Backing · July 8, 2026 · Adilet

Backing Hackroda: the bottleneck moved to verification

AI made writing code cheap. It did not make trusting code cheap. That gap is the company.

Hackroda is one of our focused product bets, and the reasoning fits in three steps.

First, the volume step. AI now writes 20 to 30 percent of code at Microsoft, per Satya Nadella in April 2025. Sundar Pichai put Google’s share of AI-generated new code above 25 percent back in late 2024. GitHub Copilot crossed twenty million users. Whatever you think of the quality, the quantity is not in dispute: more code is entering the world per engineer than ever before.

Second, the quality step. Google’s DORA research tied the rise in AI adoption to a measurable rise in delivery instability. Vendor studies point the same direction; one analysis of hundreds of pull requests found AI-generated code carrying meaningfully more issues per PR than human-written code. The precise multiplier matters less than the direction. As one industry line put it: the bottleneck is not coding anymore, it is verification.

Third, the position step. Our services arm already does QA work for clients. That means Hackroda does not start from a whiteboard. It starts from inside paid engagements where the verification problem is concrete, recurring, and billed for. We see it from both sides: as builders shipping AI-assisted work, and as the people clients hold accountable when that work has to hold up.

What Hackroda is: QA software for the AI-built era. Repeatable tests, review loops, and evaluation setups for code, AI workflows, and business-critical systems, with SDK and API connections into the tools teams already use. The direction we describe internally is the system of record for quality.

Where it actually stands, stated plainly: the first version was built and tested with buyers, and the second version is in build for them. No claim of scaled usage yet. Analysts size the software testing market in the tens of billions of dollars with double-digit growth in the AI-automation segment; we cite the range rather than one number because the estimates disagree, and the spread itself is the honest takeaway.

We fund what we understand, and we understand this one from inside the work.