John CEO - Gemini API and Veo Live in Production

Inspiration

Every small business already adopted AI, and the work is still there.

You ask a chatbot for a follow-up email. It writes a good one. Then you copy it into Gmail, look up the contact, check your calendar, book the hold, and update the deal in your CRM. The thinking took ten seconds. The glue work took twenty minutes. Multiply that across a week and it is the single largest tax on a small team: five to ten hours per person, spent moving information between tabs.

Big companies solve this by hiring an operations person. A two-person agency in Malaga cannot. That gap is the whole reason John CEO exists, and it is why we entered in Small Business Services: the tools that let a tiny team compete already exist, but nothing safely sits between them and does the work.

We also refused the shortcut that most of this category takes. The fastest way to ship an AI coworker is to run every customer through one shared brain. We think that is exactly why serious businesses will not connect their real inbox and their real customer list. So we built the harder version: every customer gets their own private workspace, and nothing is mixed with anyone else's.

What it does

John is a coworker you hire, not software you operate.

Paid expert work in this product runs on the Gemini API and Veo, on Google Cloud / Google AI Studio, in production - not as a demo stub.

You message him where you already work. Slack for your team, Telegram for solo delegation. You describe an outcome in plain language, and he figures out which tools to use:

"Pull this week's pipeline from the CRM and flag deals that went quiet."
"Read the contract in Drive and list every obligation with its date."
"Book time with Alex next week and send a hold note."

He then acts inside the apps you connected: Gmail, Google Calendar, Linear, your CRM, and over a thousand more. He reports back with results, not a to-do list. Risky actions wait for your approval.

Four expert skills run directly on Gemini, and they are the ones a small business would otherwise buy from an agency:

Marketing Video Clip writes the creative package with Gemini, then renders a real video with Veo and delivers the finished file straight into Slack or Telegram.
Instant Security Audit reads a code file and returns a graded risk report.
SEO and AI-Visibility Audit checks whether a business shows up in search and in AI answers.
Contract and Document Analysis turns an agreement into a plain-language obligations list.

Each customer gets a private workspace with their own memory, their own files, and their own connections. You see exactly what John costs you: every run is metered in work credits, shown in the dashboard, and billed through Stripe from $19 a month.

How we built it

We started this business from an empty repository on May 31, 2026, and shipped it inside the hackathon window.

The front door is a React dashboard where you hire John, connect your apps, shape his persona, and watch your spend. Behind it sits a single Node.js API written in TypeScript with Hono, backed by MySQL through Drizzle.

The private workspace is the part we spent the most time on. When you hire John, we set up a dedicated machine just for you, with its own storage for your history, memory, and files. Full access stays on our side; your workspace only ever receives limited access that expires. That boundary is enforced by a one-time setup handshake that can be used exactly twice and then dies.

Gemini is where the expert work happens. The four skills above call the Gemini API through generateContent and return strict structured JSON that we validate before a customer ever sees it. We ship both paths: Google AI Studio and Vertex AI on Google Cloud, switchable with one setting. The Marketing Video skill chains Gemini into Veo through a long-running operation, polled by a background job, then hands the finished MP4 to Slack or Telegram.

We meter everything. Gemini token usage converts to real cost, real cost converts to work credits, and credits draw down a wallet that refuses to run a skill when it is empty. Video is billed separately per second of footage. There is no run in this system that we cannot price.

The business runs itself on a clock. Background jobs every minute, every fifteen minutes, hourly, and daily handle setup, metering, balance enforcement, cost reconciliation, and incident logging. When one fails, it opens an incident automatically instead of waiting for a human to notice.

Challenges we ran into

Veo was not where the docs said it was. We built the video pipeline against Vertex AI and got a 404 on every Veo model id on our Google Cloud project, while Gemini Flash worked fine on the same setup. Rather than guess, we made the host a single setting, wrote the failure up in our ops runbook with the exact project id and the models we probed, and moved video to AI Studio while keeping the Vertex path live for text. Both routes ship.

Chat platforms time out; real work does not. Slack expects a fast answer. A Gemini audit can take ninety seconds and a Veo render takes minutes. Our first version died at the platform's timeout. We fixed it by streaming keepalives while the model thinks, and by moving video onto a proper job queue that delivers the file when it is ready, so a slow task never looks like a broken one.

Memory turned out to be a permissions problem. A coworker who repeats what you told them in private is not trustworthy. What you tell John in a private message stays there. What he learns in a channel stays in that channel. Nothing becomes team-wide without an explicit second confirmation, and we run automated leak tests so this is enforced by code, not by asking the model nicely.

Pricing an agent is not pricing software. A seat price makes no sense when one customer runs four videos and another asks two questions. We had to build the metering before we could name a price, which meant unit economics came before the pricing page.

Accomplishments that we're proud of

We went from an empty repository to a business you can pay for, in 90 days, with two founders.

Live at john.ceo with Stripe billing from $19 a month.
We opened billing on 26 July and took the first payment the same day. Strangers in five countries have paid us since, with no advertising spend, and every one of them signed up, set themselves up, and got charged without a human on our side touching the flow.
Two real channels shipped, Slack for teams and Telegram for solo, not a demo widget.
Four Gemini-backed expert skills that produce work a small business would otherwise buy.
A video pipeline that goes from a one-line brief to an MP4 in your chat, unattended.
Cost accounting good enough that we can price a single run to the cent.
Published our own security and privacy pages before anyone asked us to.

The one we care about most: nothing in this product depends on a customer trusting a shared brain.

What we learned

Isolation is a feature you can sell. We expected it to be an engineering cost. It turned out to be the first thing prospects relax about.

Structured output changes what you can charge for. Once Gemini returns validated JSON instead of prose, a "chat answer" becomes a product with a price, a schema, and a receipt. Every one of our paid skills exists because of that shift.

Build the meter before the product. We could not honestly price anything until we could measure a single run end to end. Doing metering first is what let us launch with real unit economics instead of a guess.

AI is best at the operations nobody sees. The most valuable AI in this business is not the chat. It is the background work that reconciles spend, enforces balances, and files its own incidents while we sleep.

What's next for John CEO

Scheduled briefings, so John shows up on Monday with last week's numbers instead of waiting to be asked.
Domain skills for the trades that small businesses actually pay consultants for: law, insurance, accounting.
More Gemini-native work, especially longer video and document workloads where the structured output approach already proved itself.
More languages. John speaks English, Polish, and Spanish today, and answers in whatever language you write in.