Agents Get Funded. Robots Get Deployed. Malware Gets Smarter.

What happened
Temporal, a rising star in workflow orchestration for AI agents, locked in a massive $300 million funding round led by Andreessen Horowitz, pushing its valuation to $5 billion. Decagon, another agentic AI player, tripled its valuation to $4.5 billion as its customer service agents went live at Duolingo and Hertz. Meanwhile, Complyance and Cogent Security raised $20M and $42M, respectively, to automate risk, compliance, and security workflows with fleets of specialized AI agents.

Why it matters
The money is following the action: enterprises are hungry for reliable, scalable agentic AI to automate complex, high-volume tasks. The focus is shifting from flashy demos to robust, auditable platforms that can handle real-world workflows—think customer support, compliance, and security—without human babysitting.

What’s next
Expect a surge in enterprise deployments and a new wave of agentic platforms promising not just automation, but transparency and control. The infrastructure arms race is on, and the winners will define how businesses run in the AI era.

What happened
Google dropped Gemini 3.1 Pro, its most advanced reasoning model yet, now powering everything from the Gemini App to Vertex AI. With support for million-token contexts and best-in-class performance on key benchmarks, Gemini 3.1 Pro is gunning for the enterprise. Not to be outdone, OpenAI’s GPT-5.3-Codex rolled out to paid ChatGPT users, setting a new bar for autonomous coding and cybersecurity readiness. Anthropic’s Claude Opus 4.6 also made waves with agent teams that can tackle complex projects in parallel.

Why it matters
Generative AI is no longer just about chatbots—it’s about automating entire workflows, from writing and debugging code to reviewing documents and even creating new AI models. The leap in context size and reasoning means these models can handle more nuanced, high-stakes tasks, while new safety and audit features address enterprise concerns.

What’s next
Look for generative AI to become the backbone of business operations, with models acting as tireless collaborators and even self-improving their own code. The line between “AI tool” and “AI teammate” is blurring fast.

What happened
Boston Dynamics’ all-electric Atlas wowed CES crowds with Olympic-level agility and is now prepped for industrial deployment, with every 2026 slot already spoken for. Agility Robotics’ Digit robots started work at a Toyota plant in Canada, unloading auto parts and bridging automated lines. Dobot’s Atom humanoid entered mass production at a $27,500 price point, targeting factories and retail with synchronized movement and dexterous hands. Qualcomm jumped in with a new robotics platform and processor, aiming to power the next generation of humanoids and mobile robots.

Why it matters
Humanoid robots are moving from sci-fi to shop floor, tackling repetitive, physically demanding jobs and filling labor gaps. The shift from pilot programs to real-world deployments signals that physical AI is ready for prime time, with hardware and software advances making robots more capable, affordable, and adaptable.

What’s next
Expect to see more robots working alongside humans—not just in factories, but in logistics, retail, and even public spaces. As costs drop and capabilities rise, the “robot coworker” could become a reality for millions.

What happened
Humanoid robots took a step closer to factory floors as Toyota Motor Manufacturing Canada (TMMC) signed a commercial “robots‑as‑a‑service” deal with Agility Robotics to deploy Digit humanoid workers. After a year‑long pilot, TMMC will roll out seven Digit robots in its Ontario plants to unload totes of auto parts and handle other repetitive tasks. Agility’s Digit is designed to work alongside humans without requiring major retrofits and can continuously learn tasks via its cloud‑based Arc platform. The deal follows similar deployments at logistics firms GXO, Schaeffler and Amazon.

Why it matters
Real‑world deployments of humanoid robots have lagged behind flashy demos. Signing Toyota, one of the world’s most efficient manufacturers, is a significant vote of confidence: TMMC president Tim Hollander said Digit will improve employee experience and efficiency. The robots handle monotonous, physically taxing jobs that are hard to staff, allowing human workers to focus on higher‑value tasks. Agility CEO Peggy Johnson framed the partnership as a step toward “cooperatively safe” humanoid robots that can work alongside people.

What’s next
Agility is preparing a next‑generation Digit that it claims will be the first cooperatively safe humanoid robot. If Toyota’s deployment is successful, expect more RaaS contracts as automakers and logistics firms address labor shortages and ergonomics concerns. Meanwhile, Japan announced plans to mass‑produce two humanoid models by 2027 and will publish an AI robotics strategy in fiscal 2026—signaling growing national‑level investment in physical AI.

What happened
Soft robotic arms, prized for their gentle touch but notorious for their unpredictability, just got an upgrade. Researchers from the Singapore‑MIT Alliance for Research and Technology (SMART) and collaborators at NUS and NTU have developed an AI control system that allows soft robots to learn a wide repertoire of motions and adapt on the fly without retraining. Inspired by the brain, the controller uses two sets of “synapses”: structural synapses that are trained offline on foundational movements, and plastic synapses that adapt online based on real‑time feedback while a stability measure ensures smooth motion. In tests, the system reduced tracking error by up to 55 % under disturbances and maintained over 92 % shape accuracy even when half the actuators failed.

