How I work

The plan I never opened

Five years into running my business I pulled up the plan I'd written for the SBA loan. Over a hundred pages. Customer analysis, SWOT models, revenue forecasts. I had not opened it once. Not to check a forecast, not to revisit a strategy. I opened it that day for a laugh.

It wasn't the first plan that let me down. At Microsoft I was test manager for IIS, back when the culture was shifting from cowboy to formalized, and I built a test plan template that grew into a monster. A hundred to two hundred pages. Weeks to write and longer to review. There was real value in writing them, because the process forced you to understand the problem space before anyone touched code. Then they sat in a folder on my desktop until the next release. We confused the deliverable with the discipline.

It was never a shortage of tools. OKRs for goals, Agile and Waterfall for delivery, Kanban for tracking, MVPs for product. Every one of them good at the thing it does, and not one of them connecting what I did on a Tuesday to why I was doing it. My task system was a legal pad and my inbox. If it was important I'd remember it, and if it was urgent somebody would email me. That inbox hit twenty-five thousand messages before I admitted the system was broken. I had all the tools and none of the framework.

Where to aim, and who to become

What was missing was alignment, and the distinction that made it usable I got from JB Glossinger's MorningCoach. Mission is what you're going to accomplish. Vision is who you're going to become. When I plan a year, mission tells me where to aim and vision tells me who I have to be to get there. Every goal, every project, every task traces back to those two, and without that connection you're just busy.

That got tested in an April quarterly review. Revenue was flat and I had no capital to put into sales. I was looking at my own mission statement, which was a revenue target, and it landed that the revenue had never been the point. It was a proxy for a stable financial foundation for my family. I rewrote the mission that afternoon. The plan hadn't failed. The plan was aimed at the wrong thing, and the review is the only reason I found out.

The cascade

The rhythm underneath is Glossinger's too. Six quarterly goals become six monthly projects, become six weekly, become six daily. The number isn't the point. The constraint is. Six forces a no to everything that doesn't fit, and the filter is two questions. Does this serve the mission, and does it move me toward who I said I'd become.

Yearly sets the destination. Month one builds the foundation, month two is the reality check, month three closes things out. The weekly review is where I find out whether any of it was true.

The month I stopped pretending about a health goal

Offensive work moves you toward the mission. Defensive work keeps the lights on. Both are necessary, and if defensive is eating seventy or eighty percent of the week for weeks on end, you're running in place and the alignment number on its own won't tell you.

One January I sat down with the month's data. A hundred and twenty tasks completed, and eighty percent of them aligned to my Sacred 6. That reads like a win, and I let it read that way for about a minute. Then I looked at what the aligned work actually was. Seventy-five percent of it was defensive. Core business functions, all of it necessary and none of it building anything.

Then the part I didn't want. I had a health goal sitting in my Sacred 6 all quarter. One task against it. One, the whole month, on a goal I had told myself mattered. The data doesn't argue with you and it doesn't care what you meant. If it's important, give it more focus. If it isn't, take it off the list and stop pretending.

I restructured the following month around that. I built the thing that caught me and I still didn't see it until the number was in front of me.

Fifty percent in eight hours

The system doesn't stop the chaos. I was out of town on business when one employee called in sick and came back with a twenty-one day quarantine after a measles exposure. A second went dark. Three days before we pieced it together: she'd lost her childcare, her dog had been hurt, and she just stopped answering. A third was already out on planned vacation. Eight hours, fully staffed to operating at half, run from another city through a phone screen.

Every Sacred 6 task I had lined up became irrelevant, because the only thing that mattered was keeping the operation alive. That isn't the system failing. It held different priorities for a while, and when the crisis passed I didn't have to work out where I was or what came next. The quarterly goals were still sitting there. Missing a day, the next morning resets you. Missing a week, the weekly review catches you up.

Where AI comes in

An agent runs the review mechanics now. Daily, weekly, monthly and quarterly, each level rolling up into the next, assembled before I sit down. Every task carries a three-bucket score tracing it back to why it matters, so I see a task's alignment dropping before I've sunk real time into it. The assembly never needed a human. The judgment stays on my side of the line, which is the same split I'd make for anyone else's cadence.

Everything I've built to run it started as a written PRD before any code, with a decision log recording the options rejected next to the one chosen. The published repository carries 37 numbered decisions and 18 architecture decision records, several superseding or amending each other in a traceable chain, so you can read a position change and the reason for it rather than take my word that I revise.

And I check the mechanics from outside them, because passing tests are a claim rather than a result. One feature in that repository shipped with 232 green tests and a clean review and had never once succeeded in production, because every fixture held three messages and the real mailbox held thirty-nine thousand. The property that mattered was the one the fixtures normalized away. Anything whose cost scales with real accumulated history now gets one measurement against the real thing before I call it done.

Why this transfers

I built this for a ten-person cleaning business and I run it now at a headcount of one. Scaling it past me hasn't happened yet, and when it does the hard part won't be the arithmetic. It'll be that the inputs belong to other people who have their own reasons for what they report. That is the same problem as getting three orgs to instrument against one shared session, which is the only part of this I have actually done at scale, and it was nine years ago.

The full planning system, including the years I got it wrong, is in the book. The systems themselves are public at willsad007/AIExecAssistant, including the arguments I lost.