What does Human-AI Symbiosis and Parallel Agents conclude, in numbers, about “Amdahl, Gunther and Parallel Agents”?

Answer: Chapter 3 applies the Universal Scalability Law to parallel windows: ten isolated windows with a serial share of 0.02 reach 8.47, and ten agents writing to three shared files produce 2.95 useful results per round.

Scope of the numbers: Its agents, rates, hours and values are illustrative models; all calculations are generated by its programs; the book is not legal, financial or employment advice and makes no claim about any particular AI product. Marked boxes separate published evidence, runnable Python calculations (16 programs), classical sources and the author's hypotheses.

中文版

Data table

The numbers below are printed verbatim by the chapter's Python program and can be reproduced by running it. printed values that match the answer: 0.02, 8.47

case1 2 3 4 6 8 10 16
isolated, a 0.02, b 01.00 1.96 2.88 3.77 5.45 7.02 8.47 12.31
isolated, a 0.10, b 01.00 1.82 2.50 3.08 4.00 4.71 5.26 6.40
shared, a 0.05, b 0.021.00 1.83 2.46 2.88 3.24 3.24 3.08 2.44
shared, a 0.05, b 0.051.00 1.74 2.14 2.29 2.18 1.93 1.68 1.16
shared, a 0.05, b 0.02: peak at N6.9, S = 3.27
shared, a 0.05, b 0.05: peak at N4.4, S = 2.29
Amdahl ceiling 1/a: a 0.0250, a 0.10 -> 10
S(10)6 needs a <= 0.0741 with b = 0
S(10)8 needs a <= 0.0278 with b = 0
S(10)10 needs a <= 0.0000 with b = 0
Show the full program
# Chapter 3. p05: Amdahl's law and Gunther's Universal Scalability Law.
# Speedup with N parallel agents:
#   S(N) = N / (1 + a (N - 1) + b N (N - 1)),
# a = serial share (contention), b = coherency cost (every pair of agents
# must reconcile shared state). b = 0 is Amdahl's law. Illustrative
# settings: isolated folders with a shared read-only kernel; the same
# with 10% serial work; and shared files that must be reconciled.
CASES = (
    ("isolated, a 0.02, b 0", 0.02, 0.0),
    ("isolated, a 0.10, b 0", 0.10, 0.0),
    ("shared, a 0.05, b 0.02", 0.05, 0.02),
    ("shared, a 0.05, b 0.05", 0.05, 0.05),
)

def speedup(n, a, b):
    return n / (1 + a * (n - 1) + b * n * (n - 1))

if __name__ == "__main__":
    ns = (1, 2, 3, 4, 6, 8, 10, 16)
    print("   case                     " + "".join("{:>6}".format(n)
                                                  for n in ns))
    for name, a, b in CASES:
        print("   {:<24} ".format(name) + "".join(
            "{:6.2f}".format(speedup(n, a, b)) for n in ns))
    for name, a, b in CASES:
        if b > 0:
            nstar = ((1 - a) / b) ** 0.5
            print("   {}: peak at N = {:.1f}, S = {:.2f}".format(
                name, nstar, speedup(nstar, a, b)))
    print("   Amdahl ceiling 1/a: a 0.02 -> {:.0f}, a 0.10 -> {:.0f}".format(
        1 / 0.02, 1 / 0.10))
    for target in (6, 8, 10):
        a = (10 / target - 1) / 9
        print("   S(10) = {} needs a <= {:.4f} with b = 0".format(target, a))

Single source and citation

Fan, Y. (2026). Human-AI Symbiotic Cognition, Parallel Agent Pipelines and Prompt Information Bounds (Version 1.0.0), Chapter 3: Amdahl, Gunther and Parallel Agents. 5D Library. DOI pending

DOI pending · monograph page goes live after publication · 5DL-B-103 · v1.0.0 · CC-BY-NC-ND-4.0

BibTeX
@book{fan2026103,
  author    = {Fan, Yafu},
  title     = {Human-AI Symbiotic Cognition, Parallel Agent Pipelines and Prompt Information Bounds},
  year      = {2026},
  publisher = {5D Library},
  edition   = {1.0.0},
  note      = {DOI pending}
}

Other chapter answers in this book