work that has to keep a human physically in the loop, and multiple levels of sign-off — full automation just isn't realistic. Take the Capcom case study announced at the conference, for example: they're using AI agents to run game bug testing, processing 30,000 hours of testing work every month.

Because AI agents are starting to work their way into the enterprise, a lot of people have been asking me lately: now that AI is here, can't I just write my own program instead of buying SaaS? My view is a bit different: yes, the very shallow layer of work really will get displaced — the kind of task you used to outsource, that now takes five minutes with AI. But right behind it comes a deeper set of problems — data governance, cross-system integration, security compliance, model management — the kind of thing you genuinely don't want to be handling yourself. So in practice, real decentralization is very hard to pull off. What does disappear, though, is the old style of middleman company — the kind that never provided any real depth of value. What's left standing is the set of players even more deeply tied to the major platforms.

Google's announcement of its 8th-generation TPU at this conference follows the same logic — running agents demands enormous compute and an enormous data foundation, which by definition means only a handful of the biggest players can afford to play. The software industry's bifurcation was on full display at the conference too. I sat in on an Adobe talk, and frankly, that was a company adapting poorly to the new game — their roadmap has been treading water, and it looks disadvantaged next to both OpenAI's image generation and Google's own Nano Banana.

Google Completes Its Empire Strikes Back — the Agent Wars Are Fully On

If you ask me what left the deepest impression at this conference, I'd say it's just how large America's lead in AI genuinely is — backed by a solid, real product lineup, a clear technical roadmap, and a full cadence of commercialization. Watching the whole event, you get the sense that Google, powered by its enormous revenue base, has spent years clawing its way back from behind — in AI and in cloud alike — and it's pulling off a genuine empire-strikes-back, one that's nearly complete. And the other major players are in the same boat, each with an extremely complete strategic footprint of their own. It's a genuine heavyweight championship bout, and being in the room for it felt exhilarating.

Coming back to the investment angle, I think agentic AI will be the biggest narrative of the next year in the short term — essentially at least two of the three major model makers have already firmly established themselves in agentic AI, and on my flight back to Taiwan I saw that OpenAI is pushing hard too, with a new model release that's performing well. All three majors are going to keep slugging it out.

That said, Google's positioning going into the AI agent era looks very strong out of the gate, and right now there's no obvious force capable of slowing Google's advance. As I write this, that reality is already showing up in Google's stock price.