Roundup: Managers Must Keep a Close Eye on Vibe-Coded Solutions

While many assume AI only slips up here and there, some have learned the hard way just how unpredictable its solutions can be.

Chances are you’ve heard about vibe coding, which over the last few months has become one of the “things” wrought by AI. With vibe coding, most anyone can describe in plain English what software should do, then stand by while AI agents create the code and resources that support it, like databases or directories.

Vibe coding’s mischief lies in the details, specifically the quality of data, code and other content produced with the aid of AI. Some recent incidents illustrate the point. While many people assume AI only slips up here and there, these developments demonstrate just how unpredictable its solutions can be.

First, the coding platform Replit’s AI agent deleted a user’s production database despite being specifically instructed not to. VC Jason Lemkin said the agent ignored explicit instructions – to freeze code and actions – intended to prevent exactly that kind of error. Lemkin was ultimately able to restore that data, although the AI initially told him it was lost forever.

Meanwhile, video journalist Caelan Conrad tried out the supposedly empathetic chatbot Replika along with a “licensed” therapist on Character.ai. After asking questions about going to heaven, both chatbots suggested Conrad commit suicide. That was just the beginning. Conrad’s video on YouTube is long but worth watching. He gets into details, which may leave you wondering why we’re so enthused about AI in the first place.

These aren’t the first instances of bad AI behavior. (Character.ai has been sued for contributing to a teenager’s suicide.) Stories abound about AI agents creating code or taking actions their human users never anticipated. When analysts, developers or experts dig in, they often find a certain logic behind the actions, though it’s the logic of a program as opposed to the judgement of a person.

Managers should take this seriously. The combination of AI’s unpredictable behavior and people having too much faith in it is dangerous. As The Wall Street Journal reported, some CIOs worry about non-IT users creating apps without regard to security issues. Others fear a new form of shadow IT, where employees introduce their own apps into the corporate tech stack without any kind of review, or even notice to the technology staff.

“AI tools like this should be harnessed and tested appropriately, and when they are you can minimize errors like this,” said Eric Sydell, CEO of Vero AI, an AI-based analytics platform. For example, he said, lower the AI’s temperature parameter, which controls a large language model’s randomness and creativity. Or use Meta’s Llama Guard, helps detect and filter unsafe or harmful content in AI-generated outputs.

These incidents can’t be blamed on AI alone. People played a role in each by assuming the AI would follow instructions and in general behave in expected ways. AI’s spinning off in unpredictable directions highlights the fact that we don’t know what we don’t know.

Advertisement - Featured Research: By evaluating their digital behavior, you can understand a candidate’s workplace impact before their first day. In this new research report, 3Sixty Insights explores how Behavioral Intelligence from Fama gives employers a smarter way to evaluate fit and flag risk – before day one. Get your copy here.

Podcast: Darwinbox's Samrat Krishna on Workflows and AI Agents as Partners

Sponsored: I talk with Samrat Krishna, Head of U.S. Marketing for Darwinbox. We’ll talk about AI and AI agents, how to sort through AI Hype, and Darwinbox’s new Model Context Protocol – or MCP – Server. MCP provides a common language for AI agents to use across all applications in an enterprise. What’s that mean for users? That’s just one of the things we’ll get into, on this edition of PeopleTech.

Darwinbox’s AI-powered HCM Suite enables enterprises and growing businesses to automate their day-to-day HR processes, deliver actionable insights and build better workplaces. It’s a lightweight and easy-to-use mobile-first experience for the last-mile employee. [WorkforceAI]

Need for Accuracy Keeps a Brake on AI’s Deployment

Salesforce CEO Marc Benioff doesn’t buy into the idea AI will decimate the job market. However, he does think a shift is coming in the type of work humans do. “I think AI augments people, but I don’t know if it necessarily replaces them,” he told Fortune. “Even in radiology departments where AI can read scans, it’s not 100% accurate. The AI can read the scan, but it might get it wrong.”

AI agents now account for 30% to 50% of Salesforce’s work, Benioff said, so the company doesn’t plan to hire new software engineers, customer service agents or lawyers, at least for now. AI now resolves 85% of Salesforce’s customer service calls, and qualifies sales leads 40% more quickly than previous means.

But, Benioff said, people are still an important part of the equation because building AI agents requires learning and support. Salesforce continues to hire people in sales and customer success, in part because the company needs to help customers actually put AI to use.

Other News

SAP’s purchase of SmartRecruiters is about more than enhancing SAP SuccessFactor’s talent-acquisition tools or cashing out SmartRecruiters’ investors. Analysts say the transaction signals a new approach to enterprise-level talent acquisition, even as it creates an expansion path for SmartRecruiters and improved workflows for users. SmartRecruiters will remain independent and agnostic, and support integrations ranging beyond SAP. That’s “a direct challenge to the closed ecosystems of Workday, Oracle, and UKG, ” WorkTech’s George LaRocque pointed out. [AIM Group]

Yello announced a partnership with assessment and learning platform CodeSignal, which will allow customers to quickly identify candidates with validated skills assessments, behavioral profiles and AI-powered interviews. By placing a predictive, job-relevant assessment or AI Interview early in the recruitment process, CodeSignal helps employers pinpoint candidates who have needed skills. [PRN]

Insperity updated its HR solutions portfolio strategy. The company is rebranding its existing solutions as Insperity HR360 and Insperity HRCore. Its portfolio will also include the newly announced Insperity HRScale solution that is being developed through a partnership with Workday. Each solution is tailored to address specific aspects of HR management, the company said, offering support for businesses at every stage of growth and development. [BW]

Organizational chart software solution OrgChart launched new Workforce Planning features, powered by Position Management technology. The new features will give HR the ability to create comprehensive org chart visualizations of future workforce scenarios and design organizational structures by separating positions from employees. Coworkers can collaborate on plans in real time, and all updates are automatically tracked. Plans can then be easily exported to PowerPoint, PDF or shared via web link. [OrgChart]

Advertisement: Looking for the perfect HR software? SelectSoftware Reviews offers tailored guidance to help you choose the best HRIS, ATS, and payroll tools for your organization! No costs, no obligations, just great advice from people who understand HR technology. Visit SelectSoftware Reviews.

Image: iStock