Fetching from the wire…
01
02
03
04
05
06
07
08
Public story · 2026-09-24 · high
The new managed-agent harness cuts output tokens 40% on file edits, and the old one stops working October 5.
Why now: Google detailed the change in its September 24 developer documentation, with the old harness cut off October 5.
Google has moved Gemini API managed agents onto a new harness, antigravity-preview-09-2026, built on the same design as its Antigravity coding agent. The old 05-2026 harness is deprecated on October 5. Anyone with an agent pinned to that version has about ten days before it stops working.
The new harness runs on Gemini 3.8 Flash by default, though the model is configurable per interaction, and Google isn't charging more for it. Output tokens on file-change tasks drop 40%. Multi-turn task completion rises as much as 6%, and cache hit rates climb as much as 16%, according to Google AI Studio's announcement. New Files and Credentials APIs ship alongside it, letting agents move data around and authenticate to MCP servers without custom plumbing.
The risk sits with teams that don't watch this closely. Unattended or scheduled agents running against the Gemini API are the likely failure point. Nobody's checking their logs daily, so the first sign of trouble is October 5 itself.
The announcement doesn't say whether there's a compatibility mode. It also doesn't mention a way to pin to the old harness past the cutoff, or offer a migration guide beyond the API docs themselves. If you've got agents running against Gemini's managed harness, check your configuration now, not after something breaks.
Each link below shares sources, entities, or timing with this story.
GA and stable for production across the Gemini API, Enterprise, and Antigravity, and now the default in the Gemini app and AI Mode in Search globally. Google pitches frontier-level intelligence at ~4x the speed of comparable models, priced at $1.50/$9 per 1M tokens, 1M-token c...
A June 18 model-tracking roundup reports Google set Gemini 2.5 Flash as the default across its consumer Gemini products, prioritizing latency and cost. This is single-source as of writing, so flag it pending the official Gemini blog, but it lines up with Google's other June 18...
Announced at Google I/O on May 19 and detailed here, the harness added parallel-task subagents, cross-platform terminal sandboxing, credential masking, and hardened Git policies. In early June, Google reset Gemini quota counters to zero for everyone and shipped a refreshed Gem...
"Gemini is Cooked but GCP is Cooking" argues Google quietly shelved 3.5 Pro, which industry chatter placed at roughly Opus 4.5 level, shipping Gemini 3.6 Flash as a bridge the authors call worse than Muse Spark 1.2, Grok 4.5, and tier-1 Chinese open-source models. The hard num...
arXiv 2609.15983 from Honghao Lin, David P. Woodruff, Vahab Mirrokni and colleagues explores alternative proof strategies in parallel, uses a readiness gate before decomposing a plan into interdependent subproblems, and routes verifier feedback back to the specific failing par...
Google launched agentic video understanding across Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite (Google). Instead of scanning a video start to finish at a fixed sample rate, the model runs an internal loop deciding what to watch, at what speed, and through which channel: fra...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.