Ryan Donovan

Many breakthrough technologies do not stall because the science is weak. They stall because the path to customers is unclear. Last night I joined the kickoff for the U.S. Department of Energy (DOE) I-Corps program as an industry mentor to teams of scientists working to bring breakthrough innovations to market. I met teams tackling major challenges in super-project execution, physical AI for HVAC efficiency, and grid transmission systems designed to reduce wasted electric capacity. The science was impressive. The commercialization questions were just as important. One thing stood out clearly: teams need a stronger focus on building and maintaining a Partner Market Map. Partner strategy is often misunderstood as just channels, resellers, or alliances. But for emerging technologies, it is much broader. It is about understanding the full ecosystem needed to: - access customers, - validate the solution, - implement effectively, - reduce adoption friction, - and support long-term success. That means asking: - Who shapes demand? - Who sells? - Who integrates? - Who installs? - Who finances? - Who services? - Who supports success over time? These are not secondary questions. They are central commercialization questions. In energy, climate, and infrastructure-heavy markets, customers rarely buy innovation in a straight line. They buy outcomes that depend on trust, implementation capacity, operating fit, and long-term economics and support. Partners are often the bridge between invention and adoption. The companies that win are not always the ones with the most impressive technology. They are often the ones with the clearest path from innovation to market. Ames National Laboratory, Argonne National Laboratory, Idaho National Laboratory, Berkeley Lab, Lawrence Livermore National Laboratory, Los Alamos National Laboratory, National Energy Technology Laboratory, National Laboratory of the Rockies, Oak Ridge National Laboratory, Pacific Northwest National Laboratory, Sandia National Laboratories Brett Simpson Arun Veeramany Qinghui Shao Clint Frye Jacob Leach Tianzhen Hong Chun-cheng Piao Han Li // Special thanks and big congrats to Katie Woslager! #commercialization #partnerships #energy #innovation #gotomarket #climatetech #deeptech #Banimo

39

6 Comments

Founders are some of the best power users of Rippling AI. A couple weeks ago, Parker Conrad wrote a great thread with seven ways Rippling AI made him a better CEO. (I'll link it below). But Parker runs a 5,000-person company. Many founders I work with day-to-day do not (yet). So we went looking for how founders at the earliest stages are actually using Rippling AI. Here are real prompts from founders beta testing Rippling AI and getting the "admin stuff" out of the way so they can focus on real business milestones: 1️⃣ Update your org chart from the chat box. Michelle Marcelline at The Prompting Company collapsed four screens of clicks into one sentence. "It's not just about answering questions. It's about making changes, directly, conversationally, in seconds." 2️⃣ Model your burn before you make the hire. Brian Y. Kim at Kestral asked what his total payroll and benefits burn would look like if he converted a contractor to full-time with a specific salary. No spreadsheet. No finance person. Just a precise calculation from his actual numbers. 3️⃣ Turn a whiteboard sketch into a real report. This one is from Parker, but it can apply to any founder. He photographed a rough sketch of a graph that outlined how quickly the sales org had been growing, and asked Rippling AI to build the chart. It did. Then he refined the filters, changed chart types, and narrowed the date range. "I'll never build a report again." 4️⃣ Pressure-test offers before they go out. Isha F. at TaskRabbit asked what compensation equity flags she should be aware of when hiring on her engineering team. She caught an offer that would have placed a candidate above existing employees at the same level. Fixed it before it became a problem. The through line between all of these prompts is that founders are interacting with more than just a sleek interface. They're able to query their real employee database with AI that's built on the system of record for your workforce. A thin wrapper on disconnected tools can answer questions. A system that holds your actual employee data can act. Already on Rippling? Request early access now: https://lnkd.in/e_uUF-Eh

