Matthew Li

agents lead @antigravity
formerly @windsurf, @nuro, @princeton

[Matthew Li @_matthewli]

@20thkim @kevinhou22 Yes! When an agent invokes a subagent, it can choose whether the subagent is using flash-lite, flash, or pro. If you have a preference about it you can guide it with prompting/rules

[Matthew Li @_matthewli]

@rronak_ @MichaelElabd @QuantumArjun congrats ronak! excited to see where you guys take this — also, excellent naming choice :)

[Matthew Li @_matthewli]

@seylorra they have tool inputs. the full agent context is accessible through a transcript file currently

[Matthew Li @_matthewli]

try these out! let me know what you think / want to see next

[Google Antigravity @antigravity]

Features in Antigravity 2.0 🧵
• Subagents
• Asynchronous Task Management
• Scheduled Tasks
• JSON Hooks
• Voice

[Google Antigravity @antigravity]

We’re thrilled to announce the Antigravity CLI, a lightweight way to spin up the same Antigravity agents right from the terminal. 💻

It gives you the exact same harness and same models, with a product experience tailored for the command line. It adapts entirely to you: your keybindings, your themes, your workflows.

Full Antigravity CLI Walkthrough:

[Matthew Li @_matthewli]

@socoffchar @OfficialLoganK @antigravity it’s much easier to coordinate multiple agents across multiple workspaces/projects. there’s also features like subagents and scheduled tasks

[Logan Kilpatrick @OfficialLoganK]

Very excited for all the stuff the @antigravity team has been cooking :)

[Matthew Li @_matthewli]

@premqnair you gotta update the bio now

[Matthew Li @_matthewli]

@mahmoud_me69121 @googledevs @GoogleAI should be fixed in next release!

[Matthew Li @_matthewli]

No model is perfect at instruction following. Despite refined prompting, even the best models can get confused at times.

At Antigravity, I developed "ephemeral messages" to augment the instruction following ability of models, letting us push the frontier of agentic capabilities.

[Matthew Li @_matthewli]

@michlimlim @cognition @graphite @tryflint congrats on the launch!

[Matthew Li @_matthewli]

@Sabatin02 @antigravity What do you want most that’s missing for MCP?

[Matthew Li @_matthewli]

@nub_insider @antigravity Reducing the amount of agent termination errors is one of our top priorities, we're actively working on it and there are a bunch of fixes already staged for the next release!

[Matthew Li @_matthewli]

@blankbraindead curious why you think so! one of the reminders is actually to pay more attention to the rules lol

[Matthew Li @_matthewli]

All said, ephemeral messages have helped us bridge the gap between model limitations and desired behaviors. Most reminders needed for Gemini 2.5 were removed for Gemini 3.

Our goal today is to keep finding the capabilities that are just barely possible on the strongest models.

[Matthew Li @_matthewli]

There are some downsides, like slight cache misses or thinking trace disruption. But more fundamentally, over-reliance can mask poor tool design, and too large a reminder can distract the agent. Ephemeral messages should typically only be used when it can’t be done otherwise.

[Matthew Li @_matthewli]

@samwhoo end of an era