ConceptContent and voice with AITo understand3 min
What Tier 0 is: the diagnosis before writing
The step that classifies the reader before any creative decision, what it measures and what changed when it became mandatory in an AI copy pipeline.
ReadLab
How to design, build and verify AI systems.
Concepts to understand, guides to apply and experiments with numbers from real systems. All written from what I build.
17 pieces · 3 concepts · 4 guides · 9 experiments · 1 case · updated Sep 28, 2026
Each column is a topic. Height = number of published pieces. Click a column to filter.
ConceptContext engineeringTo understand2 min
What context engineering is
The discipline of deciding what an AI model receives along with the instruction, and why it matters more than the prompt in agent systems.
ReadConceptMulti-agent systemsTo understand2 min
What a multi-agent system is
When splitting work across several AI agents pays off, the three ways to organize them and where these systems usually fail.
ReadGuideMulti-agent systemsTo apply3 min
When to use an agent and when to use a skill
The rule "agents think, skills execute", where it came from and how to decide at each step of an AI system.
Read
This is weekly.
One email a week with the densest experiments, filtered and commented. Nothing duplicated from the portal: the context that does not fit in a record.
About SinalGuideVerification and qualityTo apply2 min
How to know whether an AI answer is right
The four levels of verification, from automated test to self-assessment, and how to pick the right level for each step of a system.
ReadGuideCost and modelsTo apply2 min
How to cut Claude Code costs without losing quality
Measure before cutting, route tasks by model and automate the rules. With the numbers from 987 real sessions.
ReadGuideContent and voice with AITo apply2 min
How to keep one brand voice across several AI agents
A voice file built on limits, loaded into every agent, and a score that fails on its own. The method that kept 10 agents writing like one person.
ReadCaseCost and modelsTo go deeper
Case: cognitive-claude
987 sessions in 90 days. How a rules layer splits work across Opus, Sonnet and Haiku, what it costs and what the number does not prove.
ReadExperiment #009Content and voice with AITo go deeper2 minValidated
1 voice file, 10 agents, 9 pages in the same voice
A voice DNA file produces brand consistency even when 10+ different agents write the copy.
ReadExperiment #008Multi-agent systemsTo go deeper2 minInconclusive
Framework + AI: purchase decisions without confirmation bias
A structured decision framework with a dual mode (pure framework and AI) produces better purchase decisions than intuition or ad-hoc research.
ReadExperiment #007Context engineeringTo go deeper2 minValidated
14 sources, 9 agents, 1 discipline: context engineering as a squad
Context engineering frameworks compiled into a squad produce better agent systems than ad-hoc prompting.
ReadExperiment #006Multi-agent systemsTo go deeper2 minValidated
6 stages to turn knowledge into a product
A standard 6-stage pipeline turns any knowledge base into a marketplace-ready squad.
ReadExperiment #005Design systemsTo go deeper2 minValidated
171 tokens, 4 exports: a design system as living infrastructure
A design system built on DTCG semantic tokens produces more visual consistency than descriptive brandbooks.
ReadExperiment #004Content and voice with AITo go deeper2 minValidated
Shared narrative DNA: coherence without rigidity
A conceptual DNA document shared across every page creates narrative coherence without flattening the voice.
ReadExperiment #003Verification and qualityTo go deeper2 minValidated
2 parallel auditors catch what 1 sequential misses
Two QA agents running in parallel find more problems than one sequential auditor.
ReadExperiment #002Content and voice with AITo go deeper2 minValidated
Diagnosis before copy: what Tier 0 changes
Mapping the reader's awareness and sophistication before writing changes routing decisions and the final result.
ReadExperiment #001Content and voice with AITo go deeper2 minValidated
10 copywriters, 1 pipeline, 9 pages in 24h
A structured pipeline of specialist agents produces higher-QG copy than a single generalist agent.
Read