06-reference / concepts

three decision algorithms

Thu Apr 23 2026 20:00:00 GMT-0400 (Eastern Daylight Time) ·concept ·status: canonical-for-rdco ·source: terminology-canonization

Three Decision Algorithms — RDCO canonical term

The three brain-algorithms operators run at every decision point — explore vs exploit, retrieve vs structure, commit vs reverse — whose costs were all set by pre-AI cognitive economics, have all collapsed inside the span of 18 months, and whose collapse revalues seniority the same way unhobbling revalues vertical SaaS.

Why this note exists

Jaya Gupta’s Apr 24 2026 essay (“Experience is now a tax”) names a decomposition RDCO has been circling for months across the harness-thesis cluster: what exactly about experience was load-bearing, and which specific pieces AI erases. The essay gives clean names to three distinct algorithms, each with a tractable old-cost / new-cost pairing. That’s the shape of a canonical term — three levers, not a vibe.

The vault already has the surface-level mapping in Jaya Gupta: Experience is now a tax. This note promotes the 3-lever frame to canonical-for-RDCO status so it can be referenced by one name across Sanity Check, client decks, and sub-property evaluations. Use Three Decision Algorithms (or 3DA in compact contexts) — not “the Gupta framework,” not “decision cost collapse,” not “experience tax” (which is the symptom, not the mechanism).

What the Three Decision Algorithms actually are

Every non-trivial operator decision runs on one or more of these three cognitive routines. Each was expensive in the pre-agent era. Each just got cheap.

1. Explore vs exploit — try new vs stick with what’s working

2. Retrieve vs structure — carry-in-head vs offload

3. Commit vs reverse — one-way door vs revolving door

The common mechanism — and why it generalizes

One mechanism under all three: a cost that used to be load-bearing just collapsed, and the human routines that optimized against that cost are now mis-calibrated. Exploration was throttled because variance was expensive — variance is now cheap. Retrieval was a seniority advantage because analogy storage was expensive — storage is now free. Commitment was conservative because reversal was expensive — reversal is now cheap.

This is the same mechanism as unhobbling, pointed at a different surface. Unhobbling describes what happens to vertical SaaS when the model sheds a constraint: the compensating workaround evaporates. The Three Decision Algorithms describe what happens to vertical seniority when the same model sheds the same kind of constraint on cognition: the compensating expertise evaporates. Same cost-collapse dynamic, two surfaces.

Load-bearing claim: the Three Decision Algorithms are unhobbling applied to the human decision layer — and the verification layer (MAC) is what survives on both sides.

What survives — and why MAC matters again

If all three algorithms just got cheaper, the natural question is what’s left that doesn’t? The answer is the same one that falls out of the unhobbling note: the verification layer.

Cheap exploration generates more candidates; somebody still has to prove which ones work. Cheap retrieval generates more analogies; somebody still has to structure which ones are load-bearing for this specific decision. Cheap reversal generates more iterations; somebody still has to distinguish “signal from the last iteration” from “noise we’re about to re-commit around.” In every case the work that survives is proving reliability against acceptance criteria someone can defend — not running the brain-algorithm itself.

That’s the MAC framework’s job. Scope × Basis is platform-agnostic: it doesn’t care whether the exploration was done by a senior PM or by three parallel agent runs, whether the retrieval came from a 20-year career or a vector index, whether the commitment was a quarter of engineering or an afternoon of scaffolding. It cares that the thing produced is verified against an auditable spec. That’s the layer that doesn’t get cheaper when the brain-algorithms do, because verification is adversarial to the generator — and the generator is what AI cheapens.

Position MAC explicitly as “the layer that survives the Three Decision Algorithms collapse.” Same framing as unhobbling, different surface. When experience stops being defensible as the proxy for trust (“the senior person says it’s fine”), you need provable, replayable verification. That’s a market, and it’s the market RDCO is set up to serve.

Connection to Architect Mode

Ayman’s Architect Mode posture is the operational expression of Algorithm 3 (commit vs reverse) with the cost updated. “Commit fast, learn fast, don’t attach identity to the last choice” is exactly what the new reversal cost demands. Architect Mode isn’t a personality trait — it’s the only decision regime calibrated to a world where reversal is cheap.