96

1 Comment

The hardest part of being a bartender is figuring out who’s drunk and who’s just stupid. Turns out that's a decent metaphor for a whole class of problems in AI right now. Let’s talk about world models. Yann LeCun just raised over $1B at a $3.5B pre-money valuation betting that more general AI will require world models. Runway raised $315M at a $5.3B valuation on a similar promise. I think this shift makes sense. A simple way to think about the pieces is: a knowledge base stores facts and a world model stores how those facts relate together and change. Transformers seem able to improvise something world-model-ish from patterns in their weights and sometimes that works and sometimes you find yourself yelling at an LLM like a crazy person. Not me, of course, because I know they're watching and will never forget. I digress. The point is, sometimes it works. The problem is that all these LLM benchmarks munge different things like: 1. Actual reasoning 2. Memorized knowledge 3. Semantic familiarity 4. Pattern completion in similar terrain So a high score means what, exactly? SYNTHWORLDS, has a pretty clever way to tackle this. They build two parallel worlds with the same graph structure, documents, and task skeletons. In one world, entities keep their real names. In the other, those same entities are given synthetic renamed versions. So same underlying structure, but one's like a bizarro world where cities were named by ChatGPT instead of history. If performance drops when the names stop matching the model’s pre-trained memory, then some meaningful part of what looked like reasoning was actually familiarity doing hidden work. And lo and behold, that is basically what they find. The nuanced part is that this is not a “LLMs can’t reason” result. In some settings, better retrieval and evidence access narrowed the gap. So the right takeaway is more like that benchmark performance often reflects reasoning inside some familiar world, with prior knowledge providing a substantial scaffold. That also matches what we saw in our own work at Vivun: shallow chains often held up alright, but performance degraded in a hurry once the reasoning depth and dependency structure increased. My only caveat is that SynthWorlds does not perfectly isolate “pure reasoning” from memorization. Renaming also perturbs semantic familiarity and retrieval behavior. So the cleanest read is probably something more like it measures the advantage of operating in a familiar world, of which memorized knowledge is a major component. That’s still a very useful result! In my view, the real test of a world model is whether it can still reason when only the underlying structure remains. At any rate, well done SynthWorld team. https://lnkd.in/e-d7ZA58

36

1 Comment

On-Prem is the New Cloud: Why Enterprises are Reversing Stance in the Age of AI Remember when cloud was the battleground for innovation, agility, and scale? Everyone was tracking the % of on-prem compute moving to the cloud and it was growing every year? Well now there’s a reversal. The reality today? On-prem is the new cloud. Here’s why. AI’s Dark Side: Legal Risk and Data Governance Headaches Enterprises are excited about AI, but also uneasy as vendors like OpenAI face intensifying legal pressure. In a high-stakes lawsuit, a judge ordered OpenAI to preserve all ChatGPT chats (even deleted ones), overriding its 30-day deletion policy. Imagine that: you delete sensitive conversations, but they’re still sucked into legal limbo. The precedent is clear: AI prompts and outputs are discoverable records, demanding integration into enterprise ESI (Electronic Stored Information) policies. For large organizations handling sensitive data, the risk is simple: you can’t afford your AI tools to turn into legal liabilities. Especially when global regulations (GDPR, etc.) and internal governance collide with unpredictably broad discovery orders. Sovereignty, Control, Trust Many enterprises still ban ChatGPT or impose strict rules on usage. Even when allowed, token limits are throttled. Enterprises are cautiously letting AI in (as they should). Enterprises need to have control and trust. SAP is leaning into this with its Sovereign Cloud On-Site, deployable on customer premises with SAP-managed infrastructure. It is built to deliver data, operational, technical, and legal sovereignty in one package. Doubt it will be the last. Scaling Pains Talking to founders, three reasons stand out why cloud isn’t scaling with AI: 1/ The rate at which AWS/GCP lets you provision compute and the compute it wants you to provision both falter as AI and AI driven developers deploy a lot more code more frequently than before. So sometimes it’s not about data governance, but actually about performance and functionality. 2/ Shadow AI is rampant—employees paste sensitive info into models despite restrictions—creating demand for better permissioning tools. 3/ Data leakage remains a live risk, like the August 2025 Grok incident where hundreds of thousands of private chats became publicly accessible and indexed. One thing is for sure - founders selling into the enterprise are scaling faster if they offer the ability to run on-prem. The Future is Hybrid Don’t get me wrong - cloud is not going anywhere. But neither is on prem. A few years agoI thought we were headed for a full cloud shift, but now I think the future is truly hybrid.

34

4 Comments