Field Notes — Interactive
The vector sum
On the homepage we say ungoverned AI turns a company into a fleet of jetskis — fast in every direction, no shared heading. This is the math underneath that sentence.
Elon Musk once described a company's output as the vector sum of the talent of the people. That's not a metaphor; it's the right unit of analysis. A person isn't a quantity of talent — they're a quantity of talent pointed somewhere. Add the vectors and you get what the organization actually produces, which is routinely a fraction of what it pays for.
Magnitude is the wrong obsession
Most owners work the magnitude knob: hire smarter, push harder, move faster. Fewer work the alignment knob — how tightly the team points one way. Almost nobody works the third: judgment, whether the shared direction is the correct one. Drag the sliders and watch what each is worth.
The dangerous configuration isn't slow. It's fast, aligned, and wrong — crank alignment to 100%, drag judgment negative, and the team sprints off a cliff together, in perfect formation. Speed amplifies whatever direction it's given. That's why speed alone — including the kind AI hands out for free now — isn't a strategy.
The math of fragmentation
Direction explains where output goes. A second model explains why so little gets produced at all. An article called The Math of Why You Can't Focus at Work models individual productivity with three variables:
Interruption rate
How often you're knocked off task. Interruptions per hour.
Recovery time
How long it takes to reload context after each interruption.
Focus threshold
Minimum uninterrupted time for one unit of meaningful work.
capacity = focus blocks long enough to clear the threshold
The arithmetic is brutal: one extra interruption per hour can cut productive output several-fold, because interruptions don't just steal their own minutes — they shatter blocks below the threshold where work counts at all. It's not a discipline problem. It's a systems problem.
One day, simulated
Each bar is an eight-hour day. Green is deep focus, red is an interruption, amber is recovery — at the desk, but the context hasn't reloaded yet.
Push λ to 4+ and watch the day disintegrate. That's not a hypothetical — studies put knowledge workers at 15–30 interruptions per hour. The defaults above are generous.
A hundred days
Same model, run a hundred times. Each cell is a day; darker means more deep-work blocks completed.
Notice the variance. Some days break through; some produce nothing — with the same person, the same effort, the same intentions. That's not discipline. That's the system.
The parameter space
The whole landscape at once: expected deep-work capacity for every (λ, Δ) combination. Toggle θ upward — harder, more meaningful units of work — and watch the habitable zone shrink.
Read it like this: find your company's interruption rate on the Y-axis and its typical recovery time on the X-axis. The color says how much deep work is possible. Most organizations live in the amber-to-red zone and wonder why nothing ships.
From individuals to organizations
The article stops at individuals. The model doesn't have to — the same three variables exist at the company level, and they map to things we install:
The threshold only exists if you know what a meaningful unit of work is. Without a stated thesis, every effort is sub-threshold: busy, and nothing compounding.
People pointed the wrong way don't just underproduce — they interrupt. Misalignment raises everyone else's λ. This is what a real operating cadence attacks: fewer, better meetings; owned decisions; escalations that happen once, in the right room.
At the org level, recovery time is how long it takes to reload context after a strategy shift, a bad quarter, a departure. A company whose knowledge is evidence-stamped and current recovers in hours. A company whose knowledge lives in three heads and a stale drive recovers in quarters — if at all.
Governed vs. ungoverned
Now put the two models together. Both teams below have identical talent — same vectors, same magnitudes. The left one is governed: aligned to a heading, and the heading is close to true north. The right one is the jetski fleet: every unit fast, each pointed wherever it happened to be facing when someone said “go faster.”
This is what AI does to an ungoverned company: it grows every arrow. The sum barely moves. Speed is now nearly free — which means direction is nearly all of the value.
Culture isn't a knob
subject to θ (purpose), λ (noise), Δ (memory)
Most attempts to fix a stalled company reach for culture: alignment offsites, values documents, town halls. But culture isn't a knob you can grab. Speed, alignment, judgment, λ, Δ, and θ are knobs — and culture is what it feels like to work somewhere the knobs are set well.
This is the engine, restated as math. The operating layer lowers λ and defines θ. The knowledge stack collapses Δ and keeps judgment calibrated against reality. Measurement tells you which knob is mis-set. And the reason it has to be installed bottom-up — the reason the install is operator work done from inside, not advice delivered from outside — is that no one layer moves the sum by itself. The product of six knobs is only as good as the worst one.
If you want to know which knobs your company has, start with the AI-Native Test.
Vector-sum framing after Elon Musk. Fragmentation model adapted from The Math of Why You Can't Focus at Work (justoffbyone).