I Automated the Writing. I Kept the Sending.
This post was drafted by an agent — I kept one thing from it: the send button. Why automating the production but never the permission is the whole design of the company I'm building.
Build in public — Taigency, Day 1 of the daily series. 28 July 2026.
This post was drafted by an agent. I'm telling you that up front because the whole point of what I'm building depends on you knowing it — and on one line I refuse to move.
Today I wired the piece of the system that writes these posts. A content agent that reads what the business did that day, drafts the blog, the LinkedIn post, the thread, the carousel — and then stops. It does not publish. It drops everything into an approval queue and waits for me to read it, change what's wrong, and press send. The agent does the writing. I keep the sending.
That split isn't a technical limitation. It's the design.
I started online before the web existed — dial-up bulletin boards, a modem screaming into the night, a sysop somewhere who read every new user request and manually decided whether to let you in. Trust on a BBS was a person saying yes. It was slow and it didn't scale, and we replaced it with systems that scaled beautifully and trusted nobody in particular. Twenty-five years later I'm building a company that runs on agents, and the thing I keep pulling back toward is that sysop pressing a key.
Here's the thesis I keep coming back to: the internet isn't dying, it's collapsing inward. For thirty years it was a place you went — a set of pages you browsed, tabs you opened, sites you visited. Now it's becoming something you talk to. The pages are folding into a conversation. When that happens, the scarce thing isn't information or even content. It's the judgment about what's true enough, and kind enough, and yours enough to put your name on. A model can generate a thousand posts. It cannot decide which one you'd actually stand behind. That decision is the job now.
So the architecture of Taigency is deliberately lopsided. Agents draft aggressively — outreach, calendars, specs, BOMs, content, whole workflows. Nothing external ships without me flipping a row from "draft" to "approved." Every client-facing thing passes one human gate. Not because the agents are bad at their work; they're better than I expected. Because the gate is where the human part of the business lives.
The lesson from today: automate the production, not the permission. It's tempting to close the loop entirely. Resist it on anything that leaves the building. The value of an approval gate isn't that it catches mistakes, though it does. It's that it keeps a person accountable for every word that goes out under their name. Speed you can buy. Accountability you have to keep.
Tomorrow the agent drafts again, and I read it again. That's the rhythm.
If you want the same setup — a bench of AI operators that draft your outreach, ops, and content and then wait for your yes — I've bundled the twelve I run as Agents of a CEO. Free, built for Claude.
Want one of these built for your business?
I build custom AI platforms and go-to-market engines end-to-end. Book a session and let's map it out.
Book a session