How We Built Daffodil

Most of you know the story of 3 friends and former colleagues who left their jobs to start a new company with lofty goals requiring a multi-faceted approach to technology, compliance, and go to market. Some of their funding came from friends & family, the rest they bootstrapped.

These 3 are us if that wasn’t obvious. We were able to launch Daffodil in month 8, and close a few major partnerships before the 12-month mark.

How did a team of 3 build a compliant 501(c)(3) charitable capital custodian for donor-advised funds, an impact data AI processing platform, 3 SaaS applications, a brand, customer base and partnership program in under a year?

Obviously AI helped a lot, but let’s be honest what else went into it.

TL;DR

• high founder conviction (a function of experience and a bit of hubris)

• experimental mindset (rapid iteration and pilot programming)

• 80 hour work weeks (there’s no getting around this)

• software development capacity of ~ 20+ developers (executed by 2 software engineer cofounders throwing money at Cursor)

Ok, so what did we actually build?

Taking inventory

Today we have 100s of nonprofits across the US participating on a “zero burden” impact network and daily funds flowing from donor-advised funds custodied by our 501(c)(3) DAF. All money movement is transacted on our software as digital grant recommendations or as algorithmic allocations funded from a pooled “Discovery Fund”.

The “tangible” artifacts of our technology include 3 applications: Nonprofit App, Compass (the Giving App and DAF interface), and Mission Control (the stakeholder app that aggregates intelligence across fund flows and impact data).

Unstructured data is collected, semi-structured, vectorized and repurposed as various intelligence artifacts for different audiences. This data processing and the aforementioned grant deployments are provisioned and transacted on a durable execution workflow orchestrator, manipulating accounting entries and AI enriched data pipelines.

We have in-house and customer-facing AI workflows and agentic interfaces we use every day.

We have a cool brand, conference and sector presence, great customers and amazing partners.

So what did we farm out?

Outsourcing & Downsizing

Our original brand development was outsourced to our friends at Copper Gate Design. The 3 of us did everything else in house because we could.

We had the luxury of starting a new business and company from scratch in today’s ecosystem. Cost cutting takes a lot of forms, but most of those forms are human shaped. Having performed the layoff exercise dozens of times in our career histories, we’re grateful we launched when we did to avoid the necessary “operational efficiencies” that new AI powered tools and processes mandate.

Or is it delegation?

You should never outsource your core competency.

That said, 98% of our code is written by LLMs via Cursor IDE with strict, constant oversight. This is what we consider professionalized, productivity boosting, AI assisted delegation.

Alright, we have to talk about AI.

A modern company stack

Here I’ll get into specifics of what we’re using (vs not).

Marketing

🌐 marketing and brochure sites - Cursor & Vercel

I am a Squarespace power user and I will never use it again.

🧲 lead magnets and landing pages - Cursor & Vercel

These can be of various levels of simplicity or complexity, but we can do it all in house, faster, cheaper, AND easier. LaunchDarkly, Unbounce, etc - no need.

📰 content copy support - Notion, N8N, Perplexity Pro

We lean on Perplexity Research Mode as needed. Then we actually write, as people with brains. At times, we’ll run an AI assisted review. We don’t generate content straight from AI. We do have an in-house n8n workflow, however, that has access to our brand guidelines and writing examples, can score our content and even suggesting improvements.

R&D

🎨 product design & prototypes - Cursor & Vercel

Claude is good at this, but we use the Claude models in Cursor. This allows us to spin up a POC locally or on our hosting environment using Vercel tools just as quickly.

“But what about UX Design?” I hear you asking. None of us are UX trained, but we’ve worked with and/or managed enough excellent product designers over our years to have learned from them. We know enough about good practices, dark patterns, and maintaining consistency to be dangerous, and to keep us productive.

🧪 product development, broadly - Cursor, Superwhisper, Perplexity Pro

MVPs can validate deeper assumptions when they’re semi-functional prototypes or wizard of oz tests vs just landing pages or painted doors. We can spin the former up just as fast or faster than coordinating a UserTesting or Lyssna test with a clickable prototype.

As far as UI design, it was plenty fast and easy to combine Storybook + Shadcn + custom CSS to design and build a functional component library, all without the Figma intermediate.

🛠️ software development - Cursor & Superwhisper

We are not vibe coding, but we are definitely letting Cursor take the wheel, while lending a watchful eye. We have collectively 40+ years of software development, architecture and leadership experience on the founding team and it’s been incredible to see how models have progressed in a short amount of time.

And all the dumb shit it will do when it thinks you’re not looking.

We’ve had to absorb new “how to manage your AI” skills like appropriate prompting, plan creation, and vigilance against over-engineering, premature optimization, errant “TODOs” that would bite you had they made it to production.

It’s like overseeing an eager intern with the attention span of a toddler and the speed of an 8 core processor. When I die, someone delete my Superwhisper and Cursor histories.

Operations

⚙ automations - n8n, Notion, custom Slackbot

All user events push to n8n, which we branch off to Slack, Copper CRM, email, Sheets or Postgres. No need to spend on Segment, Iterable, etc.

Our content calendar, action items, project tracker, product roadmap, and GTM tracker are all in Notion, accessed by n8n, and interfaced by a custom Slackbot.

We can chat with our Slackbot (or via email) to answer questions, update docs, and schedule reminders for todos, read our calendars and warn of us weather happening at destinations we head to week over week. It’s all pretty weird, a little fun, and very helpful.

Running Lean

In startup lingo, running lean used to mean cheap, DIY, do only what matters, fake it ‘til you make it.

It still does, it’s just a lot easier to do now. The mental barrier to entry to “scrappiness” is a lot lower than it used to be. This is a good thing.

And when all processes become cheaper, easier and faster to do, the size of what you can put into market becomes much bigger, must faster.

Whether that is a good thing remains to be seen, but I am optimistic.

👀 ⬇️ Check out what we built

• Donate to orgs or open a donor-advised fund here

• Nonprofit? Participate in our network here 

• Advise or move philanthropic dollars? Check out Mission Control, our app for advisors, family offices, and grantmakers here