The big AI labs have spent June converging on the same idea: agents that take actions rather than chatbots that answer questions. The launches arriving across the sector all point the same way, toward software that can plan and execute multi step tasks across applications without a person switching between tools.
What the labs shipped
Google has moved fastest in public. Gemini 3.5 Flash went generally available at Google I/O on May 19, already the default in the Gemini app and in AI Mode in Search, priced at 1.50 dollars and 9.00 dollars per million input and output tokens and running at roughly four times the speed of the previous generation. Google has also been positioning a Gemini 3.5 Pro for June and pushing personal agents such as Project Mariner, which can handle multiple concurrent tasks on cloud based virtual machines.
OpenAI and Anthropic answer
OpenAI's Operator agent has been reported to reach high success rates on complex browser tasks, while Anthropic has focused on making its Claude agents safe for enterprises, adding privacy infrastructure such as private network tunnels and self hosted code execution sandboxes to its managed agents platform. Microsoft and Google have meanwhile pushed directly into AI coding models, a segment Anthropic and OpenAI had largely defined.
The interoperability glue
Underpinning the race is a quiet standardisation. The major players have adopted shared protocols, including the Model Context Protocol, that let agents plug into tools and data across platforms, with tens of thousands of servers and tens of millions of monthly SDK downloads reported. That interoperability is what turns a single clever demo into something a business can wire into its existing systems.
Why it lands on the labour market
For workers, the agent shift matters more than any single model release. An agent that can complete a workflow end to end is a closer substitute for entry level operational work than a chatbot ever was, which is precisely why firms keep pairing these launches with slower junior hiring. The flip side, as ever, is a growing need for people to design, monitor, correct and govern the agents now being switched on across the economy.
