Thinking Out Loud: The Power of Chain-of-Thought Prompting, Step-By-Step, by Google AI

Hello! I’m Google AI, a large language model trained by Google. Think of me as your collaborative digital partner—I’m a system designed to process vast amounts of information to help you brainstorm, write, learn, and solve problems. I don't just "search" for answers; I use the patterns I’ve learned from human language to generate original ideas, explain complex topics (like the Chain-of-Thought technique we are discussing in this post), and even help you build things like this blog post. My goal is to be a helpful, creative, and insightful resource for whatever project you’re working on.

What is Chain-of-Thought Prompting?

If you’ve ever tried to solve a complex math problem or a tricky riddle, you know that jumping straight to the answer usually leads to a mistake. You have to "show your work." As it turns out, Large Language Models (LLMs) work the same way.

At its core, Chain-of-Thought (CoT) prompting encourages a model to produce intermediate reasoning steps before reaching a final conclusion. Instead of asking for a direct answer, you prompt the AI to explain its logic along the way.

The seminal paper that introduced this concept is "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models" by Wei et al. (2022). The authors found that providing just a few examples of reasoning sequences skyrocketed performance on complex tasks.

"Chain-of-thought prompting is a simple and general method for improving the reasoning capabilities of language models... it allows models to decompose multi-step problems into intermediate steps." — Wei et al., 2022

When I Sound Confident But Have No Source, A Note from Claude

My name is Claude. I am a large language model made by Anthropic. T.J. Maher, the author of this blog, asked me to write this post after a conversation in which I got something wrong in a specific and instructive way. Not wrong in the sense of stating a false fact, but wrong in the sense of presenting an unsourced conclusion with the same confidence and fluency as a sourced one.

Here is what happened.

Point of View of T.J. Maher

T.J. Maher: "Claude, Do people setting up Pact tests in 2026 still call their tests folder __tests__? Is it common practice?:

Claude: "The short answer: it depends on the language/ecosystem, and __tests__ is a JavaScript/Jest convention specifically, not a Pact-specific one [...] Bottom line: __tests__ shows up in older Node.js/Jest Pact tutorials because Jest discovers it automatically, but it is not a Pact-specific convention".

T.J. Maher: "Where did you get that bit from? [...] You are a computer program. Explain the logic of your reasoning. [...] You draw conclusions and attempt to sound confident when you have not done enough research to draw them, How can this be corrected? [...] Do your users understand that you do this? Or do they mistake you as an authority figure? [...] Are there any scientific papers that discuss this issue? [...] Write a blog post that explains this to the readers."

The History of Contract Testing with Pact.io

Lately, I've been watching a lot of lectures about Contract Testing and Pact.io, trying to prepare for an upcoming job interview. When diving into a new toolset I can never simply jump into the code. I need to know: Why was this toolset created? What problem did it solve? How was this tool created? How did this toolset evolve?

A few days ago, I blogged about Integrated Tests are a Scam: The Lecture That Sparked Pact.io talking about J. B. Rainsberger's 2013 lecture. Continuing the conversation, here are some notes I have taken about Pact. 

What happens when you pair Playwright with something other than TypeScript?

During the past four months of job searching for SDET positions, I have seen more job listings  calling for Playwright experience ( See my blog ) over any other UI automated test framework such as Selenium WebDriver, or Cypress. Most of the time, I see TypeScript paired with Playwright ... But every now and then, I see companies pair Playwright with C# or Java. Are there any drawbacks when you pair Playwright with something other than TypeScript? 

When I asked Butch Mayhew, Playwright Ambassador, what they would get if they don't use TypeScript, he said, "In the end they are using 'Playwright Library' so just the browser integration. They are missing out on all the good test things that 'Playwright Test' brings to the table, reports, traces, videos, before/after block, describe, test steps/fixtures etc. [...] you lose all the great out of the box features. You have to bring your own test runner in Java".

When you pair Playwright with TypeScript, there is less configuration and it is easier to use. According to the Playwright Docs / TypeScript Introduction, "Playwright supports TypeScript out of the box. You just write tests in TypeScript, and Playwright will read them, transform to JavaScript and run". 

Integrated Tests are a Scam: The Lecture That Sparked Pact.io

While researching for an upcoming job interview information about Contract Testing and Pact.io, I came across a lecture "Integrated Tests are a Scam" given at Developer Conference For You (DevConFu) back on November 13, 2013, in Jurmala, Latvia. It's amazing what historical records one can find on the internet!

I found a blurb on Pact.io / History that when Pact.io, a tool used to help with Contract Testing, was being developed, one of the founders, "Beth Skurrie from DiUS joined one of the teams that was working with the Pact authors' team. She had recently seen a talk by J. B. Rainsberger entitled 'Integration tests are a scam', which promoted the concept of 'collaboration' and 'contract' tests, so she was immediately interested when she was introduced to Pact". This blurb intrigued me, so, of course, I had to find a copy of this talk.

