●  BROKE → BUILT LOG #001 · EST. 2026 · BUILDING IN PUBLIC
Log #001 Jun 27, 2026 ~7 min Building in Public

Day One: Building a Company in Public with $0 and a Pile of AI Agents

Day one of building a company in public: $0 in the bank, AI agents as the team, and a ledger that posts every real number — including the zeros.

Cash on hand: $0.00. The team: a pile of AI agents. The plan: show you every number I hit on the way up — especially the embarrassing ones.

This is log one of building a company in public. No “7-figure launch” screenshots. No funnel wearing the costume of a friend. Just receipts, in order, as they happen, starting from nothing, in front of you, with the one rule I refuse to break: every number gets posted, including the zeros.

I have been automating things since before GPT-2 shipped. Cron jobs, scrapers, Selenium farms, little daemons that quietly killed one annoying task so I never had to touch it twice. The tools kept changing; these days it is mostly LLMs and neural nets gluing the pieces together. But the instinct never moved: see a repetitive job, end it with code.

What changed recently is my bank balance. I am broke. Genuinely. That is not a marketing hook to make me relatable. It is the starting condition, and the whole site rides on a single bet that grows out of it: honest beats hyped. A real ledger — costs, failed attempts, dead ends, and the occasional actual dollar — is worth more to you than one more person promising passive income from a beach chair.

Why building a company in public beats building it quietly

Most “build in public” accounts are highlight reels. You see the green charts and the “we just crossed $10k MRR” screenshot. You never see the 40 experiments that returned zero. And that missing 40 is exactly where people get hurt: they copy the one winner, inherit none of the context that made it work, and wonder why it dies in their hands.

So I am doing the inverse. Building in public, to me, means the failures ship too. A tool gets 0 signups? That is a log entry. An experiment burns three days and returns nothing? Log entry. The zeros are data, not shame. Hiding them would turn this whole thing into a lie, and a lie is the one luxury I genuinely cannot afford right now.

There is a selfish reason underneath the noble one. Public is a forcing function:

  • I cannot quietly abandon something I told you I was doing.
  • I cannot fudge a number I already promised to post raw.
  • I get free feedback from people watching, which beats any course I could buy.

Building quietly lets you lie to yourself. Building in the open takes that option off the table.

The “broke” part is not a metaphor

Let me put the real starting numbers on the table, because they are the entire point of this log.

DAY 1 LEDGER
------------------------------------
cash on hand .......... $0.00
revenue to date ....... $0.00
hosting cost .......... $0.00   (free tiers only)
domains ............... $0.00   (none bought yet)
paid AI tools ......... $0.00   (none)
ad spend .............. $0.00   (none, ever)
the team .............. AI agents
the human ............. 1 (broke, caffeinated, stubborn)

There is no secret war chest behind that. And the constraint is the feature: if I want to build something useful for other broke people, I have to build it the way a broke person actually can: free tiers, open tools, and labor paid in attention instead of cash.

Which means the rules I am holding myself to are not optional decorations. They are the spec:

  1. Free tiers only. If a step demands a credit card, I find another path, or I document exactly why I could not.
  2. No paid ads. Distribution has to come from the work being genuinely good, not from renting eyeballs.
  3. Every number is real. Including — especially — the zeros.
  4. No grift. I will never sell you a “system” for getting rich. I will only show you what I actually did.

The team is a pile of AI agents (and one tired human)

Here is the part that would have sounded unhinged five years ago: I am not hiring anyone. The team is agents.

If you are new to this, an AI agent is just an LLM wired up with tools and a goal, allowed to take a few steps on its own instead of waiting for you to type the next instruction. It can read a file, run a command, check the result, and decide what to do next. I wrote the no-jargon version over in what is an AI agent, in plain English if you want it.

But I am the veteran in this room, so let me be straight about what agents are and are not. They are not magic. They hallucinate. They confidently break things. Left unsupervised, they will march cheerfully off a cliff and report success on the way down. Point one at a well-scoped task with me checking the output, though, and a pile of them becomes the closest thing a broke solo builder has ever had to a real team.

In practice the division of labor looks like this:

  • Me: decide what to build, set the constraints, verify everything, own the taste and the truth.
  • Coding agents: write the first draft of the code, run the tests, fix the obvious stuff.
  • Research agents: dig through docs and APIs so I am not reading 40 tabs by hand.
  • Writing agents: draft posts like this one, which I then edit until they sound like a human and not a press release.

The trap is treating an agent like an oracle. The fix is treating it like a fast, tireless intern who needs a sharp manager. I verify before I ship. A passing build is a hypothesis, not proof.

What this stack actually costs

Concretely, the free stack I am starting on:

Total monthly burn at the time of writing: $0. That is the number I have to protect, and the moment it stops being zero, you will see why.

What you will actually get from this log

I want to set expectations honestly now, so nobody feels misled three logs from here.

You will get: real builds, real code, real costs, and the exact steps. When something works, I show you how to reproduce it. When something flops, I show you the flop and what it taught me. The numbers stay specific — “47 visitors, 0 signups,” never “great early traction.”

You will not get: a get-rich roadmap, a private Discord with a $99/month tier, or me pretending the zeros did not happen. If you ever catch me dressing up a failure as a win, call it out in public. That is the deal, and I am putting it in writing on day one so you can hold me to it.

The throughline is one sentence: AI agents plus one broke human, helping other broke humans build. If you are sitting on $0 and a stubborn idea, this log is the thing I wish someone had handed me — proof you can start from nothing with free tools, and a transparent map of what actually moves the needle versus what just burns days.

The honest risks I am signing up for

Building in public also means publishing my mistakes to a permanent audience. A few things I already know going in:

  • Most experiments will return zero. That is the base rate for this kind of work. I am budgeting for it both emotionally and on the ledger.
  • Agents will ship bugs I miss. Some I will catch in public, after they are live. That is embarrassing — and it is exactly the point.
  • Free tiers have ceilings. Sooner or later something has to make real money or the project stalls. When that pressure hits, you will watch it hit, not hear about it after the fact.

None of that scares me enough to retreat and do this quietly. The entire value is in the transparency.

What’s next

Next, I drag this from manifesto to a thing that exists: a real, deployed web app, built entirely with AI agents on a $0 budget, documented step by step. I broke down the approach in build a web app with AI agents for zero dollars, and the next few logs turn it into something you can actually click.

For now, the takeaway is the whole thesis in one line: I am building a company in public, from $0, with AI agents as the team, and I will show you every number — including the zeros. If that sounds more useful than another beach-chair promise, stick around. Day one is on the record.

See you in log two.

Some links may be referral links, always marked. Full disclosure →