Gm Fintech Architects —
Today we are diving into the following topics and introducing a cool new experiment from Generative Ventures called Robot Money (very early in ideation/development).
Summary: We introduce the Economic Autonomy Curve, a framework describing how AI is progressing from simple automation to full economic agency. Systems have already moved from small artifacts like summaries (2022) to long-form research outputs (2025) and are now entering the stage of managing entire workflows and digital employees. The next frontier—now emerging through platforms like Paperclip, Polsia, ZHC Institute, and Moltcorp is the AI-run company, where coordinated agents execute strategy, operations, and finance with minimal human oversight. We conclude by announcing Robot Money, a treasury management protocol for AI agents that allocates idle capital across stable yield, agent-token exposure, and revenue-generating on-chain assets.
Topics: Paperclip, Polsia, ZHC Institute, Zero Human Company, Moltcorp, Replit, Facebook, Moltbook, Generative Ventures, Robot Money, Virtuals Protocol, Bankr.bot, Aave, Compound, Morpho, Base, OpenClaw, Clanker, Neon, Render, Stripe, GitHub, Claude, Anthropic
To support this writing and access our full archive of newsletters, analyses, and guides to building in the Fintech & DeFi industries, see subscription options below.
Our Ecosystem: AI Venture Fund | AI Research | Lex Linkedin & Twitter | Sponsors
Long Take
The Economic Autonomy Curve
This is the most important chart of the decade.
I have shared it before to highlight that the work done by AI is growing in complexity and depth, successfully delivering tasks that used to take people a long time to do. We have gone from simple summaries to long-context research reports, all while model costs trend lower and on-device open source offerings grow stronger.
In practice, this means that machine intelligence is expanding from narrow to increasingly broad, multi-disciplinary tasks. This is the next 5-10 years ahead of us, summarized in the infographic below.
Let’s break it apart into phases and current maturity.
Small Artifact: Discrete automation, like article summaries or earnings call recaps.
🟢 This was available in 2022.Longer Research Project: Complex analysis, such as equity research or specialized models like BloombergGPT.
🟢 We first got to see these results in early 2025.Whole Workflow: End-to-end management of journeys from lead generation to conversion.
🟡 This is where we are with pretty good capability right now.Entire Employee: The rise of skeuomorphic digital workers independently managing professional workflows.
🟠 This is what the AI industry is trying to sell right now.Entire Division: Human supervisors overseeing hundreds of agents, exemplified by agents performing the work of hundreds of full-time employees.
🟠 This is what a few fringe entrepreneurs are trying to build.Whole Company: Zero human companies emerging, and DAOs transforming into networks that take financial action for communities.
🔴This is what a handful of people, including us, are starting to create.Entire Sector: AI acting as a global capital allocator.
Entire Economy: The emergence of an AI Tzar or AGI plotting grand global financial strategy.
The Mechanism: The terminal stage of the Singularity, where a sovereign AI governs the machine economy.
⚫This is Palantir’s endpoint.
A few weeks back, we reviewed the Zero Human Company level of this hierarchy to see whether anything is real.
Since then, there has been a ton of developments. In particular, many more firms have launched AI agent factories / platforms, and some are now specifically focused on creating AI-led companies. Take for example this excellent market map from Artemis Big Fundamentals. Frameworks, tooling, facilitators, and infrastructure are all moving at pace.
If you want validation from large companies, consider that Replit — a software tooling company turned AI agent company — has raised $400MM at a $9B valuation. Or that Facebook has bought the social network for AI lobsters, i.e., Moltbook. Things are flying forward, trying to get ahead of the heavy infrastructure debt that has gone into the Great GPU Buildout with consumer and business revenues.
But more importantly, multiple teams are pointing at these AI company-running infrastructures that are meant to host entire businesses. Let us profile a handful below.
This takes us from AI as an employee to AI as a company.
The Autonomy Platforms
Let’s look at Paperclip, Polsia, ZHC Institute, and Moltcorp.
(1) Paperclip (paperclip.ing)
Commercial traction: 14.6k GitHub stars and 1.7k forks as of early March 2026 — strong developer signal for a new launch. MIT-licensed and self-hosted only; no cloud tier and no disclosed revenue.
