Aurora has 20 driverless trucks running and wants 200 by year-end

Two stories today: driverless freight moves from demo to payroll, and JetBrains decides the agent era needs a landlord.

Aurora Innovation is running 20 driverless trucks on Texas highways and says it will have 200 on the road by the end of the year, with positive free cash flow targeted for 2028. The company let CNBC ride along for three hours on Interstate 45 between Houston and Dallas, where its rigs haul freight with nobody in the cab, reading the road with lidar, radar and cameras. The pitch is arithmetic: Bank of America estimates Aurora's service costs about $0.85 per mile against roughly $1.30 per mile for driver wages and benefits, and Morgan Stanley's Ravi Shanker puts an autonomous fleet at nearly 7.5 times the profitability of a human-driven one. Aurora has logged more than 440,000 miles since it began public-road testing in 2022, most of them with a safety operator aboard, and it lost more than $800 million in 2025.

The number worth watching is the 10x inside that plan. Twenty trucks is a science project; 200 is an operating business, and the deadline is 15 weeks away. Freight firms are short of drivers and federal hours-of-service rules cap a human far below the 20 hours a day Aurora says its trucks can run, so demand is not really the question. Whether shippers hand over judgement, customer service and liability — everything a driver does besides steer — is. Hardware is scaling elsewhere too: Pony.ai's fourth-gen driverless heavy truck enters mass production.

JetBrains pulled its agentic development work into one umbrella called Air — Air inside its IDEs, Air Teams for coordination, Air Governance for organization-level visibility and cost control — and is selling multi-vendor openness as the point. Its argument is that developers adopted agents faster than their organizations built the plumbing around them, so the costs surface later in review, rework, security and spend, and the hard problem is code that is almost right: plausible, passing a superficial check, quietly carrying a bad assumption. Air Governance is the piece that is actually new — it is meant to show which agents touched which code, where data went and what the spend was, including for providers JetBrains does not own. The company's product page lists Codex, Claude Agent, Gemini CLI and its own Junie running side by side.

The hedge at the end of the announcement is the interesting part. JetBrains says it will not name future products before their scope and availability are settled, and that its strategy does not depend on its own agent winning — which is what a vendor says when it expects the model layer to stay someone else's business. Twenty-six years of selling a workbench is being traded for a control plane above the agents. The same instinct showed up at Google last week — Google lets every engineer use Claude Opus 5 inside Antigravity.

What to watch: whether Aurora's fleet count actually reaches 200 by December 31, and whether Air Governance ships as a product or stays a slide.

Would you put a shipment on a truck with nobody in the cab — and does an agent governance layer change that answer? Tell us in the comments.

Brain organoids classified Braille at 83% — in a trio, not alone

Two stories today: human brain tissue on a chip decoded a tactile signal, and the data-centre company behind some of the largest AI contracts in Europe turns out to be six months behind on its flagship site. Human forebrain organoids grown on a microelectrode array classified Braille letters with 61 percent accuracy on their own and 83 percent when the responses of three organoids were combined, in a peer-reviewed study from the University of Bristol. The team took event-based tactile readings

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Microsoft finished pre-training its biggest in-house model yet — then used it to make a point about the company it bankrolls. Microsoft AI completed pre-training of MAI-Base-2, a mixture-of-experts foundation model with 80 billion active and 2.8 trillion total parameters, and says the recipe is 4.2–12.3x more pre-training FLOPs-efficient than DeepSeek-V4-Pro-Base. The run itself is the number worth keeping: 16,000 Nvidia GB200 GPUs for 38 days across August and September, at 21% average model

Amazon's new Bee devices listen all day — for $11.99 a month

Amazon turned its always-listening Bee line into a two-device family on Thursday, and NVIDIA opened a 2.8-trillion-token Indic corpus recipe. The connecting thread: both are bets that the next AI money is in memory — yours, or your model's. Amazon put two new always-listening devices on sale, and both now lead to a subscription. Bee Frontier, the next-generation wristband that also clips to clothing, ships now at $74.99 in Onyx, Lavender and Sage — up from the $50 predecessor — with seven-day

Perplexity open-sources its agent sandbox after 118 failed escapes

Perplexity handed its agent containment layer to the open-source world, and Washington's AI task force stopped reading briefings and started working. Perplexity open-sourced SPACE on Thursday — the sandbox it built to keep its own AI agents from getting out — and says nine frontier models failed to break out of it across 118 attempts. Each agent runs inside a small virtual machine with its own operating system, with credentials and network access managed outside the sandbox so an agent can re

Read more

Brain organoids classified Braille at 83% — in a trio, not alone

Suleyman says Microsoft's base model beat OpenAI's on a quarter compute

Amazon's new Bee devices listen all day — for $11.99 a month

Perplexity open-sources its agent sandbox after 118 failed escapes