The Daily Byte
Breaking
Welcome to The Daily Byte — real stories from across Pakistan.
Saturday, 21 November 2026 · Pakistan

The Daily Byte

Stories from Pakistan that deserve to be heard
Minor World

Google Antigravity Adds Dynamic Workflows for AI Coding

By The Daily Byte ·14 Nov 2026 ·3 min read
The Googleplex headquarters building in Mountain View, California, with the colourful Google logo on its glass facade

Google has rolled out Dynamic Workflows in Antigravity, its AI coding platform, giving developers a way to run teams of coding agents through a short Python script instead of one long conversation. The feature, announced on 13 November, is driven by a new /run-workflow command in the Antigravity app and command-line interface.

When a developer types /run-workflow followed by a task, the main agent writes a Python script that coordinates the work. The script breaks the job into named phases and sends each piece to a separate subagent, which works in its own conversation and returns a structured result checked against a JSON Schema. Antigravity shows the script to the developer for approval before it runs, unless a permission preset already allows it.

Parallel reviews and staged fixes

The point of the feature is predictable control flow. Instead of asking one agent to work through a long list of files in a single conversation, a developer can ask for the same check to run across every changed file at once. In Google's example, one phase finds the Python files changed since the main branch, and a second phase reviews each file in parallel for unhandled exceptions, with a one-line verdict per file.

The generated scripts use five building blocks: phase() to group work into named stages, agent() to define a subagent, parallel() to run agents at the same time, pipeline() to run them one after another, and trajectory() to bring earlier steps from the main conversation into the script. Each subagent returns its result through a finish tool that validates it against the schema the script specifies, so the main conversation receives clean, structured answers rather than pages of working notes.

Workflows can also keep a single agent alive across phases. A persistent reviewer can list the bugs in a codebase, hand each bug to a separate fixer agent working in parallel, and then check that every fix is correct. When agents need to edit code at the same time, each one can be given its own isolated Git worktree, so they can change files and run tests without overwriting each other's work.

Who can use it and what it costs

Dynamic Workflows are available to Google AI Pro and Google AI Ultra subscribers in both the Antigravity app and the Antigravity CLI. There is no separate price for the feature; workflows consume the plan's existing model credits, although Google notes that high-fanout workflows can trigger many model calls, so choosing lighter models for routine checks is the main way to control usage.

The workflow runs in a separate process using the Python runtime bundled with Antigravity, so no local Python installation is needed, although developers can point to a different interpreter with the ANTIGRAVITY_PYTHON environment variable. A generated script can be saved as a .py file in the project and run again later, which makes recurring audits, refactors and test runs repeatable and reviewable alongside the code itself.

The launch caps a busy week for Antigravity. Google introduced the Gemini 4 Carbon and Gemini 4 Hydrogen models on 10 November, followed by Orchestrator on 12 November, which switches between the two models inside a single session. Dynamic Workflows take a different approach: rather than switching models mid-task, they hand coordination to a script the developer can read, approve and reuse. Google's own guidance draws the line clearly: use /run-workflow when the job is a repeated check or an ordered sequence of stages, and use its /swarm mode when agents need to negotiate or review each other's work directly.

Sources: AlphaSignal and Google Antigravity Docs.