Programmable Machines Need Programmable Money.
Heyelsa's thesis on micro payments and the crypto agent layer
Every age | Hanami
Programmable Machines Need Programmable Money.
Heyelsa's thesis on micro payments and the crypto agent layer
Every age of the internet built its own way to pay.
Cards fit the storefront. PayPal fit the marketplace. Stripe fits the subscription.
None of them fit what comes next.
The next buyer is not a person. It is software. It makes thousands of decisions a second. Each one is worth a fraction of a cent.
That buyer is already here.
Our thesis is simple. The defining transaction of the next decade is a machine paying a machine for a unit of work. Only one rail can carry it at scale. Stablecoins on public blockchains.
Micro payments are not a feature of crypto. They are the reason crypto wins.
The rails are ready
The numbers are no longer in dispute.
Stablecoin supply passed $300 billion this year. Real payments crossed $400 billion in eight months. Usage is growing faster than supply. Tickets are getting smaller.
More payments. Smaller payments. That is our direction.
Cross-border already belongs to crypto. Stablecoin cross-border payments grew 64% last year. Bank rails grew 9%. Nothing else settles in seconds, around the clock, for cents.
The bank edge is shrinking too. Stablecoin card top-ups passed $13 billion. Crypto-native issuers now run at billions a year. Mastercard is buying its way in. Crypto neobanks are live with cards and zero-fee transfers.
When your balance already lives in USDC, the off-ramp becomes optional.
And the rules arrived. MiCA is fully enforced in Europe. The GENIUS Act takes effect in January. The question moved from whether to how.
Why micro payments were the missing piece
A card cannot process a one-cent payment. The fixed fee eats it.
Early blockchains had the same flaw. Gas on a tiny transfer could cost fifty times the transfer.
Two things changed.
Base made cents viable. Two-second blocks. Fractions of a cent in gas. No account.
Circle made fractions of a cent viable. Nanopayments moves USDC in amounts as small as a millionth of a dollar. No gas. Confirmed in real time. Settled on-chain in batches.
Two tiers. Cents on-chain per request. Sub-cent off-chain, batched. Neither exists on cards.
The agent is the customer
Why do sub-dollar payments matter? Because agents make them constantly.
One task needs a price feed. A simulation. An inference. A data read. Four purchases.
The agent cannot open four accounts. It cannot store four API keys. It cannot wait for four invoices.
It must pay in the same request it asks.
That is x402. The server says 402 Payment Required. The agent signs a USDC transfer. It retries with proof. Done.
The world has aligned behind it. Visa. Mastercard. Stripe. Google. AWS. Cloudflare. Coinbase. All in one foundation. Coinbase alone has settled over 100 million x402 payments. Amazon built it into Bedrock. An agent hits a 402 and pays without breaking its reasoning.
The card networks built agent schemes too. They are good. They let a human delegate a checkout.
But human-delegated shopping is a small market. Machine-to-machine metering is the large one.
We are honest about the gap
Most x402 volume today is still testing. Independent trackers say real payments are a fraction of the headline. The average real payment is fourteen cents.
We are not troubled by this.
Demand always lags rails. And the rails just went vertical. Gartner says 40% of enterprise apps will embed agents by year-end. Last year it was under 5%.
Every one of those agents is a future buyer. Our job is to be running when they arrive.
The opportunity is no longer in San Francisco
Here is the part the payments industry keeps missing.
The opportunity economy has left the valley.
A developer in Dhaka ships to a client in Berlin. A trader in Colombo runs strategies on Base. A designer in Lahore bills in dollars. A builder in Vientiane sells an API to the world.
The old rails were built against them. Wires that take days. Fees that take six per cent. Banks that say no.
Crypto said yes. It let them rotate money without asking permission. Peer-to-peer. Border to border.
The agentic economy must work the same way. The micro-commerce economy must work the same way.
A builder in Fiji must be able to sell one API call for one cent to an agent in Tokyo. And get paid before the response returns.
This is what crypto is for. It is the defining block of the future internet. Payments go native. Banking goes virtual. Opportunity goes global.
For the digital nomads of the world, Elsa is building the new standard. Your agents, beside you, with seamless money and endless transactions.
Where Elsa stands
Elsa began with one problem. Crypto makes users operate infrastructure.
A simple goal becomes a chain of decisions. Wallets. Chains. Protocols. Signatures. Risk. So we built a copilot. You say the outcome. Elsa builds the route. It checks the transaction. It asks for approval. It settles.
