Sarah Pan (@sarahpan)

When AI Models Disagree, Don’t Use Majority Vote

Asking a second AI model is one of the easiest ways to check an uncertain answer. But what happens when the second model disagrees with the first? You ask a third model. Now one model says A, another

Jul 23 · 6 min readsarahtokenbay.hashnode.dev

Building a Model-Agnostic Agent Harness: What Breaks When You Switch LLMs

Six months ago, I made a bet on a project: let users pick their LLM. Three months later, I realized the dropdown was the easy part. This is what the hard part looked like. The Model Changed. Everythin

Jul 20 · 7 min readsarahtokenbay.hashnode.dev

The API Call Succeeded. The Workflow Still Broke.

We're building TokenBay, a unified API that routes requests across GPT, Claude, Gemini, DeepSeek, and a few dozen other models. A few weeks ago we ran a fairly boring internal test: send the same stru

Jul 16 · 6 min readsarahtokenbay.hashnode.dev

The Best AI Model Is Usually the Wrong Question

Developers keep asking which AI model is the best. It sounds like a reasonable question. There are leaderboards, benchmark charts, side-by-side comparisons, and a new “best model” every few weeks. But

Jul 15 · 5 min readsarahtokenbay.hashnode.dev

Three AI APIs Now Look Almost Compatible. They Still Aren’t Interchangeable.

OpenAI, Claude, and Gemini now look surprisingly similar at the API level. In some cases, switching providers requires little more than changing an API key, a base URL, and a model name. But this comp

Jul 14 · 4 min readsarahtokenbay.hashnode.dev

Sarah Pan

About

Available for

Sarah Pan's blogs

When AI Models Disagree, Don’t Use Majority Vote

Building a Model-Agnostic Agent Harness: What Breaks When You Switch LLMs

The API Call Succeeded. The Workflow Still Broke.

The Best AI Model Is Usually the Wrong Question

Three AI APIs Now Look Almost Compatible. They Still Aren’t Interchangeable.

Sarah Pan

About

Available for

Sarah Pan's blogs

When AI Models Disagree, Don’t Use Majority Vote

Building a Model-Agnostic Agent Harness: What Breaks When You Switch LLMs

The API Call Succeeded. The Workflow Still Broke.

The Best AI Model Is Usually the Wrong Question

Three AI APIs Now Look Almost Compatible. They Still Aren’t Interchangeable.

Sarah Pan

About

Available for

Sarah Pan's blogs

When AI Models Disagree, Don’t Use Majority Vote

Building a Model-Agnostic Agent Harness: What Breaks When You Switch LLMs

The API Call Succeeded. The Workflow Still Broke.

The Best AI Model Is Usually the Wrong Question

Three AI APIs Now Look Almost Compatible. They Still Aren’t Interchangeable.

Search Hashnode

Sarah Pan

About

Available for

Sarah Pan's blogs

When AI Models Disagree, Don’t Use Majority Vote

Building a Model-Agnostic Agent Harness: What Breaks When You Switch LLMs

The API Call Succeeded. The Workflow Still Broke.

The Best AI Model Is Usually the Wrong Question

Three AI APIs Now Look Almost Compatible. They Still Aren’t Interchangeable.