Why Decisions Became Free (And What Gets Built Because of It)

Every computing era has a resource that quietly goes from precious to free, and the freeing redraws the map. Compute did it, storage did it, bandwidth did it; each time, the practitioners of the scarce era found the abundance vaguely obscene ("nobody needs a gigabyte") right up until entire industries stood on it. This site catalogs what we think is the next one, small enough to fit in a sentence: structured judgment just became too cheap to meter.

Not intelligence broadly; that remains gloriously expensive at the frontier. Specifically the ability to ask, of any piece of text or state, a closed question, and get a fast, adequate answer for a price that rounds to zero. The receipts pile says it plainly: 26,000 verdicts for thirteen cents, decisions inside game frame budgets, judgment embedded in a SQL WHERE clause. Numbers as reported by their builders, and directionally unanimous.

The scarcity we stopped noticing

Software has always been split into what it could decide (anything expressible as logic over structured data) and what it had to escalate (anything requiring reading). The escalation layer was human, so it was rationed: we sampled the calls instead of hearing them, reviewed two percent instead of everything, filed the documents unread, let the tickets route by keyword and prayer. Whole institutional designs, the QA sample, the audit sample, the triage queue, the "flag for review" button, are fossilized workarounds for judgment scarcity, so old we mistook them for best practices rather than budget constraints.

Chat-era LLMs cracked the reading problem but kept the rationing: judgment at frontier prices was still judgment you deployed carefully. The decision-model turn is the un-rationing, and the tell is behavioral: within days, builders stopped asking "can we afford to judge this?" and started asking what isn't judged yet.

What abundance actually builds

Watch the catalog and three shapes recur. The coverage flip: every domain organized around sampling reorganizes around totality: moderation judging every post, QA scoring every call, invoices reviewed at 100 percent, every AI output guardrailed. Sampling survives only as audit of the judge, which is its correct job. The judging membrane: systems grow verdict layers at every boundary: inputs screened, outputs checked, agents' claims verified, models routing models. Software starts to resemble an organism with an immune system rather than a pipe with hopes. The loop invasion: judgment cheap enough for the inner loop enters places it was never allowed: game ticks, agent action-selection, editors judging keystrokes, and eventually control layers with actuators attached.

The dissenting clause, kept load-bearing because this site's whole method is receipts over rapture: free judgment is adequate judgment, not wise judgment; it inherits biases, wobbles nondeterministically, and invites adversaries. Abundance eras always mint their own failure genre: the spreadsheet error at scale, the spam flood, and, presumably, the confidently misrouted everything. The craft that matters shifted accordingly, from affording judgment to specifying it: the question became the unit of engineering, the eval became the unit of trust, and the humans moved up a level, from making the rulings to writing the law and auditing the court.

That's the thesis this directory exists to test in public. Six hundred builds in, it's holding. Bring receipts if you can break it; we'll catalog those too.

Frequently asked questions

What does "decisions became free" actually claim?

That closed-set judgment over text and state now costs fractions of a cent at millisecond speed, per builder-reported workloads, low enough that coverage replaces sampling as software's default posture.

Isn't this just "AI got cheaper"?

It's one specific capability crossing a threshold: frontier reasoning remains expensive; verdicts stopped being. Thresholds, not gradients, are what redraw architectures, and the cascade is the redrawing.

What's the strongest objection to the thesis?

That adequate-but-biased judgment at total coverage scales harm as efficiently as value; the answer isn't optimism but the audit stack: evals, calibration, humans on consequence. Where that stack is skipped, the objection wins.

What should I read next?

What Jev is for the object itself, the use-case tour for the evidence, and the limitations for the edges; the thesis only means anything with all three attached.