034 • 19 August 2025

Contents of this Newsletter

🛣️

How We Got Here

⏰ : 5 min

📓

Blog Picks

⏰ : 10 min

👀

Perplexity Eyes Google Chrome

⏰ : 7 min

🗞️

Weekly AI News

⏰ : 9 min

MAIN STORY

QA at the Speed of AI: How We Got Here?

Software is shipping faster than ever, but QA hasn’t kept pace. Scripts still dominate automation, and they often break under real-world changes. From the start, we set out to build AI that tests the way a human QA does—understanding intent, adapting mid-run, and validating the experience behind the app.

First Steps: Reporting Bugs Better

Our journey began with Quash Report, an open-source SDK for manual testing. Shake your device, and you’d instantly capture logs, screenshots, session recordings, and even an AI debugging suggestion. Over 100 developers used it actively, and it gave us the confidence that context-rich QA was a real need.

Chasing Automation: Code as Context

Next came Quash Automate. We assumed code was the ultimate source of truth: differentiate the feature branch from master, generate test cases, and validate workflows. It worked to a point but the flaw was clear: code only shows what was built, not what was meant to be built. That gap in intent became our biggest challenge.

Learning from QAs, Learning from AI

We tried to mimic how QAs think: starting from business use cases and mapping them to workflows. That led to our PRD-to-test-case generator that surprised us with strong results, further refined with Figma integration, which became our best intent source yet. Not long after, BrowserStack shipped the exact same thing. For us, that was proof we were defining the direction of QA and it also made something else obvious: test case generation alone was never going to be enough. This was because QAs don’t just generate test cases. They interact with apps dynamically just as users learned to interact with AI tools like ChatGPT. Static workflows missed that fluidity.

Building Execution That Thinks

We shifted focus to execution. First, we tried rule-based flows which ended up being too brittle. Then we let LLMs decide actions screen by screen. That worked, until we hit the stateless wall. Every AI call forgot the last screen. No memory, no progress.

The breakthrough came with episodic memory: a stateful layer that tracked goals, preconditions, actions, and history. Suddenly, the AI had a presence of mind and it could move through flows like a real QA.

That system became Mahoraga, our execution engine. The name is in direct correlation with the anime fighter who can adapt to any attack, no matter how unpredictable. The character’s ability to adjust on the fly felt like the perfect metaphor. That spirit is baked into our system: an execution engine that doesn’t just follow rules, but adapts, learns, and keeps moving no matter what the app throws its way.

It brought:

Adaptive power: on-the-fly code generation for actions not already built-in.
Visual intelligence: dual-image analysis to match elements contextually.
Cognitive planning: understanding workflows, not just clicks.

The system leverages a team of agents, each with a specialized role:

Mahoraga — the orchestrator
Ryomen Sukuna — the planner
Gojo — the executor

Now, the entire team is pushing this system toward launch. The end goal is clear: QA that feels effortless. A system that thinks like a tester, moves like a user, and scales with the speed of modern software.

Check out our Recent Blogs

SOCIAL

Perplexity Eyes Google Chrome With A Whopping $34.5b Bid

Last week, Perplexity made a $34.5B all-cash bid to buy Chrome. The browser isn’t on the market, but the move came with promises: Chromium would stay open, Google Search would remain the default, and $3B would be invested over the next two years. On paper, it looks like an appeal to continuity, but the real play is in the timing. With regulators already weighing whether Chrome should be spun out of Google after antitrust rulings, Perplexity’s bid inserted them into the center of that debate. Perplexity isn’t alone: OpenAI topped the headlines back in April when its executive testified it would also bid for Chrome if required to be sold, thereby framing the browser as the AI battleground’s prized asset.

Industry reaction has been clear: the deal won’t happen, but the signal matters. Browsers have become the front door to AI search, ads, and data. Whoever owns that door owns the future distribution channel. Perplexity knows this and even a symbolic bid elevates them into the same breath as Google. It reframes Chrome not just as a browser, but as strategic infrastructure for the AI era. Ultimately, this wasn’t about buying Chrome. It was about PR, antitrust theater, and reminding the market that control of the browser means control of the next phase of AI competition.

News

This Week in AI

Cohere Raises $500M and Hires Meta's Former AI Research Lead

Cohere has secured $500 million in funding at a $6.8 billion valuation, backed by investors including Radical Ventures, NVIDIA, and Salesforce Ventures. Alongside the funding, Joelle Pineau, the former VP of AI Research at Meta (FAIR), has joined as Chief AI Officer, strengthening Cohere’s leadership as it scales enterprise AI offerings.

Google Adds Gemma 3 270M: Compact, Efficient Model for Fine-tuning

Google has launched Gemma 3 270M, a highly compact (270M parameter) model optimized for energy-efficient, on-device use, including fine-tuning. It delivers strong instruction-following straight out of the box and achieves up to 10x greater battery efficiency (0.75% for 25 chats on a Pixel 9 Pro) thanks to INT4 quantization. Ideal for light tasks like classification or creative generation, it’s designed to run swiftly and affordably without sacrificing performance.

OpenAI Launches ‘ChatGPT Go’ in India at ₹399/month with UPI Support

OpenAI has introduced ChatGPT Go, its most affordable subscription tier yet, priced at ₹399 per month, exclusively for India. The plan delivers significant boosts over the free tier: 10× more message capacity, image generations, and file uploads, plus 2× longer memory. Payments can now be made via UPI, and all pricing is shown in rupees, simplifying access for Indian users.

All Newsletter Issues | Security | Contact Us | Blogs | Test Case Generator

Unsubscribe · Preferences

A QA Engine That Adapts 🥷🏼 + Perplexity’s $34.5B Chrome Gambit