Sign up to save your podcasts
Or
Share Episode 16 - AI Loops...This is the way
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
Episode 16 - AI Loops...This is the way
AI loops are changing how people get real work done with agents.
In Episode 16 of Not Brothers, Mark and Ryan break down what “loops” actually mean in practice — not as vague AI hype, but as a better way to structure work: define the goal, write the success criteria, pair the builder with a reviewer, and let the system iterate until the work meets the bar.
Ryan shares an example of a development task that ran for hours with a builder agent and review agent working together. Instead of babysitting every step, he gave the system clear win conditions and let it loop through implementation, critique, and revision.
They also talk about why this is not really new. Good AI workflows are starting to look a lot like good human workflows: clear requirements, tight scope, explicit assumptions, and a shared definition of done.
The episode moves beyond software development into creative and marketing work — including social content, clips, transcripts, visual QA, and repeatable production workflows. The big idea: loops work best when they are simple, scoped, and reviewed.
Chapters:
00:00 — What are AI loops?
00:27 — Why loops are not exactly new
01:10 — From prompt-by-prompt work back to real requirements
02:58 — OpenClaw, cron jobs, and early agent loops
03:35 — GPT-5.5 and goal-oriented prompting
04:55 — A four-hour builder/reviewer agent loop
06:03 — Chunking work versus letting loops run
06:44 — Good AI workflows look like good human workflows
08:34 — Why nuance and hidden requirements still matter
09:53 — The review agent is the critical piece
10:40 — Win conditions tell the loop when to stop
12:29 — The value shifts from typing to thinking
14:01 — Solving the wrong architecture before code exists
14:54 — Why knowledge work is undervalued
15:59 — The boiler repairman and the value of experience
17:13 — Value-based billing and technology leverage
19:18 — The dopamine trap of instant AI iteration
19:43 — Prototyping first, then turning reactions into requirements
20:11 — AI can help write specs, but you still have to read them
21:50 — Applying loops to creative and social content
23:34 — Getting to 90% before humans review
24:35 — Using transcripts as requirements for content loops
25:28 — Stop staring at the harness and start defining the contract
26:26 — Start with small loops, not a 50-agent circus
27:21 — Simple workflow loops like standups, notes, and PR rebases
27:53 — Wrap-up
View all episodes
By Mark Hughes, Ryan HughesAI loops are changing how people get real work done with agents.
In Episode 16 of Not Brothers, Mark and Ryan break down what “loops” actually mean in practice — not as vague AI hype, but as a better way to structure work: define the goal, write the success criteria, pair the builder with a reviewer, and let the system iterate until the work meets the bar.
Ryan shares an example of a development task that ran for hours with a builder agent and review agent working together. Instead of babysitting every step, he gave the system clear win conditions and let it loop through implementation, critique, and revision.
They also talk about why this is not really new. Good AI workflows are starting to look a lot like good human workflows: clear requirements, tight scope, explicit assumptions, and a shared definition of done.
The episode moves beyond software development into creative and marketing work — including social content, clips, transcripts, visual QA, and repeatable production workflows. The big idea: loops work best when they are simple, scoped, and reviewed.
Chapters:
00:00 — What are AI loops?
00:27 — Why loops are not exactly new
01:10 — From prompt-by-prompt work back to real requirements
02:58 — OpenClaw, cron jobs, and early agent loops
03:35 — GPT-5.5 and goal-oriented prompting
04:55 — A four-hour builder/reviewer agent loop
06:03 — Chunking work versus letting loops run
06:44 — Good AI workflows look like good human workflows
08:34 — Why nuance and hidden requirements still matter
09:53 — The review agent is the critical piece
10:40 — Win conditions tell the loop when to stop
12:29 — The value shifts from typing to thinking
14:01 — Solving the wrong architecture before code exists
14:54 — Why knowledge work is undervalued
15:59 — The boiler repairman and the value of experience
17:13 — Value-based billing and technology leverage
19:18 — The dopamine trap of instant AI iteration
19:43 — Prototyping first, then turning reactions into requirements
20:11 — AI can help write specs, but you still have to read them
21:50 — Applying loops to creative and social content
23:34 — Getting to 90% before humans review
24:35 — Using transcripts as requirements for content loops
25:28 — Stop staring at the harness and start defining the contract
26:26 — Start with small loops, not a 50-agent circus
27:21 — Simple workflow loops like standups, notes, and PR rebases
27:53 — Wrap-up