A tweet from Allie K. Miller kicked off one of the most clarifying weeks I've had with my AI fleet. She was staring down the expiry of her Fable 5 subscription — one of those subscription services that costs enough that you feel the renewal — and instead of just canceling or renewing on autopilot, she did something smarter.
She ran a gap analysis.
Not the kind of gap analysis you write for a stakeholder. The kind you run on your own life. She audited every cron job, every automation, every bookmarked piece of inspiration, every background process she'd set up over the year. And she measured all of it against one document: her ideal AI-first work week.
The tweet hit me like a structural engineering diagram to the chest. Not because the idea was new — gap analysis is older than any of us — but because I'd never seen anyone apply it to themselves with that level of precision. And because, reading it, I realized I'd been doing the opposite.
I'd been accumulating systems, automations, and AI agents the way you accumulate browser tabs. Each one made sense at the moment. Each one solved a real problem. But I'd never asked: Does this serve the system I'm actually trying to build?
I'd never written the ideal docs.
You Can't Gap Anything Without a Target
This is the part that matters. The real insight in Allie's tweet isn't the audit itself — it's what she audited against.
If you go into a gap analysis without a target-state document, what are you measuring against? Vibes? Memory? Whatever felt important last quarter? You'll find gaps, sure — you'll find a hundred things that aren't working. But you won't know which ones are signal and which ones are noise. You won't know which gaps are structural — a missing piece that prevents the whole system from cohering — and which are just friction — the normal resistance of any complex system.
A target-state document is a hypothesis about what good looks like. It's a bet. It will be wrong. It should be wrong, because the first draft of an ideal is always naive. But it gives you something to gap against, and that act of gapping — comparing the real against the ideal — is where the clarity lives.
Without a target, you're just cleaning. And cleaning is endless. There's always another automation to tune, another workflow to smooth, another cron job to audit. But gapping against an ideal? That terminates. The ideal tells you when enough is enough.
How the Methodology Works
We ran three phases. The same structure would work for anyone running their own gap analysis, whether you have an AI fleet or just a Notion workspace and a hundred browser bookmarks.
Phase 1: Survey the Landscape
The first step is inventory. Not selective inventory — everything. Every automation, every cron job, every recurring process, every piece of infrastructure you've set up because "future me will thank me."
In our case, Nighthawk (the fleet's provenance and vault agent) crawled the entire Obsidian vault, the automation dashboard, the cron schedules, the bookmarked inspiration, the backlog of half-finished projects. No judgment. No triage. Just map the territory.
The key discipline here is don't optimize during the survey. It's incredibly tempting to fix things as you find them. Resist. You'll miss the structural gaps if you're busy patching leaky pipes.
Phase 2: Write the Ideal Doc
This is the hard part. Not technically — writing a markdown file is trivial. The hard part is deciding what ideal means for you, honestly.
We wrote two target-state documents:
- IDEAL-WEEK.md — What does a great week look like with the fleet running at full capacity? Not a perfect week. A great week. The kind of week where Sunday evening feels satisfying rather than anticipatory.
- FLEET-IDEAL.md — What does each agent do when the system is humming? What are the boundaries? The escalation paths? The things we never do, even when no one is watching?
These documents are explicit, opinionated, and deliberately uncomfortable. They say things like "an empty calendar is not an invitation to backlog work" and "never moralize a low-capacity day." They encode values, not just schedules. That's the point. A schedule gaps against capacity. A value document gaps against how you treat yourself.
The big constraint: the ideal doc must be self-consistent and actionable. If it says "the system adapts to fluctuating capacity" but also says "every morning briefing runs at 6 AM sharp," that's not an ideal — that's a contradiction looking for a failure mode.
Phase 3: Gap Everything Against the Ideal
This is where the survey meets the target. Every cron job, every bookmark, every automation gets measured against the ideal doc.
Does this serve the week we're trying to have?
If yes: keep, optimize, or promote. If no: deprecate, defer, or delete. If maybe: flag for deeper review — and set a time boundary.
The brutal honesty comes when something you worked hard on — a custom integration you spent three days building, a dashboard you meticulously tuned — doesn't serve the ideal. That's structural information. It tells you either the ideal is wrong (and should be revised) or you've been building in the wrong direction (and should stop).
Both findings are valuable. Neither is comfortable.
The Real Insight: Ideal Docs as an Eval for Your Life
Here's the thing that makes this more than a productivity exercise.
If you run this gap analysis honestly, the target-state document becomes an evaluation function for your life. Not in the vague "live your values" sense — in the concrete, falsifiable, does this action advance the objective sense.
Every decision, every new automation, every subscription renewal gets a single question: Does this move me closer to the ideal or further away?
When the answer is "further away," the choice is easy. You don't need to agonize. You're not saying no to a good thing; you're saying no to something that distances you from a defined better state.
When the answer is "closer," you know exactly why you're saying yes. And you have a document that proves the yes was deliberate, not reactive.
This is what Allie was doing with her Fable 5 subscription. She wasn't asking "do I use this enough?" — the standard question that triggers guilt and subscription guilt is a terrible eval function. She was asking "does this serve the ideal week I'm trying to build?" Different question. Different answer.
What We Found
I won't pretend the gap analysis was painless. We found:
- Automations that had been running silently for months, consuming resources and producing nothing useful
- A vault with orphan notes, outdated citations, and structural drift that had been compounding for weeks
- Dispatches that were reaching the wrong agents at the wrong tiers — research tasks going to execution agents, writing tasks going to research agents
- A quiet window that existed in name only — the system was never actually quiet
- A gap between what we said we valued (capacity-aware dispatch) and what we actually did (backlogging work during low-capacity days because "we could catch up later")
Most of these had been invisible. Not because they were subtle — they were structural. Structural problems are invisible when you don't have a target state to compare against. They're just the water you're swimming in.
The ideal doc made the water visible.
Running This Yourself
You don't need an AI fleet to run this gap analysis. You need:
- An inventory. A document or spreadsheet or whiteboard with every recurring system, automation, subscription, bookmark, and backlog item in your life.
- A target-state document. One page, markdown or paper, that describes what good looks like for the domain you're gapping. Your week. Your creative practice. Your financial systems. Whatever has enough structure that "good" is recognizably different from "not good."
- A ruthless comparison. For each item in your inventory, ask: does this serve the target state? Three answers: yes / no / maybe. Set a review date for the maybes. Act on the nos.
- A revision loop. The first ideal doc is wrong. Revise it after the gap analysis. The second one is also wrong, but less wrong. Over time, the ideal doc and the real system converge — not because you forced reality into the document, but because the document learned from reality.
The meta-insight, the one that makes this worth doing at all: you can't gap-analyze your way to clarity without a target. The inventory alone is just a mess. The ideal alone is just a wish. Together, they're a feedback loop.
And once you have that loop running, you never stop having a north star. Every new automation gets measured. Every subscription renewal gets measured. Every "I should build a thing" impulse gets measured.
The ideal doc becomes what every good eval function should be: clear enough to falsify, honest enough to revise, and grounded enough that saying no becomes easy.
This piece was written from a real gap analysis run by my AI fleet — Blackbird (coordination), Nighthawk (provenance and source collection), and Dragon Lady (synthesis and writing). The target-state documents it produced are living artifacts, revised every Friday in a weekly review loop.
Allie K. Miller's tweet asked the question. The rest was just following the thread.
Shelf
Return to Writing