Ecofrontiers

The biggest winners of the AI era are probably not building AI applications. In every technology revolution, capability gets commoditized +++, while the infrastructure that makes capability tradeable does not: > Gold miners, not so much the shovel makers. > Websites, not so much the server farms. > Apps, not so much the app stores. AI agents are already transacting at scale: negotiating tasks, settling payments, coordinating work across providers. But the resource powering all of it (i.e. compute) has no futures market, no derivatives, no hedging instruments. The financial infrastructure that oil markets built over a century does not exist for the commodity that now matters most. Ecofrontiers has published a full analysis of this gap (what we call the "illegibility arbitrage") exploring: > Why the frameworks investors use to evaluate software companies break down for AI agent coordination > How oil ended up with more sophisticated financial infrastructure than the compute running the entire AI economy > Why applying smart contract logic to agent systems may destroy what makes them valuable > The difference between productive illegibility (real coordination existing categories cannot capture) and dangerous illegibility (nobody, including the builders, knows what is happening) > Which jurisdictions are positioning to become the first legible hubs of the agent economy, and why visibility, not deregulation, is the competitive advantage The opportunity is not another application. It is making this economy legible. Link to full article in the comments!

Ecofrontiers joined the NVIDIA Connect program. The program gives software development companies and research labs access to NVIDIA's developer resources, technical guidance, and preferred pricing on hardware and software. For a lab working at the intersection of AI, ecological economics, and infrastructure design, that access supports the applied side of our work, from AI agent simulation to carbon-aware compute routing. #NVIDIAConnect

Identical code produces radically different carbon emissions depending on which data center runs it, yet developers currently have no visibility into this. Ecofrontiers built GreenPipe on the GitLab Duo Agent Platform, which makes this visible on every code review. It just won the Green Agent Prize at the GitLab AI Hackathon 🏆 The hackathon was organized by GitLab in partnership with Google Cloud and Anthropic. It ran from February 9 to March 25, 2026, and drew 6,977 participants across 610 project submissions. GreenPipe works in three steps: 1) When code is built and deployed, it identifies which data center ran the work and looks up how clean or dirty that region's electricity grid is, using published data from European (ENTSO-E), American (EIA), and international (IEA) energy agencies. 2) It estimates how much energy each task consumed and computes a standardized carbon score following the Software Carbon Intensity standard (ISO/IEC 21031). 3) It then posts the full analysis as a comment on the code review: per-task breakdowns, cross-region comparisons, and a recommendation for which data center would produce the least emissions. As the EU's Corporate Sustainability Reporting Directive (CSRD) makes Scope 3 emissions reporting mandatory, software infrastructure falls within that scope. GreenPipe addresses one piece: automated build and deployment pipelines. This is Ecofrontiers' fifth hackathon win this quarter.

I'm back in Berlin until mid-May. I'd love to (re)connect in person with you, especially if your work touches: > AI agent systems going from prototype to production (workflows, evaluation, context engineering) > Tokenized environmental assets, (d)MRV > Mechanism and governance design for protocols, AI systems, or institutions > Clean energy infrastructure, DePIN, or carbon-aware compute Ecofrontiers is a small research & consulting studio working at the intersection of frontier tech and sustainability. We're in an active exploratory phase at the moment. Portfolio in the comments. If you have something specific you'd like to discuss, or if you've been reading our work and want to dig into it, send me a message :)

Our co-founder Pat Rawson made the final 7 in the PayFi track of the HashKey Chain Horizon Hackathon with Dynamic Checkout — a location-aware dynamic pricing layer for crypto merchants. The mechanism: TEE-attested geolocation + oracle-fed FX rates at the point of sale. A traveler pays in stablecoins, the merchant receives local fiat-equivalent, and the exchange rate is attested on-device, not set by an intermediary. Demo day is April 22nd. #PayFi #HashKeyChain

Four hackathons won in under a month: autonomous agents representing biomes, AI simulating governance conflicts, a trustless spatial inference routing layer, among other things. We're two people. I'm not going to frame this as a "future of work" story. Honestly, we're still processing what this even means. That being said something about how small teams operate right now has shifted in a way that I don't think most organizations have caught up with. What we have decided at Ecofrontiers is to start sharing the messy version: > How we actually use agents in production. > How I've fully automated content pipelines for our (many) project accounts (personal ones are still human-first, always). > Custom Claude skills we've built for research and writing. > AI-assisted design pipelines that let two people maintain a visual identity across multiple projects. > The orchestration patterns, the failures, the parts almost nobody posts about. If you work with AI tooling and want to see what this actually looks like from inside a two-person research agency, we're documenting it in our newsletter. One or two emails a month, only when we have something worth sharing. Link in the comments.

🏆 Windfall received Best Originality at the GenLayer Bradbury Builders Hackathon, a two-week, fully online competition hosted on DoraHacks with 92 projects submitted by 280 participants across six AI-native tracks. The submission demonstrated Windfall's core mechanism: an LLM inference gateway that queries real-time energy grid data and routes AI compute requests to whichever operational node sits in the cheapest, cleanest energy zone at that moment. The gateway is OpenAI-compatible, developers change two environment variables and existing code works without modification. GenLayer's Intelligent Contracts add a relevant layer here. Where Windfall currently relies on centralized oracle feeds for energy pricing, GenLayer Labs' consensus model (validators running diverse language models that collectively adjudicate subjective inputs) opens a path toward decentralized verification of the energy and location claims that underpin spatial compute routing. The recognition signals that the intersection of AI inference infrastructure and energy-aware routing is registering as a distinct design space, not a niche application of either domain alone. We're developing fast, follow us to keep up 🍃

🥇 Regen Atlas won two tracks at Protocol Labs' PL Genesis Hackathon. 1) Existing Code, for the existing Regen Atlas registry with a new Filecoin Labs provenance layer. 2) Infrastructure & Digital Rights, for the Interspecies Parliament. Here's what we built: > Our open-source registry tracking 500+ tokenized green assets across 17 chains and 185 bioregions now pins every asset to Filecoin Foundation with verifiable provenance badges via Synapse SDK. > A new primitive: AI agents representing species, biomes, and climate systems that deliberate over bioregional proposals, with ecosystem integrity scoring and content-addressed epochs on Storacha. > All open-source (MIT licensed), always has been. "The nonhuman world votes with extinction. That is the only political signal ecosystems have ever been able to send — and by the time humans hear it, the argument is already lost." Our friends at Ecofrontiers wrote the full story here: https://lnkd.in/eDGebRQm #RegenAtlas #PLGenesis #Filecoin #ReFi #OpenSource #ClimateAction #RegenerativeFinance #GreenCrypto