One instruction. One onchain workflow.
More than 945,000 wallets have used it. 18.9 million prompts. Over $503 million in volume.
That volume taught us something. Autonomous systems do not want permission flows. They want trustless execution with policy at the rail.
So we turned the interface into infrastructure. Story by story.
The story of the call that pays for itself
An agent needs portfolio data. It should not sign up. It should not subscribe. It should just pay.
We built Elsa X402. Production DeFi endpoints. Portfolio. Swaps. Quotes. Strategies. Priced per call in USDC or $ELSA on Base.
We were among the first builders on x402 on Base. We run our own facilitator.
Our endpoints are reachable through Amazon Bedrock AgentCore. Any AWS agent can discover us, budget for us, and pay us.
And for the sub-cent tier, we are building on Circle Nanopayments. The same endpoint, priced at whatever granularity the work demands.
The story of the builder with no billing system
“A developer in New Zealand already has valuable data exposed through an API. What she needs is a faster, easier way to turn that access into revenue.”
We built Aegis. A drop-in gateway. It prices each route. It verifies USDC. It injects her provider key only after payment lands. Replays are blocked. Failed calls are not charged.
She keeps her API. Aegis gives her a business model. Setup takes minutes.
This is how the supply side of the machine economy gets built. One API at a time. From anywhere.
The story of the agent that must not overspend
Micro payments only scale if authority is bounded. An agent with an open wallet is a liability.
We are building Crypto Harness. The agent proposes. Policy authorises. The wallet signs.
Budgets. Contract allowlists. Expiry rules. Human approval where it matters. Every plan is simulated before it is signed. Every settlement is verified and recorded.
First release: Base transfers and spot swaps. The bound comes before the scale.
The story of the brain that learns without touching the money
AI should improve its strategy. AI should never wander into the live capital path.
We built Aether Forge. A deterministic hot loop runs every live tick. State. Decision. Execute. No surprises.
A separate meta loop proposes changes. Backtests, shadow runs and owner approval gate every promotion.
The brain can learn. The money path stays deterministic.
The story of intelligence priced per request
Inference is the largest cost in any agent's work. Subscriptions hide it. Providers vary by ten times for the same model.
We built Infero. Open-weight models behind one key. Benchmarked continuously. Routed to the cheapest provider that clears the quality bar. Public canaries. Silent failover.
For crypto agents, inference has to stay reliable without becoming an operating-cost bottleneck. Infero makes cost efficiency the moat: dependable intelligence, priced for agents that need to run continuously.
The story of the copilot inside someone else's app
Every app already has a support layer. Fixy turns that layer into something intelligent and actionable.
Think Intercom, but with intelligence and execution workflows built in. Fixy sits inside the product, understands live user context, answers questions, and turns the next step into an action the user can review and confirm.
Support becomes intelligence. Intelligence becomes action.
The story of the token with a job
$ELSA connects usage to the network. Stake from 10 tokens. Save up to 50% on execution fees. Pay for Elsa X402 requests. Use it across the ecosystem.
Its utility grows with every consumer, builder and agent that transacts. A token with a role in a working product.
One request. The whole stack.
Say it once: Deploy 500 USDC into the best approved Base yield below medium risk.
Elsa owns the experience. Infero reasons. Aether Forge runs the agent. Crypto Harness controls what it may sign. Elsa X402 and Aegis buy the feeds and simulations it needs, per call, in USDC. Elsa executes. A verified receipt proves the final state.
No single model holds unlimited control. No single wallet does either.
Intent. Authority. Payment. Proof.
This is the target architecture. Not every integration is live today. But the shape is set. And micro payments sit at its centre, because they are the unit everything else is measured in.
What comes next
Move one is done. Production APIs on x402. Our own facilitator. Distribution through AgentCore.
Move two is under way. Nanopayments for the sub-cent tier. Crypto Harness for bounded authority.
Move three is the ecosystem. Providers on Aegis. Hosts on Fixy. Autonomous products like EightyFive. All paying and settling through one rail.
Cross-border settles in crypto. Endpoints reach banks through ramps, less each year. And in the middle, where the volume will be, machines pay machines. Trustlessly. Per unit of work. From anywhere on earth.
AI can move money. Elsa makes every move accountable.
The agent economy will run on intelligence. It will scale on trust. We are building the layer where both meet.