This year I spent most of my time building at the intersection of AI and Web3, with Cursor writing code alongside me and assistants shipping inside client products, and the financial infrastructure that seemed perpetually "almost there" started arriving at the same time the AI tooling became dependable enough for production work.
Agents inside shipped products
At LegalAgent this year I added a voice and chat assistant to their React Native Expo app. Chat runs on the Vercel AI SDK, voice runs on OpenAI's Realtime API, and RAG supplies case context, document summaries, and procedural guidance in Spanish and English, while an internal admin built with TanStack Start keeps system prompts and document categories editable without a client release.
What changed for me is how much of that work is now integration and design rather than model plumbing. Cursor and similar AI-powered IDEs have changed how I write code, and RAG pipelines on Supabase and vector databases put domain knowledge within reach of an ordinary application team, so the hard part moves to deciding what the assistant is allowed to do and how the product recovers when it is wrong.
Conversational interfaces for financial actions
On Bitlauncher, the batch auctions ran on Gnosis auction contracts deployed to EOS EVM, and the chatbot at bitlauncher.ai used tool calls for current news and YouTube content and RAG for longer answers about projects. That pairing is where financial interfaces are heading: a user describes an intent such as "swap 1000 USDC for ETH at best price over the next hour," and an agent turns it into an execution strategy, gas decisions, and batched transactions that the user approves.
The pieces for that are in place: vector databases for semantic search, function calling for deterministic actions, and wallet integration for transaction signing. I laid out the architecture in Agentic Finance, and the constraint that still matters most is where the agent's authority ends.
Stablecoins as payment infrastructure
After Stablecoin Summer, the regulatory environment is clarifying, and U.S. policy is shifting from skepticism to support that treats dollar-backed stablecoins as an extension of American financial infrastructure rather than a threat to it.
Living in Costa Rica, I see this from the receiving end. An international bank transfer takes days and costs me 3–5% in fees, while a USDC transfer arrives in seconds for a few cents, and that gap changes how I move money.
Latin America doesn't need to rebuild correspondent banking networks or payment processing to get there, because it can adopt stablecoins directly. El Salvador tried the direct route with Bitcoin, and the next phase is stablecoin-native services for payroll, remittances, savings, and lending on transparent, programmable rails.
Tokenized venues running in parallel
Traditional financial institutions are launching tokenized trading venues as production infrastructure with regulatory approval rather than as pilots. The NYSE is building parallel digital venues that run 24/7, settle instantly, and use stablecoin rails, and those venues are new markets built on different assumptions rather than retrofits of the existing ones.
The transition strategy is to run both systems side by side, legacy markets with T+1 settlement and limited hours next to digital venues with instant settlement and continuous operation, and let capital choose its rails. What makes the digital side work is programmable ownership combined with settlement that happens at the same moment as the trade, so there is no counterparty window and no wait for banks to clear wires.
Custody moves too, from central clearing houses toward wallets. On Bitcash I built WebAuthn key management for in-browser signing and added EOS contract multisig, and on Bitlauncher the wallet showed EVM and EOS balances together across a token bridge; those same primitives apply to tokenized equities, where custody becomes something developers build on instead of a service that requires institutional relationships.
Compliance follows the same path, with transfer restrictions enforced in the smart contract, KYC attached to wallet credentials, and accreditation checked before a transaction executes. The regulations stay, and enforcement becomes code instead of manual process. Once assets sit in a wallet and settle in stablecoins, participants no longer need correspondent banking relationships, although securities law still applies to them.
That opens composability between tokenized securities and DeFi: securities as collateral in lending protocols, automated market makers for equity liquidity, and cross-chain bridges for portability, all built from primitives that already work for crypto assets.
Wallet signing without seed phrases
Bitcash had the constraint that users had to sign transactions in the browser without an external wallet, so I built WebAuthn key management for it, and QR signature requests over Supabase subscriptions later let bitlauncher.ai and other partner apps authenticate and sign through the Bitcash session instead of asking for a browser extension.
That is the direction for wallets generally: smart contract wallets with session keys, account abstraction for gas sponsorship, and agents that translate intent into transaction sequences. ERC-4337, passkeys, Safe multisig contracts, and JSON-RPC interfaces already exist, and the remaining work is integration, because every wallet has its own signing methods, contract interfaces, and error handling, which is why standards, tooling, and developer documentation matter more now than new primitives.
DeFi as backend infrastructure
DeFi is becoming programmable financial infrastructure: composable primitives for lending, swapping, derivatives, and yield behind standardized smart contract interfaces. Institutional interest follows from resilience, since the protocols that survived the 2022 collapse have been through several stress tests, and the ones gaining traction now have sustainable economics and clearer regulatory paths.
I treat DeFi as the backend layer for the rest of this stack, because AI agents need programmatic access to financial services, stablecoin applications need yield and liquidity, and wallets need protocols to interact with.
What I'm building toward in 2026
Each of those layers reached production on its own in 2025: agents that run multi-step workflows, stablecoins with regulatory clarity, tokenized venues with institutional backing, wallets that keep self-custody usable, and audited DeFi protocols. In 2026 I expect to spend most of my time on the boundaries between them, meaning who signs, which rails settle, and what an agent is allowed to execute on a user's behalf.