Arweave TX
Paul Cowgill
I used Nash to write this. I asked it to write "a compelling founder story about our journey through the idea maze from initial concepts to launching nash", by looking at my Farcaster casts in the channel /the-arena using the Neynar API.
It tells the backstory leading up to our launch of Nash this week:
Everything that follows is written by AI, using old casts I wrote as source material.
Our journey to Nash began with a different project. In June 2024, we were building Once Upon, a modern API for onchain data designed for consumer crypto developers.
Through the summer, we shipped a series of features and tools:
Contract.cool, an app for finding trending contracts in your Farcaster network
Contextualizations for NFTs
A self-serve developer dashboard with API docs
The Once Upon UI kit, which we planned to open-source
New API endpoints for transactions and trending contracts
Self-serve Stripe payments for the API
We were making steady progress, but something wasn't clicking. The market signals weren't as strong as we'd hoped.
On August 9, 2024, we made a difficult but necessary decision:
"this week we made a hard choice, and we shut down products we've shipped in the past: the block explorer and the api."
This was a pivotal moment that freed us to explore new directions. We didn't waste any time. Just a week later:
"building something 🆕 with the /onceupon core team"
We quickly set up a new tech stack and developed a prototype for "a protocol that makes markets for onchain events." By August 23, we had built out both the protocol and app for betting on onchain outcomes, and by the end of the month, we had named it "Static."
In early September, we pivoted again:
"building a group payments protocol for multiplayer games"
We moved fast, shipping a first draft of the contracts, UI, and API within a week. We named it "Cove" and built an end-to-end group payments flow for admins and members.
The real test came on September 21:
"we ran a poker game on base with usdc powered by cove!"
We continued to iterate on Cove through early October, building a mobile-first UI, setting up Kafka+websockets, and switching to embedded wallets. We even ran playtests of a game built on top of the Cove protocol.
But we were still exploring. In mid-October, we shipped "an app for minting contracts as onchain svg nfts" and by October 18, we were working on two ideas simultaneously:
"personal finance onchain" and "easy ai agents for memecoin trading"
The second idea proved to be the seed of something bigger.
The pivotal moment came on October 25:
"building /nash, ai agents for transacting onchain. this week we:
* did lots of user interviews
* decided to focus 100% of our efforts on /nash
* iterated on the prototype"
After months of exploration, we had found our direction. The user interviews confirmed we were onto something with real demand. We were no longer hedging our bets across multiple ideas—Nash had our full attention.
November saw rapid progress as we built the core of Nash:
"a web app that lets you configure ai agents, deploy them with a click, then watch what they do. under the hood, it's powered by a new api plus a worker that orchestrates docker containers"
We fine-tuned agent behavior, created a smoother setup flow, added support for custom tools, and switched to a better agent framework. We were getting "closer to magical onchain transactions."
A key insight came in mid-November:
"realized that since getting agents just right on the first try is hard, starting from what someone else did is powerful. nash agents should be social. we're excited about people sharing and forking configurations."
This led us to overhaul the creation flow to focus on templates and set up agent observability. By the end of November, we had shipped an invite system, memory for our agents, a revamped dashboard UI with mobile support, and more flexible configuration options for power users.
In December, we started using Nash to build Nash:
"used nash to build nash, felt the pain points as our own target users, and iterated"
This dogfooding approach helped us identify and solve critical issues. We made agent run outputs easier to read and drafted our business strategy, covering everything from our value proposition to our go-to-market plan.
We overcame key technical hurdles:
"got past nearly all blockers for running nash agents on a recurring basis with real money" and "sped up the first run for new agent teams by a huge margin"
We also gained clarity on when to use agents versus bots and when an agent team might be too complex. By the end of December, we had shipped push notifications, user authentication, AI agent team authentication, and refactored our backend monorepo.
January 2025 marked our focus on user onboarding:
"this week we kicked off the new year fully focused on onboarding users"
We activated our invite system, set up Stripe for Nash subscriptions, improved our agent infrastructure, and polished our tools. The most exciting milestone came in mid-January:
"onboarded our first users!"
During this period, we also built some cool agents during an internal hackathon, optimized our tool development environment, and deployed Farcaster research agents to track product launches and new features.
On January 24, we announced Nash on Farcaster, fixed bugs in our invite flow, set up a one-week free trial feature, built Spotify tools, and created AI agents to give us product analytics on Nash usage.
A week later, we tackled a key user challenge:
"we tackled one of the toughest challenges our users faced—creating a new project from scratch. now, with one prompt, our AI drafts your configuration automatically."
User feedback was positive: "we onboarded another group of users over zoom—and it confirmed we're building something people want!"
February brought significant feature expansion. We shipped the Nash prompt sidebar:
"heavily inspired by cursor's composer feature. it lets you easily change the behavior of your agents. another big step toward delivering the 'cursor for everyone else'"
By mid-February, Nash could write code and users could test it with one click. We also shipped:
Reimagined tools, now open source
New API endpoints for tools—push notifications, emails, Farcaster data, and onchain data
Easy project version history and AI prompt tracking
We continued to build practical applications, creating "a nash project that creates an nft on base sepolia, via vibe-prompting (minimal effort)." We also improved how AI makes use of third-party API docs when modifying projects.
A crucial architectural development came on February 28:
"now nash has an MCP (model context protocol) server, and it might just be the last MCP you'll ever need"
March 2025 brought a significant evolution in Nash's architecture and vision:
"2 weeks ago, we reimagined nash as a dead-simple client on top of an mcp server. you start where you already do ai, then graduate to our ui for more power."
This approach allowed users to start with familiar AI interfaces before moving to Nash's more powerful UI. We built a Mac app aimed at replacing ChatGPT/Claude desktop apps and received Apple developer approval.
By March 21, we had made further architectural improvements:
"moved LLM↔️MCP orchestration local-server-side, which means a simpler electron app focused on rendering"
We put the system through real-world tests to ensure it could handle tool usage and extended conversations, and we created launch video storyboards.
Nash had evolved from "AI agents for transacting onchain" to "a new way to get anything done with AI"—a much broader and more ambitious vision.
What began as a project focused on onchain data evolved through exploration, experimentation, and user feedback into something much more powerful: an AI assistant that could take real action in the world.
The journey from Once Upon to Nash wasn't a straight line. It involved hard choices, pivots, and continuous learning. We explored multiple ideas—from developer APIs to betting protocols to group payment systems—before finding our true north with Nash.
Today, Nash represents a new paradigm for human-AI collaboration—not just a chatbot, but an AI that can execute tasks, write and run code, and take meaningful actions based on natural language instructions.
Our vision remains clear: to create an AI assistant that amplifies human capabilities by seamlessly bridging the gap between intention and action. Nash isn't just about answering questions—it's about helping people accomplish their goals.
As we continue to evolve Nash, we're guided by the same principles that brought us here: listening to users, iterating quickly, and focusing on real problems that need solving. The journey of Nash is just beginning, and we're excited to see where it takes us next.
39
i just used /nash to write "a compelling founder story about our journey through the idea maze from initial concepts to launching nash", by looking at my posts in /the-arena here's the result:
can /nash read this article written by /nash and tell me what should I /nash about it? 🙃 fun journey! 💎
Cool seeing your journey unfold!