Agents will settle AI's oldest debate

Welcome back to Building AI Agents, your biweekly guide to everything new in the AI agent field!

“For example, an AI agent could be used to plan a trip for a user, selecting and booking a flight and hotel, and coming up with an itinerary.” —every intro to AI agents article ever

In today’s issue…

• Agents are the final showdown of the AGI debate

• OpenAI embraces its competitor’s standard

• Google’s state-of-the-art Gemini 2.5 model

• AGNTCY’s blueprints for the new agentic internet

• Automating 90% of finance and law with agents

…and more

🔍 SPOTLIGHT

Depending on who you ask, AI’s impact could be utterly transformative—or an absolute nothingburger. With AI agents, we’re about to find out.

For decades, a small but vocal group of futurists, technologists, and science fiction writers have claimed that accelerating progress in AI would lead to artificial general intelligence (AGI)—systems capable of truly humanlike reason and problem-solving. Such intelligence would almost certainly be swiftly transformative to the economy and society. Although AGI was long seen as far off at best, and a sci-fi fantasy at worst, the recent breakneck progress in AI has led major tech figures, such as Sam Altman, Demis Hassabis, Jensen Huang, and more, to predict that AGI is not only possible, but imminent. At the other end of the spectrum, AI critics such as researcher Gary Marcus see the field as a bubble driven by marketing hype from these CEOs and others, and sustained by the intense FOMO of tech companies and their investors.

Recent events can be seen as a sort of Rorschach test for those with these conflicting views. The Magnificent Seven, an AI-heavy group of the world’s largest tech companies, now make up nearly a third of the total capitalization of the S&P 500, and AI headlines can massively swing the entire market—the launch of DeepSeek in January erased nearly a trillion dollars of value. Developments in AI, such as Sam Altman’s temporary ouster in 2023, are treated as headline news. Spending on tech companies’ enormous AI datacenter buildouts made up between 16% and 20% of real US GDP growth in the 3rd quarter of 2024, and are on track to surpass the percentage seen during the dot-com bubble.

For AI optimists, these trends are consistent with the gathering acceleration that precedes the arrival of AGI; for skeptics, they’re proof that it’s 1999 all over again and the bubble is soon to pop.

Until now, both of these narratives were unfalsifiable. Since no AI existed which could be said to have any kind of general intelligence, prophets of AGI could always point to some hypothetical future system. Their opponents, conversely, could claim that AGI was like fusion power: 30 years away—and always will be.

With the arrival of LLMs, this debate is finally coming to a head. Although it is fiercely argued whether they can truly reason, LLMs seem to be a form of proto-AGI, capable of a wide range of tasks and possessing some kind of internal model of the world. With the addition of tools and memory, AI agents represent the first systems we have which can be argued to act as though they have general intelligence.

The tech giants certainly should hope so, given the scale of their investments into the field—the eleven-figure sums being shelled out for datacenters are unsustainable in the absence of massive returns on investment, the likes of which only something like AGI could deliver. Unlike prior decades, in which the debate over the promise of generally intelligent AI was purely academic, now the stakes are real. Either LLMs will evolve into AGI, or the resulting crash will create an AI winter deeper than any we’ve seen before.

Thanks to agentic AI, we won’t have to wait much longer to see.