Soroush J. Pour

I’ve been working on the theory of why small, shippable factories are the perfect inversion to the centralized prefab factory model. It all collapsed into one line of math (credit to Gilles Retsin): Factory Efficiency = Usage × (Throughput × Product Value ) / ( CapEx + Deployment OpEx) The idea is that the closer you can drive Usage to 100 % while keeping the denominator tiny, the faster the flywheel spins.  A shippable microfactory tackles this by primarily focusing on a high Usage factor. Because the microfactory can be packed up afterwards and reused, its CapEx gets amortized over multiple projects, effectively raising its overall Usage across the year. If one project alone doesn’t fully occupy the factory, it can simply roll over to another job to maintain high Usage. Its CapEx is often an order of magnitude lower than a centralized factory (< $1M vs. $10-50M min), so the breakeven throughput is more achievable on a small pipeline of work. The real breakthrough here is portability. There is near-zero stranded capital and almost no idle time. To flesh this theory out, I wrote 3000 words on the topic for Brad Hargreaves and Thesis Driven. The article breaks down: - why “fixed factories, shipped goods” is being inverted to shipped factories, fixed goods - what a sub‑$1 M robotic cell does to CapEx per home versus a £45 M off‑site plant - early production/cost data on a shippable microfactory from Mollie Claypool, Gilles Retsin, Sam Baker and the Automated Architecture (AUAR) team - the hybrid reality of robots tackling the heavy structural components and human crews assembling faster and cheaper - technical constraints to pull this model off and future considerations on the optimal business model (own vs. rent) Major shoutout to Gilles Retsin for helping me pull this together. https://lnkd.in/eRTnbZPa

54

2 Comments

With RL techniques like GRPO, it’s become easy to fine-tune small models to outperform foundation models on vertical tasks. We’re open sourcing Osmosis-Apply-1.7B: a small model that merges code (similar to Cursor’s instant apply) better than foundation models. Links in comments! Using foundation models for high specificity, low complexity tasks like code merging can be overkill - instead, it’s better to use a specialized model instead for higher performance and latency. We fine-tuned the model on 100K real-world code commits where the model needs to merge a code edit snippet into an original code snippet. On the test dataset (subset of CommitPackFT), foundation model performance ranged between 0.77-0.93 reward score. In comparison, Osmosis-Apply-1.7B achieved a 0.98 reward score - while also being 3X-5X more affordable than the next cheaper model AND ~10X faster! Why this matters: the best AI agents use multiple models to ensure task specialization and reliability at scale - on top of cost and latency benefits as well. Links to the model and full write-up (with reward function, hyperparameters, etc.) are in the comments - reach out if you’re interested in using reinforcement learning!

240

18 Comments

Odin Dynamics, Inc write-up by Matthew Bernard. The story of how Amir and Jon came together to build something of National importance. It's worth the read. https://lnkd.in/eXq8zySA Tareyton Venture Partners

14

1 Comment

Forbes published my latest article on what many investors may be getting wrong about robotics. Humanoid robots look like the future. But investing in humanoid hardware may not be. Manufacturing asymmetry is real. Competing with Asian supply chains is not a typical venture risk profile. Outside of companies like Tesla or Amazon, most hardware startups, even unicorns, will face extreme pressure. The long term U.S. advantage is far more likely to sit in embodied AI software and the intelligence layer, not in assembling robot bodies. In robotics, the control point will not be the factory. It will be the mind. https://lnkd.in/gmcVtCa5

93

15 Comments

Skilled labor is the bottleneck in industrial ops. Elevāt's platform basically tries to “productize the expert” using machine signals + docs + prior fixes. Now they have $12M to scale. Congrats to them. In industrial AI, the tech is usually the easy part....adoption is the battle. Curious what you think is the real gating factor:  -- trust (techs believe it’s right and won’t get them hurt),  -- workflow integration (it fits into CMMS/ticketing/parts + works offline/on-device), OR  -- incentives (the field team is rewarded for using it vs “just fixing it their way”). Which one kills deployments most often in your experience?

1

The next frontier of AI isn't just about choosing the right model. It's about building moats through implementation. Across the Emergence Capital portfolio, companies like Bland and Federato are scaling faster by embedding Forward-Deployed Engineers. These engineers bridge the gap between AI and reality, ensuring real results in production. Read more from my colleagues Gordon and Kyle on why FDEs are the key to unlocking AI's full value. https://lnkd.in/gcsRRGX2

29

3 Comments

Why That Feedback Loop Is Hard At First Rays Venture Partners, we have a thesis around the fact that current tooling in AI Code reviewers falls short in reliably surfacing who should act on AI feedback and why, especially as AI begins generating significant portions of code instead of just assisting. We feel that despite all the hype in the category, the problem of taking code safely into production remains a challenge. Here is why: 1. High Velocity + AI Agents = Attribution Challenges When multiple AI agents generate or review code concurrently, it becomes harder to attribute: Which developer owns the change, which AI suggestion is accurate, who should be responsible for fixing flawed suggestions. AI accelerates production but doesn’t inherently answer “Who is accountable for this?” without human context. 2. Lack of Deep Context Signals for AI AI review tools look at code diffs and patterns, but they often lack: Product intent, Feature specifications, Business logic rationale. Human reviewers provide these context signals — and attributing suggestions back to people with domain knowledge remains difficult for AI. 3. Closing the Loop Requires Better Human-AI Interaction Most tools focus on discussing AI feedback directly within the PR context, asking follow-up questions about suggestions and clarifying rationale rather than just generating static comments. But this still requires developers to interpret and act on AI outputs. One of our portfolio companies Arnica is actively working on this problem. Nir Valtman & Eran Medan. If you are #CTO or a #devleader, do you think you organization is getting value from the AI code review tools or are you looking for something more? We would love to hear from you. cc: Alok Nandan, PhD

19

5 Comments