
Agentic AI
🔧 Databricks Gives Data Systems a Self-Diagnosing Agent
What happened
Databricks released Genie ZeroOps in beta, an always on agent that monitors jobs, data pipelines, and tables for failures or quality problems. When something breaks, it investigates the root cause and proposes a fix rather than waiting for an engineer to start troubleshooting.
Why it matters
Agents are moving from doing work when asked to watching systems continuously and deciding when something needs attention. That could turn parts of IT and data operations from reactive support into automated maintenance.
What’s next
Genie ZeroOps must be enabled by a workspace administrator and remains a beta product. The bigger test is how much troubleshooting companies are willing to delegate once agents move from identifying failures to proposing changes that affect production systems.
🛡️ Picus Lets AI Red Teams Attack Continuously
What happened
Picus Security launched Autonomous Penetration Testing, which uses AI agents to map environments, choose attack paths, chain real exploits, and test whether critical systems can actually be reached. Organizations define the allowed scope and techniques beforehand, with hard-deny rules and a kill switch the agent cannot override.
Why it matters
Security testing has traditionally been periodic, even though infrastructure and threats change constantly. Agentic testing turns the exercise into a continuous loop where AI can explore like an attacker while the business retains control over where the testing is allowed to go.
What’s next
The product is available to Picus customers today, while its fully agentic mode remains in early access. The important measure will be whether continuous autonomous testing finds meaningful attack paths without creating unacceptable operational risk.
Generative & Enterprise AI
⚡ Google Builds a Faster Workhorse for Agents
What happened
Google released Gemini 4 Hydrogen, a faster and lower cost member of the Gemini 4 family designed for coding and high volume agentic workflows. Google also introduced an Orchestrator in Antigravity that can switch between models such as Carbon and Hydrogen during the same job, using stronger models for difficult reasoning and Hydrogen for faster execution.
Why it matters
Agent economics increasingly depend on using the right intelligence for each step rather than sending every task to the most powerful model available. Routing planning, execution, and verification across different models could reduce both latency and cost as autonomous workflows generate more model calls.
What’s next
Hydrogen is rolling out globally across Google AI Studio, the Gemini API, Gemini for business, the Gemini app, Search, and Antigravity. Google reports its Orchestrator used 24% less quota and completed tasks 19% faster than a Carbon-only approach on one benchmark, a vendor-reported result that will need broader validation.
🧪 DuPont Puts Agents Into Industrial R&D
What happened
DuPont announced a collaboration with Microsoft and NobleAI that combines Microsoft Discovery, an agentic platform for scientific research, with NobleAI’s science-based modeling tools. The system is intended to help DuPont scientists evaluate more potential materials and solutions digitally before committing to physical experiments.
Why it matters
Scientific AI becomes more valuable when it helps determine which experiments are worth running, not simply when it summarizes existing research. Agentic systems can coordinate models, data, and experiments so expensive physical testing focuses on the candidates most likely to succeed.
What’s next
DuPont plans to expand access to the tools across its research organization with security and governance built into the deployment. The real measure will be whether AI-assisted experimentation materially shortens development cycles and improves the rate at which promising ideas reach customers.
Physical AI
🦾 GXO Prepares Apollo for a Thousand-Robot Rollout
What happened
GXO and Apptronik signed a letter of intent to expand trials of Apollo humanoid robots across warehouse tasks including picking, tote movement, packing, and sortation. If performance, safety, and operational targets are met, the companies intend to work toward deploying up to 1,000 Apollo robots globally over four years using a Robotics-as-a-Service model.
Why it matters
Humanoid robotics is beginning to move from isolated pilots toward commercial deployment frameworks. A service model could also lower the barrier to adoption by letting warehouse operators treat robots more like operating expenses than large upfront equipment purchases.
What’s next
Expanded trials begin in 2027, with each workflow measured against agreed productivity and safety metrics. Initial commercial deployment is targeted for North America in late 2027, assuming the robots meet those requirements.
💰 Agile Robots Raises $500 Million to Scale Physical AI
What happened
Agile Robots raised $500 million in a SoftBank-led Series F that values the company at $8.5 billion, with NVIDIA, Hitachi, and others participating. The company says it has more than 20,000 robotic systems installed worldwide across humanoids, robotic arms, mobile platforms, and other industrial automation.
Why it matters
Physical AI is entering a manufacturing and deployment race where capital matters almost as much as model capability. Building useful robots at scale requires factories, hardware supply chains, field support, and enough production capacity to move beyond demonstrations.
What’s next
Agile Robots plans to use the funding to expand into the United States, increase production capacity in Germany, and enter additional markets. Its recently opened German facility is designed for annual capacity of up to 10,000 humanoid robots, which represents potential capacity rather than confirmed output.
💡 Bottom Line
AI is becoming an operating layer rather than a tool people manually invoke. Agents are starting to watch systems, diagnose failures, test defenses, and coordinate specialized models while robots move toward service-based deployment at real scale. The advantage is shifting from having intelligence to building systems that can keep using it reliably.
⚙️ Try It Yourself
Give a Faster Model a Multi-Step Job
Google says Gemini 4 Hydrogen is designed for fast execution across agentic workflows and is available through Google AI Studio and Gemini.
Try this: Give Hydrogen a fictional product launch with five requirements: a deadline, budget, target customer, marketing channel, and approval rule. Ask it to create the plan, identify dependencies, and flag anything requiring human approval. Then change only the budget and ask it to update the plan without changing unaffected steps.
Does the model preserve the original constraints while adapting only the parts that actually depend on the new budget?
Insight: Agent performance is not just intelligence. Reliable execution means maintaining context while adapting to change.
