4 openclaw agents across 3 machines -- and i can't write a line of code. in this episode, i break down the exact folder structure, config files, and architecture that makes multi-agent actually work. because the thing nobody tells you? running 2 openclaws on 1 mac mini isn't about installing it twice. it's one gateway, two named agents, two workspace folders.

i
 
a
l
s
o
 
w
a
l
k through the 5-file setup to switch your agents to gpt-5.4, which dropped two days ago at half the price of claude opus 4.6. plus the real reason i run 4 agents -- it has nothing to do with productivity and everything to do with tinkering being my ultimate form of expression.

fair warning: there's some australian english in this one. gpt-5.4 called me a good cunt and that's when i knew the model was different.

Chapter Markers:

0
0:00 Intro - the misconception about openclaw

0
0
:
5
5
 
t
h
e 4-agent fleet breakdown (aria, clarity, sammy, gary)

0
1
:
3
0
 what an openclaw actually looks like (folder structure reveal)

0
3
:
0
0
 adding a second agent to one mac mini

0
4
:
3
0
 
gpt-5.4 pricing and context window breakdown

06:30 the 272K pricing threshold nobody mentions

0
7
:
30 gpt-5.4 called me a good cunt

0
8:30 5-file config walkthrough to switch to gpt-5.4

1
0:30 pattern recognition beats coding

1
2
:
00 why tinkering is the real reason i have 4 agents
14:00 copy-paste prompts for multi-agent + gpt-5.4

S
h
o
w
 
N
otes:

OpenClaw -- open-source AI agent framework (openclaw.ai)

G
PT-5.4 -- OpenAI's frontier model ($2.50/M input tokens)

T
hreadify -- threads automation tool (https://www.threadify.app/?utm_source=lenny-podcast)

M
ac Mini -- local agent hosting
Telegram -- agent communication layer