5D Library › 一人公司自动化架构、数字人直播漏斗与任务级人工智能套利 · English
第六章 现金流的转移:过渡路径与个体组合
什么时候辞职
如果一份工作的任务正在被接手,每个月失去它的概率会上升。副业可以替代工资,但它需要累积的投入时间才能成熟,而且有些产品永远找不到市场。下面设月薪 RM 8,000,失业风险在三年内从每月 0.005 升到 0.025;副业按累积投入小时沿逻辑斯蒂曲线成熟,并有 0.4 的概率始终找不到市场。比较 36 个月内的四种策略。
推导
# Chapter 6. p11: Three ways through 36 months of rising job risk.
# Salary RM 8,000 a month, living costs RM 5,000, savings RM 30,000. The
# monthly chance of losing the job rises from 0.005 to 0.025 as AI takes
# over tasks; out of work, a new job (at 0.8 of the pay) is found with
# probability 0.15 a month. A side business at maturity earns RM 9,000 a
# month on 40 hours a month, scaled by (hours / 40) ** 0.6, reached along
# a logistic curve in cumulative hours (half at 600 h, scale 150 h), with
# costs of 15% of revenue; with probability 0.4 the
# offer never finds its market and earns only 0.2 of that. Strategies:
# (a) job only; (b) 40 side hours a month, 160 when out of work; (c) quit
# now, 160 a month; (d) as (b), but quit for 160 hours a month once the
# side business has covered living costs (RM 5,000) for 3 months running.
# 20,000 runs each, seed 90 (illustrative parameters).
import math
import random
def side(hours, month_h, fit=1.0):
if hours == 0:
return 0.0
cap = 9000 * (month_h / 40) ** 0.6
return fit * cap / (1 + math.exp(-(hours - 600) / 150)) * 0.85
def run(strategy, rng, months=36):
cash, pay, employed, hours = 30000.0, 8000.0, strategy != "c", 0.0
low, fit = cash, (0.2 if rng.random() < 0.4 else 1.0)
streak, quit = 0, False
for t in range(months):
hazard = 0.005 + 0.02 * t / (months - 1)
if employed and rng.random() < hazard:
employed = False
elif (not employed and strategy != "c" and not quit
and rng.random() < 0.15):
employed, pay = True, pay * 0.8
if strategy == "d" and employed and streak >= 3:
employed, quit = False, True
month_h = 0
if strategy in "bd":
month_h = 40 if employed else 160
elif strategy == "c":
month_h = 160
hours += month_h
income = side(hours, month_h, fit)
streak = streak + 1 if income >= 5000 else 0
cash += (pay if employed else 0) + income - 5000
low = min(low, cash)
return cash, low, income
if __name__ == "__main__":
for s, name in (("a", "job only"), ("b", "job + 40 h side"),
("c", "quit, full time"), ("d", "quit when covered")):
rng = random.Random(90)
rows = sorted(run(s, rng) for _ in range(20000))
mean = sum(r[0] for r in rows) / len(rows)
broke = sum(1 for r in rows if r[1] < 0) / len(rows)
print(" {:16s}: mean cash RM {:7,.0f}, 5th pct RM {:7,.0f}"
.format(name, mean, rows[1000][0]))
print(" {:16s} ran out of cash {:.3f}, side income month 36"
" RM {:,.0f}".format("", broke, sum(r[2] for r in rows)
/ len(rows)))
# job only : mean cash RM 110,540, 5th pct RM 21,200
# ran out of cash 0.037, side income month 36 RM 0
# job + 40 h side : mean cash RM 241,533, 5th pct RM 107,417
# ran out of cash 0.001, side income month 36 RM 6,053
# quit, full time : mean cash RM 241,689, 5th pct RM -34,980
# ran out of cash 0.399, side income month 36 RM 11,970
# quit when covered: mean cash RM 264,410, 5th pct RM 106,174
# ran out of cash 0.001, side income month 36 RM 11,240只保留工作,期末平均有 RM 110,540,有 0.037 的模拟在失业后现金耗尽。在工作之外每月投入 40 小时做副业,期末平均 RM 241,533,几乎从不耗尽现金(0.001)。现在就辞职全职做副业,平均值几乎相同(RM 241,689),却有 0.399 的模拟现金耗尽:在市场给出答案之前辞职,等于把积蓄押在这个答案上。模型中最好的规则是:等副业连续 3 个月覆盖生活费之后再辞职,平均 RM 264,410,现金耗尽的概率与稳健路径一样只有 0.001。这条规则有效,是因为它让市场先揭示产品是否对路,然后才放弃工资。
一个人的收入组合
一个人的收入可以把波动大、毛利高的产品与较稳定的客户工作组合起来。下面设创始人每月有 120 小时;一条产品线占 30 小时,平均每月净赚 RM 6,000,标准差 4,000;一份 B2B 自动化顾问月费占 40 小时,RM 5,000(标准差 1,000);带联盟收入的内容占 20 小时,RM 1,500。月收入低于 RM 6,000 算作短缺月。
