So I Started a Company. Cool Cool Cool.
Here we go. Again.
I’ve been in technology for nearly 30 years. I’ve built ETL products from scratch and sold them to clients like Kimberly Clark, Coca-Cola, and Johnson & Johnson. I’ve managed $12M cloud budgets. I’ve overseen 7.4 million Azure resources. I’ve migrated 1,200+ servers for a Fortune 100 company. I’ve built Cloud Centers of Excellence, Zero Trust models, HIPAA-compliant environments, and DevOps cultures inside organizations that didn’t know what DevOps meant when I walked in the door.
I’ve also started two businesses before this one. So I know exactly what I’m getting into.
That’s not a flex. That’s a warning label.
This is my third company. Third time’s the charm — that’s what people say, right? Sure. Let’s go with that. What I’ll actually say is: the first time you start a business you don’t know what you don’t know. The second time you know exactly what you don’t know and do some of it anyway. The third time you sit down, look at the blank page, and think okay, we’re not doing that again approximately forty times before you write a single line of anything.
That’s the salty part — and honestly, the most valuable part.
Spec First. No Exceptions.
Here’s what nobody tells you about starting your third company after decades of enterprise experience: you lose every last shred of blissful ignorance that lets most founders just ship something and figure it out later.
Past-me skipped the spec. Past-me said “we’ll document that later.” Past-me made architectural decisions at 2am fueled by optimism and bad coffee and paid for them for years. I’ve lived that cautionary tale twice, and I’ve watched it play out inside large enterprises that had no excuse given their resources.
“We’ll figure it out later” is the most expensive sentence in software development. I have receipts.
So this time, before I wrote a single line of production code, I wrote a specification. A real one. Over 100 pages, organized into 23 modules — each one covering a distinct domain of the platform: infrastructure, database schema, API contracts, service boundaries, compliance controls, CI/CD pipelines, operational runbooks. Every decision made, documented, and locked in before any code touched a keyboard.
Overkill for a solo pre-funding founder? Maybe. The direct result of lessons learned the hard way across two previous businesses? Absolutely.
Agentic AI Changes the Math
Here’s where it gets interesting — or at least where I tell myself it’s interesting while questioning my life choices at 1am on a Monday.
Most people’s mental model of AI-assisted development is still pretty simple: ask it something, get an answer, copy-paste, move on. That’s not what I’m doing.
I’m using agentic AI — and there’s a meaningful difference. A regular AI prompt is a conversation. An agentic AI is a workflow. It doesn’t just answer questions; it takes sequences of actions autonomously within boundaries I’ve defined. Write the code, run the tests, catch the failure, fix it, commit. It works through a task end to end and hands control back to me when it hits a decision I’ve reserved for myself.
The spec is what makes that possible. Because the agent loads two to four modules at a time — a focused slice of context — it has exactly what it needs and nothing it doesn’t. Each session has a defined scope. A checkpoint gates the next phase. Nothing advances until the checklist is clean.
I decide what to build. The spec defines the rules. The agent executes — fast, autonomously, and without asking me if I’m really sure seventeen times.
What I’ve learned is that agentic AI is only as good as the inputs you give it. The quality of the output is almost entirely a function of the quality of the spec. Garbage in, garbage out — as true as it’s always been, just with a garbage can that now takes itself out.
The Company Gets the Same Treatment
One thing I’ve done differently this time — and I’m genuinely proud of this — is treating the company itself with the same rigor as the platform.
Every corporate document lives as version-controlled markdown alongside the codebase. Operating agreement, investor agreements, NDA suite, contributor framework, banking resolutions, partnership templates — all of it. Versioned, auditable, maintained. If something changes, there’s a history. If someone asks a question, there’s an answer.
Most founders treat company docs as something to sort out later. I’ve seen what “later” looks like. It’s not cute.
So Why Now?
Fair question. I’ve asked myself this roughly three hundred times since January.
Two reasons. First: I got tired of being the person in the room who knew exactly what needed to happen but didn’t have the authority to make it happen. I’ve watched good ideas die slow deaths by committee inside organizations at every size. I’ve built things that worked beautifully and watched them get shelved for reasons that had nothing to do with technology. I’m done with that chapter.
Second — and this is the real one — the tools finally match the ambition. My first company, I did everything manually and outsourced what I couldn’t. Second company, better cloud tooling but still fundamentally dependent on hiring people to execute. This time, agentic AI means one person with a solid spec and the right toolchain can move at a pace that used to require a team. I have the experience to write the spec correctly. I have the scar tissue to know where the guardrails need to go. And I finally have the execution tooling to back it up.
That combination feels like the window I’ve been waiting for. Maybe I’m wrong. Wouldn’t be the first time.
But third time’s the charm. That’s the story I’m going with and I’m not taking questions.
Full transparency: after months of building specs, company docs, and investor materials — hundreds of AI sessions, millions of tokens consumed — I asked the AI to distill it all into a blog intro. This is what came out. That’s kind of the whole point.
— Randy
