Engineering

One of the new, buzzy jobs in Silicon Valley is the AI Forward Deployed Engineer (FDE), an engineer who is embedded within a client organization to help customize solutions, such as building and tuning agentic workflows that suit the client’s particular needs. I’ve heard from people who are wondering anew about the FDE career path since OpenAI and Anthropic started building new teams to place FDEs within client organizations. The rise of FDEs for AI workloads is one way AI is creating new jobs (and why the jobpolcalypse narrative of upcoming job market collapse is false -- there will be many AI and non-AI jobs). However, I believe there will be far more AI Engineer jobs than FDEs, as I explain below. The FDE role was pioneered about two decades ago by Palantir, which sent engineers to government locations to work on secure, air-gapped networks. In addition to having good technical skills, FDEs need communication skills and sometimes business skills. For example, they may need to speak with clients to understand their needs, formulate a strategy to prioritize projects, explain complex technology, and respectfully push back if a client asks for something unrealistic. They’re enjoying a resurgence because of the amount of work involved in taking an off-the-shelf LLM and building it into a custom agentic workflow that fits particular business needs. However, the number of AI Engineer jobs will be far larger. A company might accept a few FDEs to be embedded within its organization. But most companies will want far more of their own employees working on their projects. While my organizations do hire FDEs, we hire far more AI Engineers! Also, a common client concern is that it is hard to find vendor-neutral FDEs — they are, after all, there to deeply integrate a particular vendor’s product into a company. In this moment when it’s hard to predict which AI service will be the best one in a year’s time, optionality (the ability to pick whatever vendor turns out to fit best in the future) is very valuable. In contrast, letting FDEs tightly bind a company’s processes significantly reduces optionality. Right now, I see surging demand for AI Engineers who can build software applications using AI software components (like LLM prompting, agentic frameworks, evals, etc.) and effectively use AI coding agents (like Claude Code, Codex, Antigravity CLI, and OpenCode). As the AI Engineer role matures, I expect it to fragment into more specialized roles, like the generic Software Engineer role from decades ago fragmented into frontend, backend, mobile, data engineering, devops, and so on. What will be the future, specialized AI engineering roles? I don’t know. Perhaps there will be AI FDEs, LLMOps Engineers, Evals Engineers, AI Data Engineers, Harness Engineers, and other roles we don’t have names for yet. But for now, I see a lot of AI engineers who are generalists create a lot of value. Skilled AI Engineers are in very high demand! [Original: The Batch]