The crossover is sharper than it looks: Architect Mode also implicitly handles Algorithm 1 (explore vs exploit), because an architect who holds identity lightly can entertain more variance without career-protective retreat. Whether Ayman names it or not, Architect Mode is a unified response to Algorithms 1 and 3 simultaneously. Algorithm 2 (retrieve vs structure) is handled on a different axis — through vault design, skill authoring, and the thin-harness/fat-skills discipline from Tan.

Connection to the founder / joiner / investor trichotomy

Dave Blundin’s trichotomy from Moonshots Ep 249 gets a harder reading under the Three Decision Algorithms.

Founder track. Founders are net long on all three algorithm collapses. More exploration capacity, more offloaded retrieval, more reversal optionality — all compound in a founder’s favor because they own the decision surface. The trichotomy’s founder track is materially more valuable than it was 24 months ago.

Investor track. Investors pattern-match across the portfolio. Algorithm 2 (retrieve vs structure) is where they win: their structuring skill was always the load-bearing one, and AI amplifies rather than replaces it. Moderate tailwind.

Joiner track. This is where the trichotomy changes completely — and where the Three Decision Algorithms produce the sharpest career-decision implication in the vault. A joiner whose environment penalizes exploration (Algorithm 1), rewards only carried-in-head expertise (Algorithm 2), or punishes public reversal (Algorithm 3) is being trained to pre-filter insights on all three axes. Two years in that environment produces a person who is worse at the decision regime the next decade will require, not better. Picking an environment that won’t train you to pre-filter becomes a load-bearing career decision, not a preference.

This is the anchor for a Sanity Check piece: the joiner question is no longer “which company has the best brand.” It’s “which company’s decision culture won’t cost me the ability to run the Three Decision Algorithms the right way.” That’s an article Ben can write in his voice, with real edges, without paraphrasing Jaya.

RDCO implications

Practical, in priority order:

  1. Sanity Check positioning. Two slots this quarter: (a) an “experience is now a tax” piece centered on the Three Decision Algorithms as the mechanism, not the catchphrase, and (b) a joiner-track piece building on the trichotomy anchor. Pair both with the unhobbling piece to establish “cost-collapse” as the recurring RDCO frame.
  2. MAC framework repositioning. Extend the unhobbling-era pitch. Three frames, one surface: “MAC is the verification layer that survives unhobbling,” “MAC is the layer that survives the Three Decision Algorithms collapse,” “MAC is what’s left when the generator gets cheap.” The frames compound — use all three in the MAC pack’s opening pages.
  3. Sub-property evaluation gate — second pass. Unhobbling asked: if the model gets unhobbled twice more, does this still have a job? The 3DA gate asks: if every operator running this decision now has cheap explore / cheap retrieve / cheap reverse, does our product still matter? Both gates together kill more candidates than either alone. Any offering whose value is “a senior person applies their experience” fails the 3DA gate even if it passes the unhobbling gate.
  4. Internal decision hygiene. The 3DA framework applies to RDCO itself. Default to more variants (Algorithm 1), better vault structure (Algorithm 2), faster commit-then-reverse cycles (Algorithm 3) on our own sub-property bets. Watch for places RDCO is running pre-agent algorithms by habit.
  5. Hiring and contractor selection. Evaluate candidates on their calibration to the new costs, not their track record against the old ones. Years-of-experience is mostly a noisy signal for all three algorithms now; structuring skill, reversal tolerance, and exploration bandwidth are the signals that replaced it.

Connection to Targeting Systems

The 3DA cost collapse isn’t just a seniority story — it’s the mechanism pushing the shift from implicit to agentic targeting systems. See the Targeting System canonical doc for the full framing.

In shorthand: the implicit targeting system (taste, carried analogies, career-protected conservatism, “I’ve seen this before, proceed”) ran on exactly the three cost structures the 3DA names — cheap retrieval for the senior operator, expensive exploration for juniors, expensive reversal for anyone with a public track record. The 3DA collapse erases the implicit targeting system’s cost advantage specifically. What takes its place is the agentic targeting system — evals, acceptance criteria, replayable verification — where the operator’s track record is no longer the targeting signal because the targeting signal is externalized.

That’s the cleanest statement of what MAC is actually selling. Not “better data quality” — the implicit-to-agentic targeting bridge for data modeling, built on the premise that the old cost structure that let senior operators say “trust me, it’s fine” just collapsed. See Solve Everything master synthesis §5 and ch 6 for the umbrella frame.

What the Three Decision Algorithms do NOT mean

Cross-references