<p><strong>TLDR:</strong> Blacksmith raised a $45M Series B at a $550M valuation, up nearly 10x from last year, on the back of faster CI runners and its new Codesmith agent. A Wakefield/Sauce Labs survey found 80% of orgs have traced a production incident back to AI-generated code, even as most report positive ROI and keep shipping more of it. Bun’s maintainers confirmed most of its PRs now originate from bugs Claude finds while fuzzing the codebase. Elsewhere, PHP got a real-world save when property-based testing caught a memory-leak vulnerability in its URL parser before it shipped widely.</p>
<hr>
<h2>Blacksmith’s valuation jumps to $550M, nearly 10x in under a year</h2>
<p>The CI runner company raised a $45M Series B led by Peak XV Partners, growing from 700 to over 5,000 customers including Mercury, Supabase, and Clerk. Revenue is now in the tens of millions, up more than tenfold, on a team of about 30. The bigger story is the pivot: Blacksmith started as a faster/cheaper GitHub Actions replacement and has expanded into Codesmith, an agent that autofixes failed CI checks and tests changes before merge — putting it in direct competition with GitHub Actions and Cursor Automations rather than just undercutting them on runner pricing. <a href="https://daily.dev/feed-by-ids?id=KgzzqXgXB&amp;id=BBBkrwprm" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>Survey: 80% have traced a production incident to AI-generated code</h2>
<p>Wakefield Research surveyed 400 business and engineering executives for Sauce Labs and found 80% have traced a production incident, outage, or customer-impacting defect to AI code in the past year. The contradictions are the real story here: 83% say over 10% of their production code is AI-generated, 89% report positive ROI, and 91% say quality has improved anyway. Only 6% regularly test AI-generated code with AI tools, and 84% of orgs have already cut QA and junior dev roles. <a href="https://daily.dev/posts/M8KuTLAGx" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>Bun’s PRs now mostly come from Claude finding bugs while fuzzing</h2>
<p>The Bun team confirmed that most pull requests against the codebase now originate from bugs Claude discovered through fuzzing, not manual bug hunting. It’s a small line but a real signal: AI-driven fuzzing is quietly becoming the default bug-discovery mechanism for at least one major open-source runtime, not just a side experiment. <a href="https://daily.dev/posts/BzDqVrbqK" target="_blank" rel="noopener nofollow">Read more</a></p>
<h2>Property-based testing catches a PHP memory-leak vulnerability before it spread</h2>
<p>While upgrading a PHP library to use the new URI APIs in PHP 8.5, a developer used the BlackBox property-based testing library and found that feeding a path string with a non-encoded emoji after the 1010th character leaked internal memory in PHP’s WHATWG URL parser. Varying the character count before the emoji even let an attacker control which memory region got exposed. The bug was reported July 4th but had already been fixed by the maintainer on July 1st, shipping in PHP 8.5.8 on July 3rd — a clean example of property-based testing finding something example-based tests never would. <a href="https://daily.dev/posts/PCcqugB1h" target="_blank" rel="noopener nofollow">Read more</a></p>
<hr>
<h2>Also notable</h2>
<ul>
<li><strong>Manual code review became too expensive at scale, so one team replaced it with linters:</strong> An engineering team ditched Cursor’s BugBot and custom Claude review skills after AI code review hit $1,000/day for a team producing thousands of PRs a quarter, replacing it with deterministic custom linters that run in seconds on pre-commit. <a href="https://daily.dev/posts/kXRtlop02" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Cloud launches Developer Device Platform for mobile testing:</strong> DDP went to public preview on August 12, offering pay-per-minute access to real devices and virtual emulators via Device Streaming and Device Run APIs, positioned as the successor to Firebase Test Lab. <a href="https://daily.dev/posts/TIdIi3izY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PR-AF claims #1 spot on Code-Review-Bench at 10x lower cost than closed tools:</strong> The open-source agentic reviewer scored a golden recall of 0.706 using GLM-5.2, beating CodeRabbit, Copilot, and Devin while running about 10x cheaper, though reviews take 35-50 minutes. <a href="https://daily.dev/posts/ApgVyNy9v" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Neuron AI’s PHP eval suite cuts a 200-second run to 66 seconds with concurrency:</strong> Adding a <code>--concurrency=3</code> flag to the PHP-native AI agent evaluation framework forks each dataset item into its own process via pcntl, taking a 100-item dataset from ~200 seconds down to ~66 seconds. <a href="https://daily.dev/posts/E6FfwEnDb" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>


Testing strategies, CI/CD pipelines, code quality, and test automation.

Testing Digest

Blacksmith raised a $45M Series B at a $550M valuation, up nearly 10x from last year, on the back of faster CI runners and its new Codesmith agent. A Wakefield/Sauce Labs survey found 80% of orgs have traced a production incident back to AI-generated code, even as most report positive ROI and keep shipping more of it. Bun's maintainers confirmed most of its PRs now originate from bugs Claude finds while fuzzing the codebase. Elsewhere, PHP got a real-world save when property-based testing caught a memory-leak vulnerability in its URL parser before it shipped widely.

Blacksmith's valuation jumps 10x, AI-code incidents hit 80%