AI has already become a powerful tool for the lab. Proper use and integration of AI tools can significantly speed up lab decision-making, improve workflows, and ease complex data analysis. However, blindly jumping into new AI projects without having the proper procedures, processes, and governance will create additional challenges for the lab. To learn more about how lab managers can make good decisions about implementing AI projects in the lab, we talked with Redwerk CEO Konstantin Klyagin.
Q: Many companies rushed to adopt AI, expecting immediate productivity gains, yet a large share are seeing little to no return. What is breaking between the promise of automation and the reality inside day-to-day operations?
A: The mistake is thinking AI can solve structural problems inside the business. Most companies are trying to automate processes that were never clearly defined in the first place. There’s no shared understanding of what “good output” looks like, no clear ownership, and no consistent workflow. AI steps into that environment and produces something, but the team doesn’t trust it, doesn’t know how to use it, and ends up second-guessing everything.
From what I’ve seen, AI only works well when the basics are already in place. You need structure, clear expectations, and discipline in how work is done. Otherwise, you don’t get productivity, you just get more output that needs to be managed.
Q: You argue that AI often amplifies existing inefficiencies rather than fixing them. What does that look like in practice when automation is layered onto unclear or fragmented workflows?
A: It usually looks like acceleration without direction. If your workflow is unclear, AI will happily generate more work inside that same confusion. People spend more time aligning, reviewing, and correcting than moving forward. AI didn’t solve the problem. It just made the gaps more visible.
Q: There’s a growing narrative that AI is saving time, yet employees report spending hours each week correcting its output. Where are organizations underestimating the true cost of automation?
A: They’re underestimating the cost of verification. AI can generate something in seconds, but that doesn’t mean it’s correct, complete, or usable. Someone still needs to review it, test it, and make sure it works in a real scenario. That effort doesn’t disappear; it just shifts.
Various industry polls back up our own internal observations. Sonar’s 2026 State of Code: Developer Survey reveals that a massive 96 percent of engineers refuse to blindly trust AI-authored code without human review. This skepticism has birthed a novel kind of engineering grind: teams now dedicate practically a fourth of their week (24 percent) to reviewing, correcting, and verifying AI suggestions.
There’s also a hidden cost in coordination. When AI produces output, who owns it? Who is responsible if it’s wrong? Who decides whether it’s ready to ship? If those questions aren’t answered up front, teams lose time just figuring out what to do with the output. AI reduces effort in parts of the process, but it doesn’t remove accountability.
Q: When an organization decides to automate a process, what are the early warning signs that the workflow itself is not ready, even if the technology is?
A: You can usually tell very quickly. If different people describe the same process in different ways, that’s a problem. If no one can clearly explain what success looks like, that’s another one. And if decisions are constantly being revisited or escalated, it means ownership is unclear. Another strong signal is when teams rely on “we’ll figure it out as we go.” That might work in early exploration, but it doesn’t work with automation. AI needs structure. If the process is still evolving, automation just locks in confusion.
Q: From your experience stepping into projects post-implementation, what tends to fail first when AI is introduced too early into a system?
A: Trust fails first. Teams stop trusting the output, so they start double-checking everything. That creates friction, slows things down, and defeats the original purpose of automation.
Advanced Lab Management Certificate
The Advanced Lab Management certificate is more than training—it’s a professional advantage.
Gain critical skills and IACET-approved CEUs that make a measurable difference.
After that, quality starts to slip. You get inconsistencies, edge cases are missed, and small errors accumulate. Over time, that leads to rework, which is always more expensive than doing things right from the start. The focus shifts from “how fast can we build” to “how do we fix what we’ve built.”
Q: As AI systems evolve from tools into autonomous agents that interact with each other, how does that change the way organizations should think about control, accountability, and risk?
A: It forces companies to become much more explicit about responsibility. When you have multiple agents interacting, it becomes very easy to lose track of who did what and why something happened. That’s dangerous, especially in systems that affect real users or business outcomes. You need clear boundaries. Each system should have a defined role, and there should always be a point where a human can step in and understand what’s going on. Otherwise, you end up with decisions being made in a black box.
From a risk perspective, this is less about technology and more about governance. If you don’t define accountability early, you won’t be able to fix problems later.
Interested in lab design?
Q: Traditional testing focuses on outputs, but agent-based systems behave over time and across interactions. What does “good testing” look like in this new environment?
A: Testing becomes much more about behavior than single results. You’re no longer checking if one output is correct. You’re observing how the system behaves over time, across different inputs, and in edge cases that are hard to predict.
There's also a mindset shift in how we approach testing AI-powered applications. In standard programming, tests assume deterministic behavior—the same input is expected to always produce the same output. But AI models don't work that way. They generate output based on probabilities, so a query may produce strong guidance one day, and outdated or incomplete advice the next. This can happen because of changes in the underlying model, updates to retrieval components, or gradual data drift.
AI helps us generate more scenarios and explore edge cases faster, but the real value comes from designing the right tests and understanding how the system should behave under pressure. Good testing in this environment means continuous validation. You don’t test once and move on. You monitor, you adapt, and you treat the system as something that evolves, not something that stays fixed.
Q: Looking ahead, what separates organizations that will turn AI into a real operational advantage from those that will keep investing without seeing meaningful results?
A: The difference is operational discipline. The companies that succeed treat AI as part of a system, not as a standalone solution. They invest in clear processes, strong engineering practices, and accountability before they scale automation.
The others focus on tools first. They adopt new technologies quickly, but without fixing the underlying structure. That leads to constant experimentation without real outcomes.
In our experience, the companies that get value from AI are the ones that were already good at execution. AI just makes them faster. The rest are still trying to figure out how to work, and AI doesn’t solve that for them.
