Autonomous Driving

Today's Stories

1. 1

   Waymo: AI readiness hinges on eval metrics, not just model performance

   Waymo's director of engineering for systems intelligence and machine learning, Manasi Joshi, outlined the company's approach to training, testing, and deploying AI at scale at VB Transform 2026. The autonomous vehicle company has driven more than 220 million fully autonomous, or 'rider-only,' miles, with 17 times fewer serious crash injuries than human drivers over the same distance. Waymo's evaluation-first methodology—built on continuous evaluation, carefully curated data, human oversight, and clearly defined business outcomes—demonstrates a risk-management playbook for enterprises deploying AI agents in nearly any industry. For self-driving cars, the stakes are uniquely high: models must navigate unpredictable streets, respond to human drivers, and make split-second decisions in the physical world, not merely generate text or automate back-office tasks.

   Waymo's framework prioritizes when an AI project is ready for deployment: readiness is determined by evaluation metrics meeting defined thresholds, not by raw model performance alone. This approach reflects the company's focus on measurable, business-aligned outcomes rather than benchmark scores.

2. 2

   Qualcomm declares $0.92 dividend, pushes into automotive AI

   Qualcomm declared a quarterly cash dividend of US$0.92 per common share, payable September 24, 2026, to shareholders of record as of September 3, 2026. The company is also participating in Micron's new long-term automotive supply agreements, signaling a push to secure a role in AI-enabled vehicle platforms. Qualcomm is working to shift from a smartphone-centric business toward AI-driven edge and automotive markets to offset handset demand cyclicality and China-related pressures. The automotive agreements with Micron show tangible progress on diversification, which investors view as critical to offsetting geopolitical and regulatory risks to its global chip and licensing business.

   Q3 2026 earnings and any signs of stabilization in handset demand are the near-term catalysts. The biggest risk remains geopolitical and regulatory pressure on Qualcomm's global chip and licensing business, including the possibility of tighter export controls and shrinking access to key Chinese customers.

3. 3

   GM tripled pull requests by redesigning workflows for AI agents

   General Motors' autonomous driving division redesigned its engineering workflows around AI agents to handle tasks beyond coding—analyzing vehicle data, triaging problems, running experiments, and testing fixes. The shift resulted in roughly three times as many merged pull requests, faster releases, and fewer defects, according to Rashed Haq, GM's VP of autonomous vehicles. Software engineers at GM spend only 15% of their time writing code; the other 85% goes to non-coding work that AI agents can now automate. Rather than simply adding an AI coding assistant, GM restructured entire processes around agents, eliminating inefficiencies that persist when engineers rely on chatbots alone.

   GM's approach suggests that the biggest productivity gains come not from AI writing code, but from AI handling the broader engineering workflow—a model other automotive and software organizations may adapt as they scale autonomous systems development.

4. 4

   Synopsys Unveils Autonomous Chip-Design Agents, Cutting Verification Time 50X

   Synopsys announced fully autonomous AI agents for chip design and thermal simulation, built with NVIDIA Nemotron technology. The chip verification agent delivers up to 50X faster time-to-validated RTL while achieving 20% additional coverage improvement; a new thermal-management workflow automates setup, preprocessing, and post-processing. The company also expanded its portfolio to more than 20 GPU-accelerated EDA and multiphysics products, including an 18X speedup for PrimeSim SPICE simulations. Agentic AI—agents that reason, plan, and execute complex workflows autonomously—shifts engineering from manual, time-consuming tasks to automated insight generation. By collapsing chip verification cycles and compressing thermal analysis from weeks to hours, these tools address a critical bottleneck in product development. For R&D teams, the productivity multiplier could meaningfully shorten time-to-market and reduce engineering labor intensity.

   Synopsys demonstrated these capabilities for the first time at the 2026 DAC Chips to Systems Conference on July 26, 2026. The partnership between Synopsys and NVIDIA—combining Synopsys' domain expertise in EDA and CAE with NVIDIA's accelerated computing platform and runtime security—signals deepening collaboration in agentic engineering tools.

5. 5

   AT&T urges companies to slow AI adoption, avoid 'racing from stoplight to stoplight'

   AT&T has advised businesses to reconsider the pace of their artificial intelligence deployment, warning against what it calls 'racing from stoplight to stoplight'—a metaphor for reactive, short-term decision-making driven by industry hype rather than strategic planning. The guidance reflects a shift in how major telecommunications and enterprise leaders view AI implementation. Rather than adopting every new capability as it emerges, AT&T is suggesting that companies should align AI investments with long-term business objectives and workforce readiness, reducing the risk of costly missteps or underutilization of deployed systems.

   This messaging may influence how other large enterprises and their vendors approach AI rollout timelines and procurement decisions, potentially slowing near-term deployment announcements but encouraging more deliberate, sustainable implementation strategies.

6. 6

   When Should You Say No to AI's Help?

   An essay explores when delegating decisions to AI might undermine personal growth. The author contrasts AI's convenience—drawing on more examples than any individual has seen and remaining detached from mood—with the risk that outsourcing big life choices (job, relationships, moves) atrophies decision-making skills and strips meaning from life, much like video game cheat codes initially felt rewarding but drained the fun. People rarely face major decisions, so each one is a chance to practice and learn. Relying on AI's instant, plausible answers risks trapping you in a local maximum—always choosing the least uncomfortable option—and leaves you wondering whether you are still the author of your own life. The author notes that without struggling through a choice and reaping its consequences, the feedback loop closes and learning stops.

   The author acknowledges the tension is unresolved. AI will become more useful and harder to ignore, but some difficulties—like tedious work or exploring curiosity—benefit from AI help without sacrificing meaning. The key is noticing when AI's convenience starts making your life smaller and adjusting how you use it.