OpenRouter
The routing layer for sending each task to the right model and economizing tokens by job, value, and risk.
Open OpenRouter docs
TOPIC 04
Recurring theme · Coined in Ep. 3
Tokens, spend, and routing.
"Token maxing" is the show's name for the question every AI-native operator hits: when is burning more tokens brilliant, and when is it waste? This guide collects the episodes, timestamps, tools, and the operating rule the hosts landed on — route spend by proximity to the work.
The definition
Token economics is the question of where AI spend actually creates value. Tokens are the unit of intelligence you buy; token maxing is the bet that spending more of them — on more experiments, more drafts, more agents — compounds into real output. The counterweight is routing: sending extraction, summaries, and simple drafts to cheap or local models, and reserving frontier reasoning for architecture, judgment, and high-stakes work.
The show's answer evolved across episodes. Episode 3 coined the frame and separated founder-led exploration from quota-driven token burn — incentives decide which one you get. Episode 4 resolved it into a proximity rule: stay on the frontier when you are close to the work and seeing value; route and optimize when you are far from usage and signing the check. Episode 5 extends the comparison to open models and the practical cost-capability tradeoff. The KPI remains real usage and good products, not tokens burned.
EpisodesWhere it shows up
Ep. 3
Token Maxing and the Company Brain Era
The anchor episode
39:35
Ep. 4
The Fastest Feedback Loop Wins in the AI Era
The resolution
1:02:09
Ep. 5
AI, Creative Workflows, and the New Skill Stack
Open models
1:01:05
Ep. 1
AI Agents and Knowledge Management: Building a Company Brain
The review cost
1:25:51
Token Maxing and the Company Brain Era
The anchor episode
39:35
Ep. 4
The Fastest Feedback Loop Wins in the AI Era
The resolution
1:02:09
Ep. 5
AI, Creative Workflows, and the New Skill Stack
Open models
1:01:05
Ep. 1
AI Agents and Knowledge Management: Building a Company Brain
The review cost
1:25:51
Best timestampsJump straight in
Ep. 3 · 03:29OpenRouter, model costs, and engineering tokens — the cost-optimization problem↗
Ep. 3 · 16:30Token maxing and company incentives — healthy spend versus quota burn↗
Ep. 3 · 23:34Subscription pricing changes behavior — quota psychology↗
Ep. 4 · 23:22Token maxing and measuring real output — users and value, not burn↗
Ep. 4 · 29:53Model routing and who should optimize — the proximity rule↗
Ep. 4 · 42:01Model routing as the real unlock — tenfold savings, same result↗
Ep. 1 · 53:13Token spend, agent experiments, and report generation — the human review cost↗
Related toolsThe routing stack
Ollama
The free end of the spectrum: open models on your own machine for cheap, high-volume, private, or offline work.
Open OllamaDeepSeek
The open-source frontier example from Episode 4: roughly six months behind, nearly free, improving aggressively.
Open DeepSeekFactory (Droid)
The model-routing reference for building engineering workflows across many models while spending far less.
Open FactoryCodex
One pole of the spend debate: frontier coding agents that are worth the tokens when you are close to the work.
Open Codex docsClaude Code
The other frontier harness in the comparison, running long development tasks where deep reasoning earns its cost.
Open Claude Code docsOpen questionsStill being argued
Does model value trend to pennies?
Episode 4 floats it directly: open source sits months behind the frontier at near-zero cost. If intelligence keeps getting cheaper, the brutal economics land on the labs — and routing gets more valuable, not less.
What is the right KPI for AI spend?
Both token-maxing episodes reject burn as a metric. New users, shipped product, and reviewed value are the proposed measures — but nobody has a clean dashboard for "value per token" yet.
When does the quota change the person?
Episode 3's subscription insight cuts both ways: a paid plan frees people to experiment, and it also tempts them to perform usage. Where incentives meet psychology is still the messiest part of the topic.
How much review does exploration owe?
Episode 1's version of the problem never went away: every token spent on agent output creates human review time downstream. Exploration that nobody inspects is just expensive noise.
“
Close to the work, stay on the frontier and measure value. Far from it, route models and optimize spend. The KPI is real usage and good products, not tokens burned.
Shoreline Ep. 4 · distilled operating principle