Building agent infrastructure: A recap from Union Hall Cast

Last week, our founder Remy joined Karel from Union Network and Brent from Namada for an in-depth conversation on a Union Hall Cast episode about the real work of building agent infrastructure.

Instead of speculating on the next AI trend, the focus was on what it actually takes to build: orchestrating autonomous agents, protecting their privacy, securing AI workflows, and connecting them to real-world value flows.

We've pulled highlights from our segment covering:

• Flora's debut at ETHCC Cannes
• Why AI agents need privacy and our modular approach to inference and transaction privacy
• The critical challenge of securing AI development workflows
• Moving beyond memecoins to real agent utility with Dispatch
• Why stablecoins are essential for agent-driven value transfers
• What's next for Flora

Flora’s public unveiling at ETHCC Cannes

The conversation kicked off with Karel asking about our recent keynote at ETHCC in Cannes, where we officially unveiled Flora after months of quiet development.

"Flora has been under wraps for a while. We've been socializing behind the scenes, and now we're ready to share an orchestration layer focused on AI agents."

Remy Carpinito, Cofounder & CEO of Flora

This debut also marked the start of our Dispatch integration — a system for on-chain event triggers and automations that will power parts of Flora’s agent infrastructure. From here on out, we’re building in public.

Key takeaways

• Stealth → Vision unveiled: We pulled back the curtain on Flora’s orchestration layer after months of building behind the scenes.
• Cross-chain vision: Flora is being built as a coordination and execution layer that connects multiple blockchains, ecosystems, and off-chain systems.
• Dispatch on deck: We’re working on integrating Dispatch to enable agent-driven on-chain triggers. More updates coming as we build in public.

Do AI agents need privacy?

When Karel brought up “verifiable agents” and asked whether AI systems need built-in privacy, Remy shared some of the thinking that’s guiding Flora’s early design conversations: Agents will likely need control over both how they think and how they transact.

One of the directions we’re exploring is a modular privacy system that developers can adjust based on each agent’s role. For example, payment agents might use shielded pools for private transactions, while inference-heavy agents could lean on private reasoning.

"Whether you need shielded pools for transactions or private inference, you'll be able to enable exactly the privacy you need. We're lining up those partners now, and you'll see it in our upcoming testnets."

This reflects where the team’s thinking is today — not a locked-in roadmap. We’re looking at ways to work with existing privacy protocols from both Web3 and Web2 so builders won’t have to start from scratch.

Key takeaways

• Design exploration: We’re thinking through how agents can control both their computations and transactions.
• Modular privacy is on the table: Developers may be able to select different privacy tools depending on the agent’s role.
• Work in progress: We’re in conversation with potential partners, but nothing is set in stone.

Securing your AI tooling

The conversation shifted to a common problem teams face when working with AI tools: accidentally exposing sensitive data to public models. It’s more common than you’d think — developers sometimes send entire codebases or confidential files to external AI endpoints without realizing the risks.

At Flora, this is part of a broader challenge we’re thinking about: How can developers use different AI models — like Google Gemini, Anthropic’s Claude, open-source models, or local inference — without losing control of where their data goes?

One of the ideas we’re exploring is building better visibility into data flow. Developers should know when they’re about to expose sensitive information to public services and have simple ways to route private work through safer channels.

Key takeaways

• Data control matters: Teams need ways to avoid leaking sensitive information to public AI endpoints.
• Model flexibility is part of the vision: We’re exploring how to support multiple models while maintaining control over data flow.
• Visibility first: We’re thinking about ways to help developers spot risks early.

Beyond memecoins: Real agent utility

A lot of what's being called “AI agents” in crypto today still relies on human input behind the scenes. At Flora, we’re thinking about what it will take to move beyond that to enable agents that actually operate on their own across systems.

Dispatch started as a simple “if this, then that” automation layer. Now we’re exploring ways to make it smarter by incorporating AI models that let agents handle real-time decisions and adapt to changing conditions — things you can’t fully pre-program.

The longer-term goal is to support agents that can work across both Web3 and Web2 ecosystems. That means interacting with smart contracts, APIs, databases, and other tools companies already use — not just staying inside crypto.

Key takeaways

• Smarter automation is the goal: We’re exploring how to move beyond basic triggers toward agents that handle real-world conditions.
• Cross-system orchestration: The vision is to build agents that work across Web3 and Web2.

Stablecoin rails: The future of agent payments

An important part of building agents is often overlooked: how they handle money. While payments are easy for humans, they’re more complicated for autonomous systems. That’s where stablecoins could unlock real progress.

We’re exploring how stablecoins could become part of the agent economy — giving agents a reliable, programmable way to send and receive value.

Another challenge we’re thinking about is secure key management. Agents will need a safe way to control their funds and execute actions autonomously — whether they’re running locally or in the cloud. This is a complex problem, and it’s a priority area for ongoing R&D.

Key takeaways

• Payments are core to autonomy: Agents will need ways to handle payments reliably and securely.
• Stablecoins make sense: We’re exploring how stablecoins could power agent-driven payments, from simple transfers to complex workflows.
• Security matters: Key management for agents is an open problem we’re actively researching.

What’s next for Flora

The conversation closed with a look ahead — both at the broader AI ecosystem and where Flora is headed next. One of the big open questions: Can open-source AI models catch up to proprietary systems, and how will developers build with them safely?

At Flora, this is something we’re actively thinking about. Here are a few areas we’re currently exploring:

• Smarter AI interactions: We’re experimenting with ways to make agent interactions feel more natural — moving beyond basic prompts and toward richer, context-aware orchestration.
• Unified developer tools: We’re thinking about how to help developers work with different AI models without sacrificing security or data control.
• Public DevNet: We’re planning a public DevNet where builders can experiment with Flora’s core infrastructure and help shape what comes next.

Building the future of agent orchestration

This Union Hall Cast gave us a chance to share more about Flora’s direction and the infrastructure we’re building. We’re continuing to develop in the open, and the DevNet will be one of the next steps where builders can get involved.