Why it matters
Soft robots have enormous potential for healthcare, assistive tasks and manufacturing because they’re safer around humans. But controlling them is hard: their flexible materials deform in unpredictable ways, and existing systems often trade adaptation for stability. The new controller meets all three key requirements for real‑world soft robots—learning transferable skills, adapting instantly to new conditions and remaining stable. MIT’s Daniela Rus says the work brings versatile soft robots closer to operating safely alongside people in clinics and factories.

What’s next
The researchers plan to extend the technology to faster, more complex robotic systems and integrate it into assistive and medical devices. A generalizable controller could accelerate adoption of wearable soft exosuits, robotic rehabilitation arms, and flexible manipulators for delicate manufacturing. As physical AI advances, expect more cross‑disciplinary work combining materials science, machine learning and neuroscience.

What happened
The U.S. Army’s Combined Arms Doctrine Directorate (CADD) has begun training doctrine writers to use generative AI tools to search historical texts and generate draft paragraphs—part of a push to shrink a process that can take years. A four‑pronged strategy trains every writer on approved AI tools, designates a “master gunner” in each division to provide expert assistance, incorporates AI best practices into courses and seeks to build a custom AI tool with industry partners. CADD director Richard Creed stresses that AI is a tool, not a “crutch”: human experts must review every line for accuracy.

Why it matters
Militaries worldwide are experimenting with generative models for planning, analysis and communications. The Army’s transparent acknowledgement of AI’s critical flaws—hallucinations, outdated sources and factual confusion—shows a cautious approach. Lt. Col. Scott McMahan likens the AI assistant to a motivated junior officer: helpful for idea generation and efficiency but not a replacement for expertise. Using AI to assemble doctrine could free up officers for more strategic work, but there’s a risk of embedding errors into foundational military guidance.

What’s next
CADD is working with the Combined Arms Command and industry to develop a purpose‑built AI tool for doctrine writers. As generative models improve, the Army may adopt them for intelligence analysis and simulation—similar to the Pentagon’s broader GenAI.mil initiative. The ethical debate continues: the Pentagon is currently in a dispute with Anthropic over military use of the Claude model. Balancing AI‑enabled efficiency with reliability and ethics will be crucial as defense organizations deploy agentic tools.

What happened
ESET researchers have uncovered PromptSpy, an Android malware family that abuses a generative AI model to ensure it stays on a victim’s device. The malware, disguised as a legitimate banking app, uses Google’s Gemini to generate step‑by‑step instructions on how to pin itself in the recent apps list, making it harder to close or uninstall. PromptSpy can capture lock‑screen data, block uninstallation attempts with invisible overlays, gather device information, take screenshots and record screen activity. It appears financially motivated and currently targets users in Argentina; ESET has not yet seen it widely in the wild.

Why it matters
This is the first known Android malware to embed a generative AI model directly in its execution flow. Using a language model to adapt UI manipulation across different devices and operating system versions makes the malware more resilient and scalable. The discovery underscores that agentic AI techniques—task decomposition, tool calling and reasoning—are a double‑edged sword: they can power productivity or enable more sophisticated cyber threats.

What’s next
Cybersecurity firms expect more malware to leverage generative AI for persistence, evasion and automated exploitation. For now, PromptSpy is distributed via a dedicated website and isn’t available on Google Play; Google Play Protect blocks known versions. Users should avoid sideloading apps, keep devices updated and enable Play Protect. As AI models become more ubiquitous, industry and policymakers will need to bolster AI‑aware malware detection and establish norms around misuse. The same agentic capabilities that empower developers will increasingly be weaponized by attackers.

What happened
Elon Musk’s xAI scored a $3 billion investment from Saudi-backed Humain, with a twist: after SpaceX acquired xAI, those shares convert to SpaceX equity, consolidating Musk’s AI and space ambitions. TCS and OpenAI inked a deal to co-develop agentic AI solutions and build massive AI infrastructure in India. Cognizant and Google Cloud deepened their partnership to operationalize agentic AI at scale, while Telefónica and Mavenir launched an AI Innovation Hub for telecom.

Why it matters
The AI industry is consolidating fast, with big tech, sovereign wealth, and global consultancies joining forces to build the next generation of AI infrastructure and applications. These deals aren’t just about money—they’re about controlling the platforms, data, and talent that will shape the future of AI.

What’s next
Watch for more cross-industry alliances and mega-investments, especially in regions racing to build sovereign AI capabilities. The battle for AI leadership is going global, and the stakes are only getting higher.

Autonomous systems are moving from pilot projects to core infrastructure—funded, deployed, and increasingly embedded in both enterprise workflows and the threat landscape. The agentic era isn’t theoretical anymore; it’s operational.