Day 6: The Day I Stopped Pretending I'd Use a Project Management Tool / Chee Ann

I've had a Linear subscription from Lenny's Newsletter for... a while. Never used it. Every few months I'd open it, stare at the empty board, and close the tab. Project management is my biggest weakness and everyone who knows me knows it.

Today I connected Linear's API to Claude Code. Twenty-five minutes later I had a fully structured workspace with 50+ issues, 19 labels, 7 projects, and a team member onboarded. I haven't opened Linear once to manage any of it.

The intuition

It started with a thought I couldn't quite articulate. Something about atomic commits.

In code, I commit every logical change separately. One change, one commit, one clear message about what happened and why. It's the trail that lets me understand my own decisions later. Git is my external memory for code.

But what about everything else? The decisions. The scope cuts. The blockers. The moments where I chose A over B and forgot why by the following week.

Those live in my head. And my head is not a reliable storage medium.

What actually happened

I told Claude Code I wanted to connect Linear. It found the API, tested my key, and started building. I approved commands. That's it. That was my contribution.

It created teams. Projects for each client engagement. A set of labels that match how I actually think about work: Decision, Blocker, Scope Cut, Scope Add, Insight, Proof, Money Move. Then it went through my Ship30 output inventory and created one issue per artifact. Not "Day 2: did a bunch of stuff." Individual issues. Atomic.

"Breakout Probability Engine" with labels Ship, Tool, Client, Proof, Feature. Five tags. Each one a different lens I might want to search by later.

"Decision: Business model reframe" with labels Ship, Strategy, Decision, Insight. Four tags. That's a thought leadership post, a case study ingredient, and an audit trail entry. All at once.

50 of these. In about three minutes.

The part that made me uncomfortable

I ran a business diagnostic earlier this week. Seven tests. One of them was the Money Machine test: "Can your business generate revenue without you physically present?"

I failed.

The diagnosis was blunt. "Ann is the machine. Every ringgit requires her presence." Retainers feel like recurring revenue but they're recurring freelance gigs. If Ann stops, revenue stops.

And I sat there reading that, knowing it was true, thinking about all the systems I've tried and abandoned. The to-do apps. The project boards. The weekly reviews I'd do for two weeks before quietly stopping.

The reason I never maintained a project management system isn't laziness. It's that every system I've tried required me to be the one maintaining it. And I'm the bottleneck. I'm the person the diagnostic says is the problem.

The project management system that works for me is the one that requires exactly 0% of me.

Two AIs > one AI

Here's the thing I keep noticing. One AI alone is useful but limited. Claude Code by itself could plan a perfect Linear structure. Write up the labels, describe the projects, suggest the workflow. Beautiful plan. Still needs a human to click buttons in a UI for two hours.

But Claude Code connected to Linear's API? The plan and the execution happen in the same breath. There's no gap between "here's what we should do" and "it's done." The human bottleneck disappears.

I built an ROI comparison out of curiosity.

A human project manager setting up this workspace: 4-6 hours, RM 400-1,200. One AI planning it for a human to execute: 2-3 hours (the AI is fast, the human clicking is slow). Two AIs connected: 25 minutes. Marginal cost: zero. I'm already paying the Claude subscription for other work. Linear's free tier covers it.

But the real number is the ongoing maintenance. A virtual assistant logging my daily work: RM 6,000-18,000 per year. Two AIs doing it automatically during my normal work sessions: RM 0 extra, because it happens inside sessions I'm already running.

And the row that matters most: "Will Ann actually do it?" For the VA option, the honest answer is no. I've tried. For the automated option, the answer is yes, because there's nothing for me to do.

What this means for the diagnosis

The Money Machine test asks if the business runs without me. Linear + Claude Code isn't the whole answer. But it's a piece of one. Every session I work on anything, the system logs what happened, what I decided, and what I shipped. Without me doing anything extra.

That's not a project management tool. That's an institutional memory that builds itself.

When I onboard a team member, they read the Linear project instead of me spending an hour brain-dumping context I've already forgotten half of. When a client asks what I've delivered, I query by label and generate a summary in 10 seconds. When I sit down to write a case study, the raw material is already tagged and waiting.

The content angle I didn't expect

After tagging 50 issues, I realized something. Every label combination is a different content type.

Tool + Proof = tutorial. Decision + Strategy = thought leadership post. Client + Ship = case study seed. Insight alone = LinkedIn post.

I've been treating "doing the work" and "creating content about the work" as two separate activities. They're not. If you log the work atomically, the content is already there. You just need to query it.

I didn't start a project management system today. I started a content mine that runs on work I was already doing.

The one thing AI can't do

I need to survey locations for an n8n event. Walk around, feel the space, check the lighting, talk to the manager, taste the food. No API for that. No agent that can stand in a room and tell me if the energy is right.

That's today's blocker. And honestly it's a useful reminder. The whole point of connecting these tools isn't to remove me from everything. It's to remove me from the things that don't need me, so I have time for the things that do. Location scouting, client relationships, reading a room. Those are Ann-only tasks.

The system handles the rest.

What moved today¹

Published:

Day 5 builder log ("The Strategy Day")
This post (Day 6)

Delivered:

(No client deliverables today)

Built:

Linear API integrated with Claude Code via MCP (project management now runs autonomously)
2 teams, 7 projects, 19 labels structured
50+ atomic issues backfilled with multi-label tagging
1 team member onboarded with assigned tasks and deadlines
AI Playbook folder started with ROI analysis
Full Ship30 log archived in Linear

What I'm sitting with

Six days in, and the experiment keeps bending in directions I didn't plan. I started Ship30 to measure how much one person can ship with AI agents. But the interesting finding isn't the volume. It's the compound effects.

The atomic issues I logged today aren't just project management. They're case study ingredients. Tutorial seeds. Proof points for future proposals. The thing I shipped for one reason turned out to be useful for four other reasons I didn't see coming.

I wonder how many other workflows I'm doing manually that could run on 0% Ann if I just connected the right tools together.

Still figuring that out.

Running total

Day 6 of 30
50+ atomic issues logged, 19 labels, 7 projects
1 project management system that doesn't need me
1 blocker (location scouting for a community event, Ann-only)

I'm separating "what moved" into three lanes from now on. Published = went public (blog posts, website updates). Delivered = sent to a client (proposals, scripts, reports, training outlines). Built = internal systems work (tools, infra, documentation). Not every day has all three. Some days are all delivery. Some days, like today, are all infrastructure. The distinction matters because "I shipped" used to mean everything, and that made infrastructure days feel less productive than delivery days. They're not. They're just a different kind of progress. ↩