Case study · AI-native planning & judgment
Cadence tells you when to look. Alignment tells you if it matters.
This is how I think about the planning and rhythm-of-business problem, built AI-native: an agent runs the review cadence, I keep the judgment call. It caught a week that looked like 90% mission-aligned success, when 85% of that was actually defensive work.
Situation
I believe meetings, reviews, and reporting should be built around decisions and trade-offs, not status. I said exactly that in a Chief of Staff interview, describing the playbook I run. Every meeting is focused on decisions and trade-offs, not status, backed by an AI-generated pre-read that surfaces the questions before the room, not during it. This system is that belief built into software.
Most personal productivity systems run on willpower and memory. Most corporate rhythms of business run on status decks that report activity without ever checking whether the activity still matters. I designed something different. It’s an operating system, built on Notion, where an AI agent executes the review mechanics on a fixed cadence, daily up through quarterly. I spend my energy on the judgment those reviews surface.
I built it while managing a high-stakes transition. I was closing a business I’d run for years and returning to a technology career. I applied the same operating discipline I’d used at $1M ARR and across 26 years at Microsoft to my own execution instead of an organization’s.
Task
Inside one connected review, planning, and execution cadence, the system needed to do three things that a typical personal system, and most corporate rhythms, don’t. Treat the whole stack, from purpose down to the daily task, as one connected hierarchy, so any piece of work could be traced back to why it mattered. Apply the same KPI, SLI, SLO, and SLA discipline I’d used running production systems to my own execution, with trend history instead of a single point-in-time snapshot. And make alignment to mission and values a measured number instead of a slogan, so drift would show up in the data before it cost me something.
There was a constraint underneath all three. The real objective was redirecting cognitive energy away from tactical, mechanical work, the low-level tasks that eat attention without needing judgment, and toward the deeper work that requires it. That deeper work means reflection, planning, and the tradeoffs only a human should make. A system that costs more thought to run than the decisions it frees up fails on its own terms.
Action
I split the design into two axes. A horizontal axis of cadence loops: a Daily Compass and Daily Plan each morning, a Daily Review each evening, then Weekly, Monthly, and Quarterly reviews. Each one ingests the layer below it and emits up to the layer above. A vertical axis, the alignment spine, runs purpose (why, mission, vision, values) through every one of those loops. It’s enforced by a True North filter, a cascade that traces every task up to the mission, and a three-bucket score that separates chosen priorities from necessary-and-aligned work from real drift.
I’d seen what happens without that spine. At Microsoft, I watched two teams sit in the same staff meeting and hear the identical mission and vision from the same VP. They walked out executing in opposite directions, each one certain they’d interpreted it correctly. Nobody was lying. They just had no mechanism forcing their work back to a shared true north, so the drift compounded silently until it showed up as conflict between them. That’s the failure mode the alignment spine exists to catch before it costs something.
An AI agent runs the mechanics on schedule: pulling the data, calculating the metrics, drafting each review, and writing the output back into the system. That way, the next loop already has its inputs. I also design and evolve the underlying database structure the same way. I describe what I need in plain language and let Notion’s API, its MCP server, and its AI build or reshape the schema. That way I have to understand my data, not the mechanics of how Notion implements a relation or a rollup. I kept the parts that actually need a human: reflection, tradeoffs, and the calls where judgment matters.
The first version of the alignment spine was wrong in a way I didn’t catch for months. The three-bucket score read each task’s current Sacred 6 status, not the status it held at the time, so reprioritizing a task today silently rewrote how aligned every past week looked. Apparent drift was reading at 46%, alarming until I traced it back and found most of it was that flag rewriting history, not real drift. The fix was completion-stamping, a slowly-changing-dimension pattern. Stamp each period’s status when it closes. Never read it live. Real drift dropped to about 4% the same week the fix landed.
The clearest proof the design works came from a judgment call. A weekly review that looked great on paper turned out to be running on defense when the objective needed offense. The numbers are in the Result section below, in full.
The system evolves on its own input, too. Mechanism in the deep-dive below.
Architecture deep-dive optional, for the technically curious
The system runs on two axes, feeding one cascade, tied to one small measurement link.
Every task inside those loops rolls up through one cascade:
Goal + single Key Result.
PARA Objective tier.
PARA: the execution tree.
The executing layer.
Each Key Result carries a scoreboard row, linked to its own value history:
The self-improving loop:
- Input. I listen to a leadership podcast daily. AI captures it into my second brain as a searchable, referenceable artifact.
- Detection. AI reviews that artifact for anything that could sharpen a prompt, a skill, or the cadence itself. Open and closed loops was a recent example. It flags candidates but doesn’t get to decide on its own.
- Judgment. I decide whether a flagged idea is worth acting on.
- Change. If it is, we workshop it into the system directly: a new prompt, a new skill step, a new review section.
- The result. About a dozen enhancements have landed that way across the cadence over the last six months.
Result
I’m using the same honesty standard here that runs through the design itself. I label shipped and designed results separately here, not blending them into one number.
- Shipped and running. A Quarterly Review skill (a 17-step guided workshop), two new databases (a metrics scoreboard and a linked snapshot history for trend data), and the Daily, Weekly, and Monthly reviews. All of it runs with completion stamping and the three-bucket alignment score.
- It caught me, too. Early this year, a weekly review showed 90% of my work aligned to mission and vision, a week that looked good on paper. About 85% of that aligned work was actually defensive, keeping the business wind-down running and the lights on. Only about 15% went to the offensive work of finding a job, the one thing on the list that counted. The review surfaced the split as worth a hard look. That forced the trade-off conversation, and I reprioritized within the week.
- Designed impact. Weekly, monthly, and quarterly reviews all carry the same built-in reflection moments that caught the offensive-versus-defensive imbalance above. They flag something worth a hard look on a regular basis, not just that one week. Daily execution now traces up through the cascade to a 90-day objective.
- The system evolves, too. About a dozen enhancements have landed across the cadence over the last six months, sourced from daily outside input, filtered by AI, decided by me. Mechanism above.
Learning
The organizational version of this problem is the one I care about most. I ran a business on the same discipline, a weekly, monthly, quarterly, and annual rhythm. It started as a paper VTO, the EOS vision-and-traction tool, and moved to a digital system as it matured, including the 13-week rolling forecast that carried it through its hardest stretch. This system is that same discipline, automated with AI and turned on myself, because the planning and rhythm-of-business problem doesn’t change shape between a ten-person company and an individual. The test is the same at any scale. Look often enough, and make sure what you’re looking at still matters.
The score matters less than what it triggers; it’s there to flag what’s worth a hard look and force the conversation before it’s too late to matter. In my case that surfaced as offense versus defense, a 90%-aligned week that was actually 85% defensive. It could just as easily surface as a stalled key result or a value going quietly unlived. AI can watch for the pattern. Only I can decide what to do once it’s flagged.
This runs on an audience of one today, and I already know where it goes next. The same horizontal loops and the alignment spine map onto a team without a different design, only a different audience. The six chosen priorities become a team’s top priorities and key bets, the weekly review becomes a WBR, and the alignment spine becomes strategy-to-execution governance. Building that version, and proving it holds up when the judgment isn’t mine alone to make, is the next problem I want to solve.
All selected work