Google's Antigravity team introduced Custom Agents on 12 August with first-class support in Antigravity 2.0 and the Antigravity CLI, with the IDE following shortly. The premise stated in the announcement is that software engineering has shifted from writing lines of code to orchestrating agents, and that the productivity gain comes from division of labour: breaking a complex project into specialised agents that can act, verify, and run tasks in parallel.
The mechanism is configuration rather than a new model. A custom agent is defined in seconds and given capabilities the team describes as unique to Antigravity, which means the differentiator a vendor now competes on is how narrowly an agent can be scoped and how its verification is wired, not which model sits underneath.
That is the product layer this site has been tracking as the coding-agent workbench becoming its own product layer, and it inherits a problem already visible at scale: once specialised agents multiply, hundreds of agent skills create a lifecycle problem rather than a capability problem. The coding agents hub follows both.
The post describes the feature and its rollout order, not measured outcomes from division of labour. The condition to watch is whether custom agents ship an ownership and deprecation story, because a workspace that accumulates specialised agents without one converges on the same unmaintained sprawl as a scripts directory.