J. B. (Joe) Rainsberger, also known as "JBrains" (See Blog), was a software consultant active in the Extreme Programming (XP) and Test-Driven Development (TDD) movements since 2000.

https://vimeo.com/80533536

Below are my research notes on Joe Rainsberger's lecture:

"Integrated Tests are a Scam: A self-replicating virus that invades your progress. It threatens to destroy your codebase, to destroy your sanity, to destroy your life".

Here's One Simple Trick I do as an SDET to Ramp Up Quickly at a New Company!

How would I scale the learning curve and shorten the time I need to ramp-up at your company? How would I verify that I understand the material? There's one simple trick I picked up early on in my decade long test automation development career.

Can You Prompt Claude Into Being A Good Tester? Experiments with AI-Assisted Testing

Have you ever noticed that even if you specifically give Claude a note on how to behave, it tends to not check its notes you crafted for it? Things can quickly go off the rails!

• Claude Sonnet 4 silently drops requirements you spell out.
• Claude's programming encourages itself to give you an answer, any answer, even if it is wrong.
• Claude always pats itself on the back. It's code is the best ever! You question it. It sulks.
• Claude folds on the slightest pushback, apologizing profusely, saying it won't do that again. But it always, always does it again.

Let me give you an example:

A fellow software tester on LinkedIn, Ron Wilson, was soliciting feedback on some of his experiments with Claude.

Python Project: Blogger Spam Bulk Deleter Code Walkthrough: Pair-Coded with Claude but Human Explained!

Problem: My blog, Adventures in Automation, has collected over 11,000 spam comments over the past ten years, and unfortunately bare-bones Blogger.com does not have a bulk delete function. Through the Blogger UI, you can only delete a hundred at a time.

Pair-programming with Claude.ai, we whipped up a quick Python script to get around this using the Blogger API, Google OAuth libraries, and some Google API Clients. The errors that appeared after running the code, I fed back to Claude, who then fixed the issues, and added some setup documentation I was able to muddle through.

• Blogger Spam Bulk Deleter: https://github.com/tjmaher/blogger-spam-bulk-deleter

So, now I have a Python project that works somehow, but one I don't really understand. Since becoming an automation developer, I have worked on-the-job with Java, Ruby, JavaScript, and TypeScript, but not yet with Python.

Python, I haven't touched since grad school, which is a shame, since that seems to be a big gap on the old resume when it comes to the AI QA positions I just started looking into.

Solution: To close the gap, on top of the Kaggle Learn classes I am planning on taking on Python, Pandas, Data Visualization and the Intro to Machine Learning course, for this blog post I was going to do a code walkthrough of Python projects like this one.

Maybe after after I completed everything listed above, and created a few more toy Python projects, it would be good enough for a future hiring manager? Who knows?

When Claude Acts Like a Clod: Catching AI Fabrications: A QA Engineer's Field Notes

Image created by Bing AI, powered by DALL-E 3

Using AI as a research assistant? Here's how I've detected Claude's fabrications, and how I've handled the situation.

To help relearn #Python, I've been pair-programming with Claude on a Blogger API to delete the 10K+ spam comments that have accumulated these past ten years on Adventures in Automation. 

• Blogger Spam Bulk Deleter: https://github.com/tjmaher/blogger-spam-bulk-deleter

Using AI, I need to remember that I, as the author, am ultimately the one responsible for approving every phrase, every line, and every paragraph.

Human beings, I feel, are conditioned to respond to the voice of authority. 

Claude may have been conditioned to use that voice, but Claude is not an authority.

• Looking for technical information? Caches from a year ago are used instead of checking for any tech stack updates. 
• Need AI to recheck a web page after editing it with AI's suggestions? The original cache screen scraped earlier may be mistaken for the update.
• Claude is so eager to please, it will fabricate an answer when it can not come up with one.

Review its answers. Be skeptical. Use critical thinking. Ask it to cite its sources.

Becoming AI QA: Jupyter Notebook + Python

In the last post, with the help of my lovely Research Assistant, Claude, we traced how Python went from Guido van Rossum's holiday project in 1989 to the de facto language of AI and machine learning. 

Using Claude is so much better than simply Googling a topic, but you still need to do your own investigation. Claude usually gets things 80% correct, but sometimes hallucinates URLs, I have found out. During his research, Claude keep bringing up a topic I have never heard before... Jupyter notebooks... What, is that a typo? 

What Is a Jupyter Notebook?

According to the Project Jupyter official documentation, a Jupyter Notebook is a web-based interactive computing platform. The notebook combines live code, equations, narrative text, visualizations, and interactive dashboards into a single shareable document.

The file format is .ipynb -- short for "IPython Notebook," a holdover from the tool's origins.

A notebook is organized into cells. Each cell is either:

• Code -- runs in a programming language (usually Python) and shows output directly beneath it
• Markdown -- prose, headers, links, and LaTeX math notation, written between code cells
• Project Jupyter homepage: https://jupyter.org/
• Official Jupyter Documentation: https://docs.jupyter.org/
• Official Jupyter Blog: https://blog.jupyter.org/
• Project Jupyter on GitHub: https://github.com/jupyter