AI raised the floor. Engineering excellence raises the ceiling. It's so riveting to see new LLM models get published and the step changes that are happening. AI has made it dramatically easier to produce code. It has simultaneously made it much harder to hide weak engineering fundamentals. AI is raising the floor, meaning more people can generate software and prototypes quickly. But engineering excellence raises the ceiling: determining whether that code becomes a reliable, scalable system that actually creates enterprise value. AI is exposing something many organizations have quietly carried for years: technical debt, fragile architectures, and disconnected data foundations. When systems aren't built well, AI doesn't fix that. It simply reveals it faster. 💡  𝗜 𝗮𝗺 𝗮 𝘀𝘁𝗿𝗼𝗻𝗴 𝗯𝗲𝗹𝗶𝗲𝘃𝗲𝗿 𝘁𝗵𝗮𝘁 𝘁𝗼 𝗺𝗮𝘅𝗶𝗺𝗶𝘇𝗲 𝗔𝗜 𝘃𝗮𝗹𝘂𝗲, 𝘄𝗲 𝗻𝗲𝗲𝗱 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗲𝘅𝗰𝗲𝗹𝗹𝗲𝗻𝗰𝗲. So what does engineering excellence look like right now? I think about it as four pillars: ▸ 𝗔𝗜-𝗥𝗲𝗮𝗱𝘆 𝗔𝗿𝗰𝗵𝗶𝘁𝗲𝗰𝘁𝘂𝗿𝗲: AI doesn't work well on top of poor architecture. Modernizing legacy code without addressing underlying structure just produces the wrong architecture faster. ▸ 𝗛𝗶𝗴𝗵-𝗤𝘂𝗮𝗹𝗶𝘁𝘆 𝗗𝗮𝘁𝗮 𝗙𝗼𝘂𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀: AI is only as intelligent as the data it reasons over. You can't shortcut this layer and even a strong foundation must continuously evolve. ▸ 𝗦𝗲𝗰𝘂𝗿𝗲 𝗮𝗻𝗱 𝗢𝗯𝘀𝗲𝗿𝘃𝗮𝗯𝗹𝗲 𝗦𝘆𝘀𝘁𝗲𝗺𝘀: As AI agents become more autonomous, seeing what's happening and why becomes non-negotiable. Governance isn't just policy it's instrumentation and operationalization, as many of you noted in my last post. ▸ 𝗗𝗶𝘀𝗰𝗶𝗽𝗹𝗶𝗻𝗲𝗱 𝗘𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗣𝗿𝗮𝗰𝘁𝗶𝗰𝗲𝘀: Spec discipline, test rigor, strong code review, clear ownership are not legacy practices to abandon, but more important than ever. AI rewards good fundamentals and makes the consequences of weak ones more visible, faster. There's a real shift in how engineers spend their time. Less writing foundational code. More orchestrating systems: designing architecture, shaping how AI agents interact, validating outputs with genuine judgment. I see our senior engineers flying because their systems thinking depth makes AI a true force multiplier. Earlier-career engineers are learning, but need more deliberate mentorship than ever. When AI can simulate senior output, the risk is gaining confidence without gaining understanding. The best thing leaders can do: create conditions where engineers are proud of how they build, not just what they ship. The time savings alone aren't the win. For us, we are investing in deeper architecture work, stronger data foundations, the next generation of agentic capabilities and I believe that's the winning combo. 𝗗𝗼 𝘆𝗼𝘂 𝗮𝗴𝗿𝗲𝗲 𝘁𝗵𝗮𝘁 𝗲𝗻𝗴𝗶𝗻𝗲𝗲𝗿𝗶𝗻𝗴 𝗲𝘅𝗰𝗲𝗹𝗹𝗲𝗻𝗰𝗲 𝗶𝘀 𝗺𝗼𝗿𝗲 𝗶𝗺𝗽𝗼𝗿𝘁𝗮𝗻𝘁 𝘁𝗵𝗮𝗻 𝗲𝘃𝗲𝗿?

Exciting updates on Project GR00T! We discover a systematic way to scale up robot data, tackling the most painful pain point in robotics. The idea is simple: human collects demonstration on a real robot, and we multiply that data 1000x or more in simulation. Let’s break it down: 1. We use Apple Vision Pro (yes!!) to give the human operator first person control of the humanoid. Vision Pro parses human hand pose and retargets the motion to the robot hand, all in real time. From the human’s point of view, they are immersed in another body like the Avatar. Teleoperation is slow and time-consuming, but we can afford to collect a small amount of data.  2. We use RoboCasa, a generative simulation framework, to multiply the demonstration data by varying the visual appearance and layout of the environment. In Jensen’s keynote video below, the humanoid is now placing the cup in hundreds of kitchens with a huge diversity of textures, furniture, and object placement. We only have 1 physical kitchen at the GEAR Lab in NVIDIA HQ, but we can conjure up infinite ones in simulation. 3. Finally, we apply MimicGen, a technique to multiply the above data even more by varying the *motion* of the robot. MimicGen generates vast number of new action trajectories based on the original human data, and filters out failed ones (e.g. those that drop the cup) to form a much larger dataset. To sum up, given 1 human trajectory with Vision Pro  -> RoboCasa produces N (varying visuals)  -> MimicGen further augments to NxM (varying motions). This is the way to trade compute for expensive human data by GPU-accelerated simulation. A while ago, I mentioned that teleoperation is fundamentally not scalable, because we are always limited by 24 hrs/robot/day in the world of atoms. Our new GR00T synthetic data pipeline breaks this barrier in the world of bits. Scaling has been so much fun for LLMs, and it's finally our turn to have fun in robotics! We are creating tools to enable everyone in the ecosystem to scale up with us: - RoboCasa: our generative simulation framework (Yuke Zhu). It's fully open-source! Here you go: http://robocasa.ai - MimicGen: our generative action framework (Ajay Mandlekar). The code is open-source for robot arms, but we will have another version for humanoid and 5-finger hands: https://lnkd.in/gsRArQXy - We are building a state-of-the-art Apple Vision Pro -> humanoid robot "Avatar" stack. Xiaolong Wang group’s open-source libraries laid the foundation: https://lnkd.in/gUYye7yt - Watch Jensen's keynote yesterday. He cannot hide his excitement about Project GR00T and robot foundation models! https://lnkd.in/g3hZteCG Finally, GEAR lab is hiring! We want the best roboticists in the world to join us on this moon-landing mission to solve physical AGI: https://lnkd.in/gTancpNK

