Three Factors Won’t Cut It

Our brains aren’t dumb; they’re throttled. In Bandwidth Economics I argued that modern life bombards us with far more data than our working memory can juggle—about seven items, according to Miller’s famous “7 ± 2” limit . To cope, we lean on blunt heuristics: anchoring on the first number we see, saliency bias for whatever shines brightest, and “good‑enough” selection that reduces 50 inputs to three. My son, Tigin, choosing a college? Rankings and peer choices drown out faculty depth, alumni firepower, housing, commuting, progression, or career outcomes . These shortcuts aren’t laziness; they’re how a 128‑bit wet CPU survives a terabyte world. The price shows up later in regrets, mis‑priced risks, and missed upside.

We rarely look at the two blindspots:

1. Probabilities. Most decisions are wagers, yet we evaluate them like coin flips. Venture data show only ~18 % of Series‑A startups reach break‑even within five years , but many candidates treat a tech offer as “secure” or “risky” with zero odds attached. Anchored to anecdotes, we overweight rare shocks (layoffs) and underweight common drags (sector slow‑downs).

2. Time. We love “now”: starting salary, cool office, instant prestige. Future states—skills a job will build in three years, demand for those skills in five, residual brand equity in ten—fade into haze. Decisions ripple across short‑term (≤1 yr), mid‑term (3‑5 yrs), and long‑term (5 yrs+) horizons; treating them as snapshots distorts the whole frame.

Ignoring probability and time warps trade‑offs: a fat paycheck in a sunset industry, or a cheap house beside a rising floodplain, looks brilliant—until it doesn’t.

Structured frameworks already exist—Multi‑Criteria Decision Analysis (MCDA), weighted decision matrices—but manually filling them for 30 variables is spreadsheet CrossFit. Enter state‑of‑the‑art AI:

• Factor harvesting. A no‑code system scrapes job boards, Glassdoor, SEC filings, and even LinkedIn churn to surface variables humans skip.

• Bayesian weighting. The model assigns importance scores you set (1‑10) and updates likelihoods as new data arrive.

• Temporal simulation. Decision trees project best‑, base‑, and worst‑case paths across three horizons, visualized as heatmaps or radar charts.

• The machine does the math; you remain the pilot.

Let's look at an example where I helped my young friend Sarah consider two job offers:

1.      List & rank 22 factors and group them: Financial, Career Capital, Culture, Logistics, Personal Fit.

2.      How the AI processed it: Factor weights came from Sarah (financial 35 %, career 25 %, culture 15 %, balance 15 %, risk 10 %). Probabilities pulled from Crunchbase, LinkedIn churn, and sector forecasts. A temporal decision tree projected outcomes across 0‑1 yr, 3‑5 yrs, and 5 yrs+. Startup X glowed green on short‑term cash and mission buzz but bled red on stability, culture, and work‑life balance. SaaS Y skewed yellow‑green across most rows, with deep green on balance and runway.

3.      Temporal map Year 1: Startup X crushes salary but drags on commute. Years 3‑5: SaaS Y’s stronger mentorship and sector CAGR compound into higher expected skill marketability. Year 5+: Equity at X pays out in only 8 % of Monte‑Carlo runs; Y’s brand equity delivers steadier optionality.

4.      Matrix & heatmap The dashboard splashes red on X’s stability and green on Y’s growth curve. Sarah sees an 18‑month cliff for X: if a Series‑B round fails, her résumé stalls. Y’s downside is slower salary growth.

5.      Bias audit sidebar † Sarah opens Explain Weights: stability risk isn’t gender‑coded but tenure‑coded, so she keeps the feature. Compliance box ticked, human judgment engaged.

6.      Sarah asked X to triple the equity to 0.30 % and offer hybrid work to compensate for 52‑hour weeks and 45‑minute commute. X refused—inflexibility that signaled deeper governance and culture risks the AI had already flagged. That final data point tipped the scale: she chose Y, trading a flashy bump for compounding skills, saner hours, and a sturdier sector tailwind.

Bandwidth limits push us toward three‑factor snap calls, blind to odds and time. A rigorously audited AI‑enhanced framework scales cognition: 30 variables scored, probabilities refreshed, futures charted. Miller said our mental RAM is seven items; AI is the external SSD—your files, just more space to think. Next quest for the precious ring of career stability? List twenty factors, let the model crunch, keep your judgment on the throne, and watch the road to Mordor light up on the dashboard.

I hate this essay to sound like the self-help shit that it does. I set out to write about the AI mayor of a city with all the complex decisions done automatically etc. Then I slammed on the brakes: we don't even make simple decisions reliably and here I'd be so far in the future and be booed out of the room. Self-help shit is better for LinkedIn anyway. I'd gain 3 more followers. Yay! (Note to self: buy domain name mayor.ai and start cookin').