Sources say Google has created a strike team to improve its coding models; Sergey Brin told DeepMind staffers they must aggressively pivot to catch up on agents
The InformationErin Woo
Context & Ripple Effects
Google’s AI effort has repeatedly been reorganized around competitive catch-up: leadership consulted Sergey Brin and Larry Page amid the ChatGPT challenge in early 2023, then consolidated its research groups into Google DeepMind to push research toward products.
Related coverage later ties the coding-model effort to a broader attempt to close ground on Anthropic, while earlier reporting showed progress—and internal concern—around reasoning models. This makes the agent push an escalation of an existing productization and capability race, not an isolated initiative.
First-order effects
Google DeepMind’s near-term priorities shift toward coding capability and agents, with a dedicated team concentrating work that may previously have been dispersed.
Sergey Brin’s intervention raises the operational urgency for DeepMind staff, increasing pressure to translate research work into competitive agent-oriented models.
Second-order effects
A focused Google effort intensifies competitive pressure on other frontier-model developers, particularly where coding performance and agent behavior are becoming key comparison points.
Within Google, the move is likely to sharpen trade-offs over research talent and model-development attention between reasoning, coding, and other AI product priorities.
Third-order effects
If such internal strike teams become a recurring response to model gaps, frontier AI competition will increasingly be organized around rapid capability-specific reallocation rather than broad, unified research agendas.
The pattern points toward agents becoming a core measure of model competitiveness; whether that produces durable differentiation will depend on whether teams can convert coding gains into reliable deployed systems.
The trend: Frontier AI labs are moving from general model races toward tightly managed competitions for agentic and coding capabilities that can be productized.
The Gemini models are probably the best base model of any of them, but the behavior and RL on the model is... not amazing It's REALLY good, but also super inconsistent If they can nail the RL environment on top of the base model, that'll be huge deal
This is going to be interesting because it's not clear Google can duplicate what Anthropic has done with agents, not because the technology is hard, but the cost of turning one prompt into hundreds are too high, too unprofitable. Google, a public company, can't light money on fi…
Google DeepMind formed a strike team to improve its coding models, with Sergey Brin directly involved. It's surprising to me that Google has the world's largest internal codebase (>2B LOC), yet lags behind Anthropic and OpenAI in coding + agents. “Google's AI writes 50% of [image…
find it odd that every Deepmind person I come across works on post training but gemini is probably the worst post trained model out of all frontier models
Please let this be before Google I/O 2026 Solid coding models from Google would be amazing since they have so many surfaces I can use it: - CLI - Antigravity - Jules - AI Studio - API in another harness using my $10 cloud credits
yeah bro sergey brin is trying to internal coding model maxx because anthropic is mogging them on agentic coding. chat is google cooked. is google unc. is google chopped https://www.theinformation.com/ ...