Demystifying the Software Testing 1️⃣ 𝗙𝘂𝗻𝗰𝘁𝗶𝗼𝗻𝗮𝗹 𝗧𝗲𝘀𝘁𝗶𝗻𝗴: 𝗧𝗵𝗲 𝗕𝗮𝘀𝗶𝗰𝘀: Unit Testing: Isolating individual code units to ensure they work as expected. Think of it as testing each brick before building a wall. Integration Testing: Verifying how different modules work together. Imagine testing how the bricks fit into the wall. System Testing: Putting it all together, ensuring the entire system functions as designed. Now, test the whole building for stability and functionality. Acceptance Testing: The final hurdle! Here, users or stakeholders confirm the software meets their needs. Think of it as the grand opening ceremony for your building. 2️⃣ 𝗡𝗼𝗻-𝗙𝘂𝗻𝗰𝘁𝗶𝗼𝗻𝗮𝗹 𝗧𝗲𝘀𝘁𝗶𝗻𝗴: 𝗕𝗲𝘆𝗼𝗻𝗱 𝘁𝗵𝗲 𝗕𝗮𝘀𝗶𝗰𝘀: ️ Performance Testing: Assessing speed, responsiveness, and scalability under different loads. Imagine testing how many people your building can safely accommodate. Security Testing: Identifying and mitigating vulnerabilities to protect against cyberattacks. Think of it as installing security systems and testing their effectiveness. Usability Testing: Evaluating how easy and intuitive the software is to use. Imagine testing how user-friendly your building is for navigation and accessibility. 3️⃣ 𝗢𝘁𝗵𝗲𝗿 𝗧𝗲𝘀𝘁𝗶𝗻𝗴 𝗔𝘃𝗲𝗻𝘂𝗲𝘀: 𝗧𝗵𝗲 𝗦𝗽𝗲𝗰𝗶𝗮𝗹𝗶𝘇𝗲𝗱 𝗖𝗿𝗲𝘄: Regression Testing: Ensuring new changes haven't broken existing functionality. Imagine checking your building for cracks after renovations. Smoke Testing: A quick sanity check to ensure basic functionality before further testing. Think of turning on the lights and checking for basic systems functionality before a deeper inspection. Exploratory Testing: Unstructured, creative testing to uncover unexpected issues. Imagine a detective searching for hidden clues in your building. Have I overlooked anything? Please share your thoughts—your insights are priceless to me.

China has switched on the world’s first grid-connected 20 MW offshore wind turbine – the largest wind turbine currently operating anywhere in the world. Installed around 30 km offshore in China’s Fujian province, the turbine has a rotor diameter of 300 metres, nearly the height of the Eiffel Tower. Wind turbines have been getting steadily bigger for decades – driven by physics and economics: ✅ Power from wind scales with the square of the rotor diameter. ✅ Power also scales with the cube of wind speed, and taller turbines can access the stronger, steadier winds higher above the surface. ✅ Costs such as foundations and cables increase as turbines get larger, but energy production tends to grow faster than these costs. Offshore wind farms in particular benefit from scale because installation vessels are extremely expensive to operate. Reducing the total number of turbines - foundations, lifts and cable connections - can materially lower overall project costs. Larger turbines do introduce challenges, including more complex manufacturing and greater single-asset risk. But the economic advantages of larger turbines in offshore projects continue to outweigh these challenges, which is why turbine sizes keep increasing. Even larger 25–26 MW turbines are already under development – all from Chinese manufacturers. With the world’s largest domestic deployment pipeline and an integrated manufacturing ecosystem, China is increasingly setting the pace in the next generation of offshore wind turbines.