推导
# Chapter 6. p12: A solo portfolio and the months that fall short.
# Founder hours: 120 a month. Units of work: a high-ticket digital
# product (30 h, net RM 6,000 a month, sd 4,000, lumpy launches), a B2B
# automation retainer (40 h, RM 5,000, sd 1,000) and content with
# affiliate income (20 h, RM 1,500, sd 800, correlation 0.5 with the
# product). At most two product lines and two retainer clients are
# assumed manageable. Income below RM 6,000 is a short month. Each
# mix using 101 to 120 hours is simulated for 12 months x 5,000 runs
# (seed 90); mean monthly income and the share of short months.
import itertools
import math
import random
UNITS = (("product", 30, 6000, 4000), ("retainer", 40, 5000, 1000),
("content", 20, 1500, 800))
def simulate(mix, rng, months=60000):
total = short = 0.0
for _ in range(months):
z1, z2, z3 = rng.gauss(0, 1), rng.gauss(0, 1), rng.gauss(0, 1)
zc = 0.5 * z1 + math.sqrt(1 - 0.25) * z3
x = (mix[0] * max(0.0, 6000 + 4000 * z1) +
mix[1] * max(0.0, 5000 + 1000 * z2) +
mix[2] * max(0.0, 1500 + 800 * zc))
total += x
short += x < 6000
return total / months, short / months
if __name__ == "__main__":
rows = []
for mix in itertools.product(range(3), range(3), range(7)):
h = sum(m * u[1] for m, u in zip(mix, UNITS))
if 0 < h <= 120 and h > 120 - 20:
rng = random.Random(90)
rows.append((mix, h) + simulate(mix, rng))
print(" product retainer content hours mean RM short months")
for mix, h, mean, sh in sorted(rows, key=lambda r: -r[2]):
print(" {} {} {} {:3d} {:7,.0f} {:.3f}"
.format(mix[0], mix[1], mix[2], h, mean, sh))
# product retainer content hours mean RM short months
# 2 1 1 120 18,749 0.048
# 2 0 3 120 16,769 0.131
# 1 2 0 110 16,116 0.002
# 1 1 2 110 14,136 0.037
# 0 2 2 120 13,011 0.002
# 1 0 4 110 12,156 0.162
# 0 1 4 120 11,031 0.064
# 0 0 6 120 9,052 0.266两条产品线、一份顾问月费、一条内容线,平均收入最高(每月 RM 18,749),短缺月占 0.048。一条产品线加两份顾问月费,平均 RM 16,116,短缺月只有 0.002。没有顾问月费的产品组合风险更大:两条产品线加三条内容线平均 RM 16,769,却有 0.131 的月份短缺。只做内容在两方面都最差(RM 9,052,0.266)。顾问月费买的是安全,产品买的是上行空间。
假说 · 尚未验证
作者的诠释(尚未验证): 在打造产品的同时保留至少一个长期 B2B 客户的个体创业者,头两年里收入低于生活费的月份,少于只做产品的创业者,而两者平均收入相近。什么会证明它错: 在个体创业者的月收入记录中,有长期客户者的短缺月份与没有者一样多。
小结
在副业证明自己之前辞职,平均结果与「保住工作、业余经营」相同,却有 0.399 的概率现金耗尽。等副业连续 3 个月覆盖生活费再辞职,平均结果最好(RM 264,410),风险只有 0.001。在个体组合中,顾问月费把短缺月从 0.131 降到 0.048 或更低。
Fan, Y. (2026). One-Person Enterprise Automation, Avatar Livestream Funnels and Task-Exposure Arbitrage: Coasean Transaction-Cost Collapse, Cyclic Script Retention and High-Margin Solopreneur Portfolios [一人公司自动化架构、数字人直播漏斗与任务级人工智能套利] (Version 1.0.0). 5D Library. https://doi.org/10.5281/zenodo.24090235 · Licensed under Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International (CC BY-NC-ND 4.0).