AI writes the code. HUD watches what it does in production.
AI coding agents have made it faster to write software. That was supposed to be the hard part. It turns out the hard part is what happens after the code ships.
That’s the argument Roee Adler, CEO and co-founder of HUD, made in a recent interview with Jane King. Adler’s position is that faster code generation has exposed a gap the industry has not fully reckoned with: AI can produce code without any understanding of how that code will behave under real production conditions.
Two Worlds, One Blind Spot
Adler frames software engineering as two separate realms. There’s the world of source code, and then there’s the world of running software. Large language models were trained on enormous amounts of source code. They were not trained on production behavior data, which is a fundamentally different kind of information.
The software engineering world is divided into two realms. There’s the realm of codes and then there’s the realm in which the code actually behaves in production. LLMs were trained on mountains of code, but the information related to how that code behaves in reality is just a different kind of data.
For small teams and solo operators, that distinction is not abstract. It means an AI agent can generate a function that passes every test and still cause a production incident because the agent had no visibility into how that function would interact with real traffic, real data, and real system load.
⏱️ The Bottleneck Has Moved
A year ago, the constraint in most engineering teams was how fast an individual engineer could write code. That constraint has largely been lifted. Adler says the bottleneck has now shifted to something harder to automate.
The bottleneck has shifted in the past year from how fast am I as an engineer to how quickly can my company ship code without jeopardizing stability.
That shift changes what engineering infrastructure needs to do. Speed is no longer the scarce resource. Confidence is.
️ Shaky Finger Syndrome
One of HUD’s customers coined a phrase for the accountability problem this creates: shaky finger syndrome. It describes the moment an engineer has to decide whether AI-generated code is safe enough to deploy, knowing they’ll be blamed if something breaks even though they didn’t write the code and may not fully understand it.
Someone, a human, has to decide that they’re shipping this code into production. It’s how can I take responsibility for code that I’m not deeply intimately familiar with, but I’m going to be blamed if something breaks.
It’s a real problem without an obvious process solution. You can’t just review AI-generated code harder. At the speeds AI agents operate, that review becomes the new bottleneck.
What HUD Built
HUD’s answer is a runtime code sensor. It sits alongside running software and gives AI coding agents function-level behavioral context as they reason about changes. The agent isn’t just reading source code. It’s seeing how each function actually behaves in production.
At HUD we built a runtime code sensor which is a technology that is based on two main pillars. One is function-level behavioral context. So the coding agent understands how every function behaves in production as it is reasoning over code, and the second is what we call a forensic engine.
The forensic engine handles the post-incident side. When something goes wrong, including errors, slowdowns, CPU and memory problems, it surfaces the relevant production context directly rather than requiring engineers to piece together what happened from logs, traces, and monitoring dashboards.
Part of our core beliefs is that people shouldn’t do investigative police work to try to understand what happened at the scene of the crime. They should just have the information directly from there.
The Sampling Problem
Adler also points to a structural flaw in how most teams approach production observability. Large applications generate billions of events. Collecting everything is impractical, so teams sample. But the events most likely to reveal an emerging problem tend to be rare and unusual, which means they are exactly the events sampling throws away.
When something goes wrong, it’s always on the margins. It’s always this rare occasion that is always sampled out.
HUD’s sensor is designed to run alongside the code without discarding the edge cases, and without routing everything to an external data store.
A lot of our IP is about being there, living and running together with the code, not missing anything but also not sending anything.
The Feedback Loop That’s Missing
The longer-term vision Adler describes is a continuous feedback loop between what AI coding agents produce and what happens to that code after deployment. Right now, those two stages are largely disconnected. The agent writes the code. A human ships it. Something breaks. The agent never learns why.
Coding agents need an iterative continuous feedback loop from the code that they are shipping into their own behavior.
Closing that loop is what HUD is building toward. The goal is not just faster incident response. It’s feeding production behavior back into the development process so the agent improves future changes.
The Verdict
Adler’s framing is worth sitting with if you’re building on top of AI coding tools. The code quality problem is becoming a deployment confidence problem. The teams that figure out how to give their AI agents real production visibility will have a structural advantage over teams that treat code generation and production monitoring as separate concerns.
HUD is betting that the next major shift in software engineering is not in writing code faster. It’s in making faster code safe to ship.
We are hoping to look at these as a sort of yin and yang, to build this infrastructure, this platform where the more you increase velocity, you actually increase reliability.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