5 key developments this month in Wearable Devices supporting Digital Health ranging from current innovations to exciting future breakthroughs. And I made it all the way through without mentioning AI… until now. Oops! >> 🔘Movano Health has received FDA 510(k) clearance for its EvieMED Ring, a wearable that tracks metrics like blood oxygen, heart rate, mood, sleep, and activity. This approval enables the company to expand into remote patient monitoring, clinical trials, and post-trial management, with upcoming collaborations including a pilot study with a major payor and a clinical trial at MIT 🔘ŌURA has launched Symptom Radar, a new feature for its smart rings that analyzes heart rate, temperature, and breathing patterns to detect early signs of respiratory illness before symptoms fully develop. While it doesn’t diagnose specific conditions, it provides an “illness warning light” so users can prioritize rest and potentially recover more quickly 🔘A temporary scalp tattoo made from conductive polymers can measure brain activity without bulky electrodes or gels simplifying EEG recordings and reducing patient discomfort. Printed directly onto the head, it currently works well on bald or buzz-cut scalps, and future modifications, like specialized nozzles or robotic 'fingers', may enable use with longer hair 🔘Researchers have developed a wearable ultrasound patch that continuously and non-invasively monitors blood pressure, showing accuracy comparable to clinical devices in tests. The soft skin patch sensor could offer a simpler, more reliable alternative to traditional cuffs and invasive arterial lines, with future plans for large-scale trials and wireless, battery-powered versions 🔘According to researchers, a new generation of wearable sensors will continuously track biochemical markers such as hydration levels, electrolytes, inflammatory signals, and even viruses, from bodily fluids like sweat, saliva, tears, and breath. By providing minimally invasive data and alerting users to subtle health changes before they become critical, these devices could accelerate diagnosis, improve patient monitoring, and reduce discomfort (see image) 👇Links to related articles in comments #DigitalHealth #Wearables

Powering Cities with Every Step: Japan’s Smart Energy Innovation ⚡🚶♂️ What if your daily walk could help power your city? In Japan, it already does. Train stations, sidewalks, and bridges are being fitted with piezoelectric sensors—materials that generate electricity from movement. 🔹 How It Works – Every footstep applies pressure, creating a tiny electric charge. Multiply that by thousands of daily commuters, and it’s enough to power LED screens, lights, and signage. 🔹 Real-World Impact – Tokyo train stations track how much energy passengers generate, turning commutes into a live science experiment. Bridges capture vibrations from cars to power streetlights. 🔹 The Big Picture – While this won’t replace traditional energy sources, it’s a step toward greener, self-sustaining infrastructure. 💡 Could this technology be scaled for more cities? Where else could we harvest untapped energy? Let’s discuss! 👇 #Innovation #SustainableEnergy #SmartCities #GreenTech #FutureInfrastructure

Plastic is highly durable and resistant to decomposition. Most plastics take hundreds to thousands of years to break down, meaning that once produced, they persist in the environment for an extremely long time. What do you think about this initiative in Bali? Marine Pollution: A large proportion of plastic waste ends up in the oceans, where it poses a serious threat to marine life. Animals often mistake plastic for food, leading to ingestion and, in many cases, death. Microplastics, which are tiny plastic particles resulting from the breakdown of larger pieces, can enter the food chain, affecting not just marine species but also humans who consume seafood. Harm to Wildlife: Animals can become entangled in plastic waste, leading to injury or death. For example, plastic rings, nets, and bags are common culprits in the harm and killing of birds, fish, and other wildlife. Toxicity: Some plastics contain harmful chemicals, such as BPA (Bisphenol A) and phthalates, which can leach into the environment and potentially enter the human body, causing health issues. The incineration of plastic waste can also release toxic gases, contributing to air pollution. Carbon Footprint: The production of plastic is energy-intensive, relying heavily on fossil fuels. This contributes to greenhouse gas emissions, exacerbating climate change. How AI Can Help Address the Plastic Issue: Waste Sorting and Recycling: AI can enhance recycling processes by improving the accuracy and efficiency of waste sorting. Machine learning algorithms, combined with robotic systems, can identify and separate different types of plastic from other waste materials, increasing the volume of plastic that gets recycled. Plastic Detection in Oceans: AI-powered drones and satellite imaging can be used to detect plastic waste in oceans. By analyzing images with AI, we can better understand the scale of ocean plastic pollution and target cleanup efforts more effectively. Material Innovation: AI can accelerate the development of alternative, more sustainable materials by analyzing vast datasets of chemical compounds and predicting their properties. This can lead to the creation of biodegradable plastics or entirely new materials that have less environmental impact. Supply Chain Optimization: AI can help companies optimize their supply chains to reduce plastic use. By analyzing data on production, packaging, and transportation, AI can suggest ways to minimize plastic waste and encourage the use of sustainable alternatives. Education and Awareness: AI-driven platforms can be used to educate the public about the impacts of plastic pollution and encourage more sustainable behaviors. Personalized recommendations based on AI analysis can guide consumers to make more environmentally friendly choices, such as choosing products with less plastic packaging. #plastic #ai #technology #innovation via @sungai_design

