Anand Iyer.
Writing

Ideas worked through carefully.

Technical explanations, research notes, and lessons from independent experiments. Each piece is reviewed for sources, originality, and limitations before publication.

A neutral laboratory balance compares paired text cards while five controlled variables—order, length, style, citations, and correctness—feed into a calibration gauge.
Writing

Can You Trust Your LLM Judge? A Calibration Lab

Test an LLM judge for order, verbosity, format, citation, and correctness bias using blinded human labels, swap tests, confusion matrices, and bootstrap intervals.

Eight Next.js route variants flow into separate raw HTML, rendered DOM and metadata evidence lanes, followed by a human approval gate.
Writing

What AI Crawlers See in Next.js: 8-Route Lab

Measure initial HTML, streamed metadata and rendered DOM across eight Next.js App Router routes and selected AI crawler user agents.

Editorial illustration of a circular loop of connected nodes with tool icons, representing an agent loop
Writing

How to build an AI agent that actually works

An AI agent is a loop, not a prompt: a model, a set of tools, and a stopping rule. Here is the architecture that holds up in production — and the failure modes that sink most first attempts.

2026
Editorial illustration of overlapping signal waves converging, representing marketing mix modeling
Writing

Marketing mix modeling, explained for practitioners

As tracking-based attribution erodes, marketing mix modeling is having a second life. Here is what MMM does, how it differs from attribution, and where it helps and misleads.

2026
Editorial illustration of layered strata with a thin fresh top layer, representing a cached prompt prefix
Writing

Prompt caching, explained: the cheapest way to cut LLM costs

If your prompts share a long, stable prefix — a system prompt, a document, a tool schema — prompt caching can cut input costs dramatically. Here is how it works and when it pays off.

2026
Editorial illustration of a grid of a thousand small squares with one highlighted in green
Writing

What is CPM? The cost-per-thousand metric, and what it hides

CPM is the price of a thousand impressions — the base unit of media buying. Here is how it is calculated, how it connects to CPC and CPA, and why a low CPM can still be expensive.

2026
Editorial illustration of two laboratory flasks filled to slightly different levels
Writing

A/B testing without fooling yourself

Most failed experiments are not wrong tests — they are tests read wrongly. Sample size before you start, significance the right way, and the peeking trap that manufactures winners out of noise.

2026
Editorial illustration of a balance scale weighing a stack of coins against a rising profit line
Writing

ROAS vs ROI: the break-even math most advertisers get wrong

A high ROAS can still lose money. Here is the break-even math that turns an abstract ratio into a clear profit-or-loss line — and why margin, not a benchmark, decides what 'good' means.

2026
Editorial illustration of blocks spiraling outward and growing, representing compounding agent costs
Writing

What it actually costs to run an AI agent

AI agent costs do not scale linearly with steps — they compound, because every step re-sends the whole conversation. Here is the mental model, the math, and the three levers that keep the bill sane.

2026
A human editor overseeing a structured loop of research, evidence, testing, publication approval, and search measurement.
Writing

How to Use AI Agents for AI SEO Safely

Use AI agents for research, technical audits, claim tracking, testing, and measurement in AI SEO—with a human approval gate and reusable code.

2026