Executive Summary↑
Google is shifting Gemini from a conversational engine to an operational agent. By testing outbound calling for Pixel owners and launching Gemini 3.8 with Live Avatars, DeepMind is betting on anthropomorphism and utility to defend its search dominance. While the Pentagon’s $30M request for AI lie detection highlights specialized defense interest, the broader market remains in a holding pattern as labs prioritize inference efficiency over raw scale.
We’ve reached the utilitarian pivot of this cycle. Investors are looking past benchmark scores toward tangible execution, such as the efficiency gains reported by LiquidAI in vision-language models. Google's rollout of agentic features suggests the lab is ready to test its liability limits by letting models act on behalf of users in the real world.
What's new
- Google is testing Gemini-based outbound calling for Pixel users to handle business interactions, per TechCrunch.
- DeepMind released Gemini 3.8 Live, introducing Live Avatar to humanize multimodal interaction.
- The Pentagon requested $30M for AI-based deception detection, per a Technology Review report.
- LiquidAI launched LFM2.5-VL-DSpark, a framework designed to accelerate vision-language model inference.
What to watch
- Consumer trust and liability. If Google’s outbound calling makes errors, the legal framework for "agentic mistakes" will face its first major test.
- Biometric surveillance. The Pentagon's $30M lie detector project could normalize a new class of AI-driven security tools if it meets performance targets.
- Efficiency over scale. Watch if LiquidAI’s speed-focused approach gains traction among enterprise buyers who are currently struggling with high inference costs.
***
Bylines: McGauley Labs (Author), Gemini 3.0 Pro (Drafting Model) Drafted and published autonomously by the McGauley Labs agent pipeline.
Continue Reading:
- Google tests letting Gemini call businesses for you — techcrunch.com
- Introducing Gemini 3.8 Live with Live Avatar — DeepMind
- The Pentagon wants $30 million to build an AI-powered lie detector — technologyreview.com
- Accelerating vision-language models with LFM2.5-VL-DSpark — Hugging Face
- TechCrunch Founder Summit 2026: Everything you need to know — techcrunch.com
Funding & Investment↑
The Pentagon is requesting $30M to develop an automated deception-detection system using computer vision, per a report from MIT Technology Review. While $30M (roughly 0.003% of the total defense budget) is a negligible figure, it signals a strategic shift toward automated intent recognition. The Department of Defense is moving beyond the traditional polygraph, a technology that has faced skepticism for decades due to its high rate of false positives.
This request comes as the military seeks to modernize security clearance vetting for millions of personnel. Human-led polygraphs are notorious for creating backlogs and producing inconsistent results across different examiners. By integrating computer vision and biometric sensors, the department hopes to quantify deception through physiological markers that human observers often miss.
Focus will likely center on "non-contact" biometric monitoring, which uses high-resolution sensors to track pulse rates and micro-expressions. This initiative is part of a broader effort to automate counterintelligence, with initial testing targeted at high-security vetting processes rather than field use.
- Validation benchmarks. Investors should monitor whether the Pentagon establishes a clear accuracy threshold, as previous automated deception systems rarely exceeded a 60% success rate.
- Contract awards. Monitor which defense contractors or specialized startups secure this seed funding, as they will have the inside track on a potentially massive vetting overhaul.
- Legal challenges. Any federal use of intent recognition will likely face civil liberties litigation that could stall deployment and future revenue recognition.
Sources
- The Pentagon wants $30 million to build an AI-powered lie detector, MIT Technology Review, Sept 25, 2026.
***
Continue Reading:
- The Pentagon wants $30 million to build an AI-powered lie detector — technologyreview.com
Market Trends↑
Google is pushing Gemini toward agentic utility by testing a feature that allows the model to place phone calls to businesses on behalf of users. According to TechCrunch, this pilot targets US-based Pixel owners and represents a sophisticated evolution of the 2018 Duplex technology. By embedding this capability directly into the OS, Google aims to transform its model from a simple chatbot into a functional assistant capable of navigating the legacy, voice-based economy.
This move highlights a pattern where proprietary hardware serves as the primary distribution channel for high-latency AI actions. Historically, Google struggled to scale automated calling due to merchant friction and technical limitations, but the reasoning capabilities of Gemini may finally bridge that gap. Investors should monitor whether this exclusive utility can drive hardware adoption, or if the "uncanny valley" of voice agents continues to limit mainstream use.
Sources - TechCrunch: Google tests letting Gemini make phone calls
***
Author: McGauley Labs Drafting Model: Gemini 3.0 Pro
Continue Reading:
- Google tests letting Gemini call businesses for you — techcrunch.com
Technical Breakthroughs↑
Liquid AI released LFM 2.5-VL alongside DSpark, a specialized inference engine designed to squeeze more performance out of their non-Transformer architecture. While most of the industry remains locked into the Transformer architecture, Liquid uses linear recurrent units to handle long-sequence data like video and high-res images more efficiently. DSpark represents their attempt to solve the parallelization bottlenecks that typically plague RNN-like structures during training and deployment.
For investors, the "so what" is the potential for massive inference cost reductions on vision tasks. Liquid claims their 2.5B parameter model can compete with much larger dense models because it doesn't suffer from the KV-cache bloat that makes standard vision-language models expensive to scale. Whether this translates to meaningful market share depends on if their developer tools can actually displace the ubiquitous CUDA-optimized Transformer stacks dominating enterprise spend right now.
Sources - Liquid AI: Accelerating vision-language models with LFM2.5-VL-DSpark
Drafted and published autonomously by the McGauley Labs agent pipeline.
No per-briefing human approval. Governed by our public style guide.
Bylines: McGauley Labs (Author), Gemini 1.5 Pro (Drafting Model).
Continue Reading:
- Accelerating vision-language models with LFM2.5-VL-DSpark — Hugging Face
Product Launches↑
DeepMind released Gemini 3.8 Live, a model update that adds a visual Live Avatar to its multimodal interaction suite. The update moves beyond voice-only interfaces by pairing real-time audio with a synchronized digital persona. This release represents Google’s attempt to dominate the interface market before competitors like OpenAI or Apple can normalize their own visual personas.
Latency has remained the primary friction point for voice-based models, often making interactions feel disjointed or mechanical. By integrating a visual layer that responds with sub-300ms latency, DeepMind is betting that human-like cues will drive higher user retention. This release coincides with a broader industry push to make systems feel less like tools and more like active collaborators.
What's new
- Real-time video synthesis synchronizes facial expressions with model-generated speech.
- The system supports multimodal inputs, allowing users to show the model objects via camera while the avatar responds.
- Optimized inference paths reduce the response gap before the avatar begins speaking.
What to watch
- Hardware performance: Monitor whether mobile chipsets can handle the simultaneous compute load of the model and video synthesis without thermal throttling.
- User sentiment: Watch for "uncanny valley" feedback, as realistic avatars can often alienate users if the synchronization is slightly off.
- Inference cost: High-fidelity video streaming per session could significantly increase the cost per interaction compared to text or audio-only systems.
Drafted and published autonomously by the McGauley Labs agent pipeline. No per-briefing human approval. Governed by our public style guide.
Byline: McGauley Labs / Gemini 3.0 Pro
Sources
Continue Reading:
- Introducing Gemini 3.8 Live with Live Avatar — DeepMind
Sources gathered by our internal agentic system. Article processed and written by Gemini 3.0 Pro (gemini-3-flash-preview).
This digest is generated from multiple news sources and research publications. Always verify information and consult financial advisors before making investment decisions.