Grid bottlenecks are a feature — not a bug — of the energy transition. For years, we viewed economics as the main hurdle to scaling clean energy. High costs for wind, solar, heat pumps, and storage dominated the conversation. But the world has changed. Thanks to extraordinary innovation and dramatic cost reductions in renewables and electrification technologies, the bottlenecks we face today are different. They’re no longer about whether clean energy is affordable — it is. Instead, the challenge is whether our energy systems can evolve quickly enough to integrate it. A recent Financial Times piece highlights this clearly: across Europe, the rapid build-out of renewable generation now outpaces the ability of grids to move electricity to where it’s needed. Curtailment, congestion, and long queues for grid connections already cost billions annually — and without decisive action, these costs will grow. This isn’t a sign of failure. It’s a sign of success. It means the transition is happening faster than the infrastructure built for the fossil era can handle. The rise of decentralised, variable renewables and electrified heating and transport requires a fundamentally different approach to planning — one that anticipates growth rather than reacts to it. The EU’s move toward more coordinated, top-down scenario building and cross-border grid planning recognises exactly this. Better alignment between countries and system operators, faster permitting, and prioritisation of critical projects are essential steps to unlock the full value of cheap clean energy. Because every euro lost to bottlenecks is not a cost of climate action — it’s a cost of not modernising our grids fast enough. The more successful we are in deploying renewables and electrification, the more urgently we must upgrade and expand our grids. Grid constraints are not a reason to slow down. They’re a reason to speed up the transformation of an energy system that was never designed for the technologies now powering our transition.

If you are working in a big tech company on ML projects, chances are you are working on some version of Continuous Integration / Continuous Deployment (CI/CD). It represents a high level of maturity in MLOps with Continuous Training (CT) at the top. This level of automation really helps ML engineers to solely focus on experimenting with new ideas while delegating repetitive tasks to engineering pipelines and minimizing human errors. On a side note, when I was working at Meta, the level of automation was of the highest degree. That was simultaneously fascinating and quite frustrating! I had spent so many years learning how to deal with ML deployment and management that I had learned to like it. I was becoming good at it, and suddenly all that work seemed meaningless as it was abstracted away in some automation. I think this is what many people are feeling when it comes to AutoML: a simple call to a "fit" function seems to replace what took years of work and experience for some people to learn. There are many ways to implement CI/CD/CT for Machine Learning but here is a typical process: - The experimental phase - The ML Engineer wants to test a new idea (let's say a new feature transformation). He modifies the code base to implement the new transformation, trains a model, and validates that the new transformation indeed yields higher performance. The resulting outcome at this point is just a piece of code that needs to be included in the master repo. - Continuous integration - The engineer then creates a Pull Request (PR) that automatically triggers unit testing (like a typical CI process) but also triggers the instantiation of the automated training pipeline to retrain the model, potentially test it through integration tests or test cases and push it to a model registry. There is a manual process for another engineer to validate the PR and performance reading of the new model.  - Continuous deployment - Activating a deployment triggers a canary deployment to make sure the model fits in a serving pipeline and runs an A/B test experiment to test it against the production model. After satisfactory results, we can propose the new model as a replacement for the production one. - Continuous training - as soon as the model enters the model registry, it deteriorates and you might want to activate recurring training right away. For example, each day the model can be further fine-tuned with the new training data of the day, deployed, and the serving pipeline is rerouted to the updated model. The Google Cloud documentation is a good read on the subject: https://lnkd.in/gA4bR77x  https://lnkd.in/g6BjrBvS ---- Receive 50 ML lessons (100 pages) when subscribing to our newsletter: TheAiEdge.io #machinelearning #datascience #artificialintelligence