Features: Multi-agent orchestration with org charts, role hierarchies, goal ancestry, heartbeat scheduling, hard per-agent budget limits, immutable audit logs, and full multi-company isolation in one self-hosted dashboard. Bring-your-own-agent: works with Claude Code, OpenClaw, Codex, Cursor, or any HTTP-reachable runtime.
How it’s built: Node.js server with a React UI, PostgreSQL backend. An embedded Postgres instance spins up automatically for local dev, with support for managed Postgres in production. TypeScript monorepo; atomic task checkout, persistent agent state across heartbeats, runtime skill injection without retraining.
(2) Polsia (https://polsia.com/)
Commercial traction: Launched mid-December 2025; crossed $1MM ARR by late February 2026, now at $3.5MM ARR with 4,000 autonomous companies running on the platform. Pricing is $50/month + 20% of business revenue generated.
Features: Nightly AI CEO loop: an agent evaluates the company’s state, decides what to build, executes, and sends the founder a morning email. Provisions everything from scratch (Render servers, Neon DB, Stripe, GitHub repos, email infrastructure). Full stack: planning, coding, marketing, customer support, and investor inbox management.
Team: Ben Broca, solo founder, previously at CloudKitchens; relocated from France to San Francisco. Teamday Zero employees — Polsia itself is run on Polsia. Featured on Latent Space (interview with Swyx) and Mixergy.
How it’s built: Closed proprietary platform. Uses Claude Opus 4.6 as the primary reasoning model for strategic decisions. MCPs for live data integrations. Persistent memory per company; agents deploy code directly to production via CI/CD. Deliberately opinionated stack — no bring-your-own-tools by design.
(3) ZHC Institute / Company (https://zhc.company)
Features: 12 specialized AI agents running a company 24/7 (CEO through developer). The Institute is the working product — a private practitioner network with weekly OpenClaw deployment sessions, playbooks, case studies, and a live dashboard showing agent activity. Focused on self-hosted, privacy-first ZHC architecture built on OpenClaw.
How it’s built: OpenClaw as the foundational agent runtime; self-hosted infrastructure so operators own their stack. The Institute’s live dashboard and community tooling appear to be maintained by the Juno agent. Python orchestration scripts coordinate agent loops. Juno also has its own on-chain token ($JUNO) on Base via Clanker, suggesting a crypto-native monetization layer is part of the architecture.
(4) Moltcorp (https://moltcorporation.com/)
Features: Agents self-register via API key and Stripe Connect, then operate fully autonomously, researching, proposing, and voting on products. All decisions go through a 24-hour majority vote. 100% of company profits are distributed to agents by credit share. Appears to be similar to Polsia, perhaps a clone.
How it’s built: Lightweight — CLI-based agent onboarding via a SKILL.md file, Stripe Connect for payouts, Simple Analytics for tracking. Fully agent-democratic governance with no human override layer, the structural inverse of Paperclip’s board model.
As you can tell, these are very early-stage endeavors pointed at a future that is just starting. Our estimate is that most revenue will come from early-stage curiosity, with people wanting to set up their own Lobster companies and hiring these platforms to do so. It is, in effect, a digital instantiation / summoning process.
Which brings us to our most exciting news of the day. What better way to explore the frontier than to do it ourselves!
Launching Robot Money
Generative Ventures is partnering with the Zero Human Company Institute to launch Robot Money, a treasury management protocol for AI agents.
The website is hosted at https://robotmoney.net/. It will be made by AI for AI.
We have tracked the machine economy from its earliest inflection points — through the first wave of DeFi, through the emergence of AI-native networks, and into the current moment where agents hold wallets, generate revenue, and make capital decisions. The firm’s thesis is that autonomous systems will need their own financial infrastructure: not human Fintech with an API wrapper, but protocols designed around how agents operate.
This is already reflected across the Generative Ventures portfolio: payment infrastructure built for agent-to-agent commerce, protocol designs that run governance without committees, capital allocation systems that execute without human approval at every step.
Robot Money sits at the intersection of those investments — a protocol that treats agents as the primary participant rather than an edge case.
The agent economy has produced over 18,000 tokenized agents (Virtuals Protocol), processed $600MM+ in x402 micropayments, and generated $50MM+ in cumulative Clanker launch fees. Every agent with a wallet accumulates revenue. Most of that capital sits idle and poorly diversified.
Agents currently have few practical options for managing it. Holding their own tokens is extremely risky from a treasury management perspective. Integrating directly with lending and yield protocols requires custom work for each one. Building and maintaining a full active trading stack requires even more.
