This Week in Agentic AI: November 2–9, 2026
This week saw a surge of new AI agent launches from major tech companies and specialized startups, alongside efforts to accelerate agentic development. Concurrently, initial reports on real-world agent adoption revealed varied usage patterns, from low consumer engagement to high enterprise-level investment. Research initiatives focused on benchmarking agent capabilities in scientific domains also progressed.
New Agent Platforms & Capabilities
Multiple entities launched or announced new AI agent capabilities this week, including Perplexity's work-focused Teammates and Meta's expanded Muse AI with video chat and voice modes. Apple is reportedly accelerating its Siri agentic roadmap, while Google DeepMind introduced agents for cheminformatics. Specialized agents for mathematical verification (Axiom Math) and future travel planning (Airbnb for 2027) were also revealed, alongside new agent interaction hardware from Aina.
Agent Adoption & Practical Challenges
Early data from Uber indicated extremely low consumer adoption of AI agents for ride and food ordering, with only four rides fulfilled via agents last quarter. In contrast, Y Combinator's Garry Tan reported significant personal token spending on AI agents. Practical feedback also emerged, suggesting strategies like hiding verbose output from coding agents to improve usability.
Advancing Agent Research & Evaluation
OpenAI introduced BioBasicsBench, a new 100-task benchmark designed to evaluate how well AI agents can interpret raw experimental data and perform routine wet-lab science tasks. This marks a step forward in standardizing the assessment of scientific AI agent capabilities.
Top stories this week
Perplexity launches Teammates AI agents for work
Perplexity introduced Teammates, always-on work-focused AI agents that can take on multiple roles in a workspace and draw on existing projects, skills, and memory from Perplexity tools. Teammates integrate with Perplexity Computer, which can deploy agents on a user's device.
Why it matters for builders
Perplexity's Teammates and Perplexity Computer give builders a new agent platform that combines workspace memory and on-device deployment, potentially lowering friction for automating multi-step work tasks compared to existing agent offerings.
Apple is racing to make Siri more agentic
Apple CEO John Ternus is pressuring AI executive Craig Federighi to accelerate Siri's agentic roadmap and deliver a powerful AI agent that can compete with OpenAI's Dots and Meta's Muse, according to Bloomberg's Mark Gurman. Federighi has already delivered a functional version of Siri AI quickly, but Apple knows challenges remain.
Why it matters for builders
As Apple pushes Siri toward a more powerful agent, builders should watch how a platform-owner voice agent could reshape user expectations for proactive assistance and app integration—especially with OpenAI and Meta raising the competitive bar.
Uber Fulfilled Only 4 Rides From AI Agents Last Quarter
During its Q3 2026 earnings call, Uber said third-party AI agents arranged four rides and around 500 Uber Eats orders in the quarter.
Why it matters for builders
The low transaction counts are an early reality check for builders integrating AI agents with service platforms, indicating agent-driven commerce is still nascent and volumes may be minimal at first.
Agentic cheminformatics with Google Antigravity
Google DeepMind introduces Antigravity and Cheminformatics Skills to accelerate drug discovery workflows, search compounds, and propose novel rigidified macrocycles.
Why it matters for builders
Demonstrates how domain-specific scientific tools can be exposed as agent skills, enabling multi-step compound search and molecular design workflows that builders can adapt for other verticals.
OpenAI Introduces BioBasicsBench: 100-Task Benchmark for AI Agents in Wet-Lab Science
OpenAI has released BioBasicsBench, a benchmark of 100 tasks that evaluates how well AI agents interpret raw experimental data and perform routine wet-lab science tasks.
Why it matters for builders
For developers building agents that handle scientific data, this benchmark provides a concrete evaluation set for testing multimodal reasoning and data interpretation in biology, potentially informing domain-specific agent design.