Practical signal on agentic AI

Researching, building, and testing what actually works in agentic AI.

Field notes, product experiments, and practical frameworks for leaders and builders navigating the next wave of AI. Less hype, more useful signal.

Subscribe to the Newsletter → Explore Products

Research Products Book Builder Notes

What we do

A lab for practical agentic AI.

Research & Content

Field notes and analysis on agentic AI, AI commerce, and the systems reshaping how work gets done.

◨

Product Lab

Small AI products, real user problems, fast experiments. FitSwap, Menu Visualizer, NeuraCast.

⌘

Claude Code Book

A beginner-friendly guide to practical AI — for professionals who don't want to become programmers.

Featured signal

This week's useful reads.

View all →

Agentic AI 8 min read

Agentic Commerce: How AI shopping agents are quietly redrawing retail.

Autonomous buyers don't browse — they negotiate, route, and commit. A field guide to the stack, the economics, and what breaks when the shopper isn't human anymore.

Read analysis →

Market Signal 6 min

The commodity cycle: every AI API you pay for will eventually be free.

Read →

AI Safety 7 min

When agents go rogue: real failures, and the guardrail patterns that stop them.

Read →

The product lab

Not every idea needs a deck. Some need a prototype.

View the lab →

FitSwap

Live

Shoppers can't visualize clothing on their own body before buying — high returns, low confidence.

Consumer·Agentic commerce

Menu Visualizer

Live

Text menus don't tell diners what the plate actually looks like. Visual menus change the order.

Consumer·Vision + generative

NeuraCast

In development

Podcast workflows still need too many humans. NeuraCast runs the loop end-to-end.

Creator·Agentic media

Claude Code for Non-Programmers (2nd Edition) by Vikas Sah

Book · 2nd Edition · Now on Amazon

What if your stack already ran itself by the time you opened your laptop?

Maya hires a small staff of named specialists (Iris, Atlas, Reuben, Echo, Pheme). By Chapter 27, nine jobs run unattended. None of them happen in a chat window. The operator's manual for Claude Code, written for non-programmers.

27 chapters 9 appendices Kindle $21.99 5.0★ on Amazon

Get the 2nd Edition on Amazon → See the full overview

The lab notebook

What I'm researching, building, and testing — right now.

Researching 3 threads

Agentic CommerceHow shopping agents negotiate, route, and commit — and where the protocol layer is missing.

EvaluationA practical eval rubric for multi-step agents that doesn't require a rollout harness.

Market SignalThe commodity curve for frontier APIs — what stays valuable when inference is free.

Building 3 active

NeuraCastEnd-to-end agentic podcast pipeline — research → script → voice → publish.

Vibe Guard SkillsSafety skills that run pre-commit on AI-generated code: security, resilience, comprehension.

CCC ScreenerWeekly small-cap alpha screener with self-learning weights and kill-switch monitoring.

Testing live signals

FitSwapDoes virtual try-on reduce return rate enough to justify inference cost? Early data: yes.

Menu VisualizerDo generated plate images change ordering behavior? Running against pilot restaurants.

Claude Code SkillsWhich skills get used at 10× rate vs. created-and-abandoned. Shapes the book's appendix.