Cheaper AI production may swell output without broad entrepreneurial gains: as more creators enter low-marginal-cost markets, scarce human attention dilutes per-producer returns and concentrates rewards among a few winners.

Main Finding

The paper formalises the "Builder Saturation Effect": when AI dramatically lowers the fixed cost of building digital products (k → 0) in markets with near-zero marginal costs, free entry causes the number of producers (B) to rise but aggregate human attention (A) remains limited. As B grows, average attention and profit per builder fall toward zero, while reinforcement and quality heterogeneity generate increasingly skewed, power‑law or condensation (winner‑take‑most) outcomes. Thus democratised production is more likely to intensify competition and produce concentrated winners than to create broadly distributed entrepreneurial success.

Key Points

• Core scarcity: attention (A = M · a) is the binding scarce resource in digital markets; production can scale elastically but attention cannot.
• Baseline model:
  • Consumers allocate attention via a logit rule. With B symmetric products and outside‑option weight z = e^{β(q0−q)}, average attention per builder is s̄(B) = A / (B + z), which strictly decreases in B.
  • Builders monetize attention at rate p; profit π = p·s − k. Under free entry, expected profit = 0, yielding equilibrium B* = max(pA/k − z, 0).
  • Lower entry costs (smaller k) increase B* but do not raise average profits — they dilute attention.
• Extensions (heterogeneity & reinforcement):
  • Introduces attention stocks xi(t), reallocation fraction δ, quality qi drawn from F, and reinforcement parameter α in allocation probabilities: pi(t) ∝ xi(t)^α e^{β qi}.
  • Special cases: α = 0 → static logit; α = 1, homogeneous qi → linear preferential attachment (power‑law P(x) ∝ x^{-3}); α = 1 with fitness qi → Bianconi–Barabási fitness model, which can produce fit‑dependent power laws and condensation (one product capturing macroscopic share).
• Empirical motivation/examples: App Store and GitHub statistics showing extreme concentration (top 1% capturing most revenue/downloads) and maintainer overload consistent with attention scarcity.
• Conceptual synthesis: combines attention economics (Simon), information‑goods cost structure (Shapiro & Varian), monopolistic competition, superstar/convex reward logic (Rosen), preferential attachment and cumulative advantage (Barabási, Bianconi), and network effects.

Data & Methods

• Methodological approach: applied theoretical formalisation and synthesis of established results rather than novel mathematical primitives.
• Baseline analytic model:
  • Population: N agents with M consumers each having attention budget a → aggregate attention A = M·a.
  • Product quality qi and outside option q0; logit allocation with sensitivity β.
  • Free entry equilibrium determined by zero-profit condition.
• Dynamic/heterogeneous extension:
  • Mean‑field discrete‑time reallocation: xi(t+1) = (1 − δ) xi(t) + δ A · pi(t) with pi(t) ∝ xi(t)^α e^{β qi}.
  • Parameters: reinforcement α ≥ 0, sensitivity β > 0, reallocation share δ ∈ (0,1].
  • Uses known results from preferential-attachment and fitness models to import distributional conclusions (power laws, condensation).
• Empirical/illustrative evidence: cited platform-level statistics (App Store downloads/revenue shares, GitHub AI-generated contributions, stable per-user app counts) to motivate the attention constraint; a simple simulation parameter table is included (e.g., M=10,000; A=10,000; B=1,000; z=100) though paper focuses on analytic and imported stochastic results.
• Assumptions and limits: near-zero marginal cost, free entry, fixed aggregate attention, logit (or multiplicative) form for allocation, exogenous monetization rate p and attention budget A; dynamic entry/exit and endogenous attention growth not modelled.

Implications for AI Economics

• For founders and investors:
  • Lowered build costs will increase the number of entrants but not the number of successful ventures proportionally. Expect intensified competition for attention and a smaller fraction of outsized winners.
  • Strategies that affect discovery, compatibility, bundling, or network advantages (platform integrations, pre‑existing installed base, strong signal/branding) become more valuable than marginal improvements in build efficiency.
• For platform designers and marketplaces:
  • Curation, ranking algorithms, and mechanisms that expand effective attention (better recommendation, personalization, demand stimulation) materially affect how entry converts into realized value.
  • Without thoughtful discovery mechanisms, open entry risks overwhelming human evaluative capacity and collapsing average returns.
• For policy and welfare analysis:
  • Democratic ease of production does not automatically imply democratized economic gains; welfare impacts hinge on demand-side responses and distributional effects (increased consumer surplus from variety vs. wasted effort/duplication and concentrated producer rents).
  • Interventions that expand aggregate attention (e.g., demand stimulation, lowering search costs) or reduce winner‑take‑most externalities (e.g., interoperability standards, platform non‑discrimination) could change outcomes.
• For empirical work:
  • Testable predictions: (i) average attention/per-producer engagement falls as entrant counts rise (controlling for A); (ii) outcome distributions become heavier‑tailed as α and heterogeneity increase; (iii) condensation/winner‑take‑most regimes emerge where early or small quality differences are amplified.
  • Useful data: platform downloads/usage, revenue concentration, time series on entrant counts vs per‑product attention, measures of reinforcement (visibility effects), and experiments on discovery algorithms.

Caveats: the paper abstracts from endogenous attention growth, multi‑period strategic entry/exit, and richer monetization dynamics. Its main value is synthesising established mechanisms into a targeted framework that highlights attention as the key constraint in AI‑enabled democratised production.