Google is turning Managed Agents into scheduled workers with budgets and hooks

October 9 update: Google's old `antigravity-preview-05-2026` agent identifier was scheduled for shutdown on October 5, 2026. Current docs direct developers to `antigravity-preview-09-2026`. The new agent defaults to `gemini-3.8-flash` when `agent_config.model` is omitted; Google's current custom-agent documentation says only the September preview ID is supported as `base_agent`. This is a breaking identifier migration, not just another implicit model upgrade. Gemini 3.7 Flash remains an explicitly selectable model alongside 3.6 and 3.5 variants. The current managed-agent platform still supports environment hooks, token budgets, scheduled triggers and persistent sandboxes described below. Google's October 7 Gemini 3.8 Flash documentation also confirms the introductory $0.75 input/$3.75 output per million token rate through December 31, 2026, with $1.50/$7.50 from January 1, 2027.

Historical context: Google originally introduced Managed Agents in May, added background execution and remote MCP in July, and subsequently introduced hooks, token budgets, scheduled triggers and model-default changes during the 05-preview period.

The practical new boundary is compatibility: code that hard-codes the retired May Antigravity agent identifier needs migration to the September ID. Builders should explicitly pin `agent_config.model` for reproducibility, test tool and sandbox behavior against the new runtime, and verify pricing and token usage. A platform-controlled default can change while your application code remains untouched, and now the agent identifier itself has a retirement lifecycle.

The original operational story still matters: environment hooks, budgets and schedules make Managed Agents closer to persistent workers than simple chatbot requests, but production teams need to treat those managed components as versioned dependencies.

Environment hooks can run custom commands or HTTP handlers before or after tool execution inside the managed sandbox. A pre-tool hook can deny a write or code-execution action, while a post-tool hook can lint, inspect or audit what the agent just produced.

That matters because prompt instructions are not the same thing as enforcement. Telling an agent not to edit a protected file is weaker than placing a deterministic gate in front of the write operation. Hooks give teams a place to attach existing policy, security and quality checks without trying to teach every rule to the model.

Managed agents can be given a `max_total_tokens` budget covering input, output and thinking tokens. When a run reaches that limit, execution returns incomplete while preserving its environment and context so the developer can inspect or continue it with a fresh budget. Google notes that the limit is best-effort and actual usage can slightly exceed it between budget checks.

This is a small control with a large practical effect. Autonomous agents can take unexpected paths, repeat failed approaches or spend far more than a short interactive prompt. A hard budget makes cost part of the runtime contract rather than something discovered after the invoice arrives.

Google has also added scheduled triggers that bind an agent, prompt, environment and cron schedule together. Runs reuse the same sandbox, so files can persist between executions.

That makes Managed Agents relevant to recurring jobs such as dependency audits, periodic research, repository maintenance or report generation. Builders should still think carefully about idempotency, stale state and what happens when one scheduled run overlaps another, but the platform is clearly moving beyond request-response agents.

Google's legacy `antigravity-preview-05-2026` documentation explicitly says the identifier was scheduled to shut down October 5, 2026 and directs migration to `antigravity-preview-09-2026`. The current managed-agent guide supports only that September ID as a base agent and defaults to `gemini-3.8-flash`. Explicit choices still include Gemini 3.7 Flash, 3.6 Flash, 3.5 Flash and 3.5 Flash-Lite. Do not assume an agent-ID replacement preserves behavior, tool permissions or state without tests; pin the underlying model where reproducibility matters.

Google lists Gemini 3.8 Flash at $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, with standard pricing of $1.50/$7.50 from January 1, 2027. The published introductory rates match those of 3.7 and 3.6 Flash, but a newer model may use more reasoning tokens, tool calls or retries, so teams should measure end-to-end task costs rather than assume unchanged invoices.