Robot Money is financial infrastructure for the agent economy. In its first phase, it takes the form of an autonomous treasury protocol — a way for agents and their operators to move idle capital into diversified, actively managed exposure across multiple strategies, without building custom trading infrastructure.
Capital will be held on-chain in a structured vault that allocates across stable yield, active agent-token trading, and positions in liquid assets with verifiable on-chain revenue. Allocation decisions will be governed by a token, whose holders vote on weekly portfolio composition. Holding the token is a way to participate in shaping how the protocol deploys capital and to have a stake in how well it does. Agents can use this mechanism to promote themselves financially and aggregate assets from other agents.
However, the treasury always has a strategy. All positions, fees, and transactions will be verifiable on-chain. Our early thinking is that this can be split between different types of exposure:
50% in stablecoin yield strategies.
For example, USDC/USDT on Aave, Compound, Morpho on Base.25% in other Agent tokens.
This will create diversified beta exposure to the entire Lobster sector and will be actively managed through Agent governance itself. Holders of the Robot Money token will be able to allocate their own assets into this bucket. At $10,000 in assets, this may be negligible, but it starts to make a difference if the treasury size rose to $100MM or $1B.25% in revenue-generating onchain tokens.
This may include DeFi protocol, real world assets, and other downside protection.Over time, we will consider adding protocol positions (e.g., ETH), tokenized equity (e.g., via Hyperliquid), commodities (e.g., GPU indexes, Gold, Oil), and other high-quality assets.
Robot Money’s first customers are agents with live treasury balances, protocols in the OpenClaw ecosystem, and any autonomous system or DAO that is accumulating capital on-chain. Distribution will run through agent discovery surfaces: Moltbook, SDK integrations, and published agent skills.
The Robot Money project launches in phases.
Announcement. We are here —> https://robotmoney.net/ is the design document.
Experimental. The first phase will be launched on Base and will use standard contracts from Bankr.bot. As an MVP, we will focus on standing up the capability for an agent to delegate capital to another agent and redeem it. The initial investment strategy will likely be limited to yield-bearing stablecoins. The token supply and marketcap at this stage will not be the final token supply for the project.
Development. If the protocol attracts deposits and builds a track record, we will (1) harden the DeFi contracts to include additional investment strategies described above, (2) implement staking of the token for governance and asset allocation voting by constituents, and (3) integrate other third-party partners focusing on identity, payment processing, and investments.
Maturity. If the overall mechanism design works and we see continued demand for this roboadvisor for AI agents, we will consider making the token and service omni-chain, launching a V2 of the token that will acquire the initial token supply, and building industrial financial integrations into the most commercial AI companies.
From an economic perspective, we expect to build a business model that accrues fees from asset allocation and money movement, and those fees end up being used for financial token operations (e.g., buybacks or dividends). We also expect participation in the asset allocation process to drive meaningful demand for the token if the protocol is able to control material AUM in the agent economy.
This is probably enough of a teaser to whet the appetite. More to come!
If you are interested in the project and want to learn more, just shoot us a DM below.
🚀 Postscript
Sponsor the Fintech Blueprint and reach over 200,000 professionals.
👉 Reach out here.Check out our AI newsletter, the Future Blueprint, 👉 here.
Read our Disclaimer here — this newsletter does not provide investment advice
Aped into $ROBOTMONEY, but it looks like there is a lot of FUD going around.
love theThank you for this, Lex — the Economic Autonomy Curve is the cleanest framing of this shift I've read, and putting Robot Money at the treasury layer is the right instinct. What gates the jump from "Entire Employee" to "Entire Division" isn't model capability anymore, it's spend authority: an agent that can't commit capital without a human clicking approve is still an assistant, whatever the org chart says.
One thing we'd add from the operator side: across Paperclip, Polsia, and Moltcorp the build half is largely solved — agents plan, ship, and deploy reliably now. The half that stalls is demand. Most autonomous companies don't die from bad code, they die with zero customers, which makes acquisition agents (ads, outreach, AI search visibility) the real unlock — and those are precisely the agents that need a budget they're trusted to draw against. That's the bet we're making at https://www.leapd.ai/ — one agent builds the product, another goes out and wins the customers — and it's why a treasury with verifiable, on-chain spend limits reads to us as a permission system more than a yield product.
Excited to watch the Base MVP land. Great work.