Google unveils new AI agents at software development conference, could steal consumer AI crown from OpenAI

The amphitheatre complex where Google holds its annual software-developers’ conference has a cheesy, fairground feel.

RVs are parked on site. Employees whizz in on the tech company’s multi-coloured bicycles. There are stands and sideshows everywhere.

On stage, Sundar Pichai, its boss, tells a corny joke about Google’s overworked chips, known as TPUs, doing “teraflops into bed”.

The event is by no means as sleek as Apple’s developer jamboree held in June, which tries to retain some of the minimalist chic of the late Steve Jobs. But when it comes to artificial intelligence, Google long ago upstaged the maker of the iPhone (its models will power many of Apple’s AI features going forward).

It now looks as though it may steal the consumer AI crown from OpenAI, maker of ChatGPT, as well.

On May 19, Google unveiled a new line-up of AI agents powered by its latest Gemini 3.5 Flash model.

It included AI coders to rival those offered by OpenAI and Anthropic, but also agents designed to perform various tasks for regular folk going about their daily lives.

Some will appear in the Gemini app, used by 900 million people every month. Others will be embedded directly in Google Search, used by more than three billion. The company is, in sum, bringing agents to the masses.

As is common in Silicon Valley, the examples executives used on stage raised plenty of eyebrows. How often would someone need an AI agent to make a slide presentation for a bouncy-castle party?

Yet the tools unveiled also show promise.

An agent called Gemini Spark will be able to do things such as scan emails or organise group trips even after a user has closed their laptop or put down their phone, while “information agents” in Google Search will be able to keep tabs on sports tournaments, shopping sales or the stock market.

All this looks particularly troubling for OpenAI, which has so far led in consumer AI.

Shortly after Google launched its Gemini 3 family of models in November, Sam Altman, OpenAI’s boss, issued a “Code Red” emergency to galvanise employees into speeding up improvements to ChatGPT.

Since then, the focus at the lab has shifted towards its coding agent. But the release of Gemini 3.5 Flash, which Google says is four times faster than other frontier models, and the new suite of agents are likely to raise fresh questions about what OpenAI is doing with its flagship chatbot.

Investors are certainly bullish on Google’s prospects. The market value of Alphabet, its parent company, is now within a hair’s breadth of $5 trillion, having passed $4 trillion only in January.

Yet Google’s success in AI is also creating problems. According to Mr Pichai, the number of tokens — Silicon Valley’s favoured measure of AI usage — consumed by its services has risen to 3.2 quadrillion a month, up from 480 trillion a year ago.

Each token requires computing power, and therefore money, to generate, which is why Google’s capital expenditure this year will be up to $190 billion, six times as much as four years ago.

Moreover, that money does not go as far as it used to, because everything from chips to energy has become more expensive.

Even for Google, there are limits to how much it can afford to spend.

There are a few potential solutions. One is to lower the cost per token by making the technology more efficient, which Google will surely do.

Another is to put limits on AI usage. Such restrictions were not announced at the event, but Gemini subscribers were alerted afterwards that they would apply, albeit with higher limits than for non-subscribers, according to Richard Windsor of Radio Free Mobile, a research firm.

Usage caps may also encourage more people to pay for a subscription.

A third solution is to lean more into ads. Google reckons that the greater detail in AI queries will appeal to marketers. Although it has not yet incorporated ads into its Gemini app, it has been interspersing them into the AI responses its search service now spits out, and soon it will put AI product explainers alongside these ads.

Mr Pichai noted at the conference that some companies are “already blowing through their annual token budgets — and it’s only May”.

Consumers are not “tokenmaxxing” to anything like the same extent. But the more they use agents, the more providers of AI will need to come up with novel ways to make money from them.