038 • 16 September 2025

Contents of this Newsletter

😲

QA Needs This ASAP

⏰ : 4 min

📓

Blog Picks

⏰ : 10 min

🐈

We Hired a CHO!

⏰ : 2 min

🗞️

Weekly AI News

⏰ : 9 min

MAIN STORY

Automation Without This Is Dangerous

A recent research paper, “Why Language Models Hallucinate,” points out something important: AI doesn’t usually go wrong because it’s careless, but it goes wrong because it doesn’t actually know. When the data isn’t there, models face epistemic uncertainty. But instead of saying “I don’t know,” they’re trained to guess, because benchmarks reward confident answers over honest silence. That’s how you end up with hallucinations: plausible, polished, and wrong.

For organizations building software, this matters more than ever. If your tools are built on guesswork, trust erodes fast. To solve this, the next generation of AI systems needs something beyond better prompting—they need context. Context fills the gaps, grounds decisions, and makes outputs dependable instead of decorative. There was a time when people preferred GPT over other models largely because of the personalization feature that could store memory which was a simple but powerful use of context. Today, many see context as the next big breakthrough in AI. It marks a shift from models that just generate to systems that actually understand. And for QA, the impact is even sharper: without context, testing is just scripted guesswork. With it, accuracy, coverage, and confidence go up by an order of magnitude.

It feels fitting to be talking about this right after World Engineer’s Day. Engineers don’t just build tools; they build systems that others can trust. Context engineering may well be the foundation of that trust for the next decade.

The Challenge of Context in QA

If context is the differentiator, it’s also the hardest part to get right. Most QA automation today focuses on steps: run a script, check a result, mark pass or fail. What it misses is everything around those steps—the device state, network conditions, prior bugs, and user flows that shape whether the app actually works in the real world.

That gap is why context-based testing is both so valuable and so difficult. Capturing context means ingesting data that changes constantly, deciding what matters, and aligning it with how teams actually work. It’s not just an engineering problem; it’s a trust problem.

This is where contextual decision-making comes in. AI-driven tools can collect enormous amounts of context, but not every piece should be applied blindly. Users need the ability to approve, deny, or refine the context at every step. That balance: automation powered by context, with humans in control of how it’s used, is what makes context-aware QA trustworthy.

Building that balance is tough, but it’s also the frontier. Without it, context either overwhelms the system or gets stripped away. With it, QA becomes something new: adaptive, reliable, and aligned with how products evolve in reality.

Quash: Built on Context from Day One

At Quash, context has been our foundation from the start. Our earliest product delivered context-rich bug reports with screenshots, logs, steps, and crash data bundled together so developers didn’t have to dig for answers. That same principle has guided everything we’ve built since.

Today, context powers our entire testing flow. When users upload an app, a PRD, or Figma designs, Quash ingests that information to guide workflows end to end. Test runs don’t just check isolated scripts or steps; they account for device conditions, execution triggered by a single prompt, verifying with users whether the app matches expectations, even pulling public information to build a broader understanding of the product. Prompts only work because they’re backed by this context and testers stay relevant because their judgment is applied within a context-aware system, not lost in automation.

Scaling context is the next challenge. It means building QA tools that continuously pull from multiple sources like: product docs, designs, Mixpanel, Sentry etc while staying aligned with both development and QA teams. Done right, context-aware QA isn’t just automation; it’s a system that learns from the past, adapts to the present, and predicts future failures. The future of testing will belong to tools that don’t just execute steps, but understand the environment they’re running in. And that future is context.

Check out our Recent Blogs

TEAM

Say Hello to Our Chief Happiness Officer

At Quash, we’ve always believed that building great products takes more than just code and pipelines, it takes people (and now, one very fluffy cat).

Meet Bombay, our new Chief Happiness Officer. His role is simple: keep spirits high, paws on keyboards, and morale unshakably strong. From enforcing nap breaks to overseeing snack quality, he’s already redefining what “work-life balance” means in the office.

He might not know much about mobile testing, but Bombay has mastered the art of presence, reminding us that sometimes the best way to solve a hard problem is to step back, stretch, and maybe chase an imaginary bug across the room.

We’re excited to have him join the team. He says meow!

News

This Week in AI

OpenAI Upgrades Codex With GPT-5
OpenAI has rolled out a new version of Codex powered by GPT-5, giving its coding agent longer “thinking” cycles, ranging from seconds to seven hours depending on task complexity. The upgrade improves performance on code generation, large-scale refactoring, and reviews, with fewer incorrect comments and more actionable suggestions. Codex with GPT-5 is now available in IDEs, GitHub, and ChatGPT across paid tiers, with API access to follow.

Anthropic Gives Claude Memory and File-Editing Powers
Anthropic has upgraded Claude so it can now remember you and your team’s projects and preferences, create and edit files in Excel spreadsheets (with formulas), Word docs, PowerPoint decks, PDFs, etc. This feature is in preview for Max, Team, and Enterprise users (Pro plan coming soon). Uses include turning raw data into insights, converting PDFs to slides, automating repetitive reporting; users can download files or save them to Google Drive. There are warnings about needing internet access and risks for sensitive data.

Replit Raises $250M, Launches Agent 3
Replit has closed a $250 million funding round led by Prysm Capital with participation from a16z, Google’s AI Futures Fund, Amex Ventures, YC, Coatue, and several others, pushing its valuation to $3 billion. The company has seen explosive growth, with annual revenue soaring from around $2.8 million to $150 million in under a year as its user base surpassed 40 million. Alongside the funding, Replit introduced Agent 3, an autonomous AI agent capable of testing and fixing its own code, building workflows, and handling tasks lasting up to 200 minutes.

All Newsletter Issues | Security | Contact Us | Blogs

Automation Without This Is a Disaster Waiting to Happen 💥