Cleverbit addresses “vibe-code drift” with Nissy workflow safety
The era of simple AI-assisted coding is over. The industry is moving into the age of Agentic SDLC (Software Development Life Cycle). In this new way of working, developers are no longer just writing code; they are managing the intent behind it.
Managing Intent
The main challenge with agents is that they make autonomous decisions. They don’t just execute your instructions; they interpret them. This can lead to problems. An agent might introduce security risks, build features you didn’t ask for, or change your architecture in ways you don’t intend.
The engineering company Cleverbit Software calls this vibe code drift. It happens when the gap between what you want and what the agent actually builds gradually widens, creating “dark code” that the developer no longer fully understands, controls, or can reliably explain, despite remaining accountable for it.
Organizations are seeing the quickest gains with AI-generated code, but as soon as developers stop writing code, it has a ripple effect on the rest of the flow. “You’ll have bottlenecks, you’ll have new risks that you need to handle, and you’ll have challenges with collaboration within a team,” explained Eman Zerafa, CTO at Cleverbit. “And that means the team needs to figure out how to adopt AI across the entire SDLC responsibly. At Cleverbit, we’ve been using AI throughout the SDLC for a while now. We’ve built tools to help with requirements gathering and testing and pipelines, and we’ve created various skills along the way.”
With developers now writing much less code, the job now is to set the intent and verify the result. Developers must be clear about what they are building, how it’s tested, and which decisions they are supervising.
Best Practices
Beyond tools, success in agentic engineering requires discipline.
Among the best practices are:
Curate Context: Do not feed the agent everything. Too much information creates noise
and slows down the model. Keep your context files short and focused.
Automate Verification: Use automated tools, such as linting and static analysis, to
check the code. Make these a standard part of your pipeline. Zerafa noted that "if you
have automated verifiers, they should be separate from the code generators, or make
sure the builders have completely separate context."
Keep Records: Treat AI plans as important documents. Saving these plans helps your
team understand why certain decisions were made, even if an agent did the actual
writing.
Protect Human Attention: Use AI throughout the SDLC but always use human
judgment for high-impact decisions. Nissy helps surface these moments, so you only
spend time reviewing what really matters.
Introducing Nissy
To address this new work regimen, Cleverbit created Nissy — a workflow tool that keeps developers tethered to their AI agents, putting humans back in the loop exactly where it matters.
Nissy works in two key ways: checking your intent before code is created, and checking all the actions the agent carried out. Before a prompt reaches the agent, Nissy reviews it locally for PII and sensitive information, acting as a safety net for accidental leakage. After that, Nissy checks the intent against best practices and project specific rules, helping you understand when instructions are likely to go beyond the guardrails. Once the agent creates the code, Nissy compares it against the original intent. It catches silent errors, such as over-building a simple feature or ignoring constraints you set.
Nissy is an R&D project designed with privacy at its core: work is processed with no data retention, so proprietary code and information never leak out.
Nissy also answers the harder question: is your organization using AI correctly? It builds a measurable, evolving picture of usage over time, surfacing repeated failure modes and recommending concrete actions to improve.
“Nissy exists because we set out to measure vibe code drift by having a local tool that checks the way you’re working. That quickly evolved into an assistant that helps you stay more in control of your own code,” Ethan Zammit, an AI solutions engineer at Cleverbit, said. “So this does this by attaching to the different parts of the agentic instruction loop and it helps come up with better prompts and be more selective with what you put in.”
That, he explained, reduces the verification load of supervising the agent and it essentially closes the loop between what the code generated is and what the initial prompt was.
Watch a webinar recording here: Agentic Software Development Explained | Webinar
Moving Forward
The goal of this new engineering style isn’t to replace developers; it is to help them own the software you produce. The difference between a high-performing team and a struggling one will be the structure they build around their agents.
With tools like Nissy, users can keep the speed of AI while maintaining high standards. Developers can build faster, but with more confidence, better security, and clearer goals. The agents do the heavy lifting, but the developer remains in charge.
The post Cleverbit addresses “vibe-code drift” with Nissy workflow safety appeared first on SD Times.
Tech Developers
No comments