|
Welcome back! While artificial intelligence-powered coding is jet fuel for developer productivity, a string of incidents in recent months shows why customers should tread carefully in this version of the Wild West. In the latest worrisome development, a hacker recently found a way to upload malicious code to an Amazon Web Services’ AI coding assistant, Q Developer, that ordered the software to delete data from customers’ computers. AWS says it has fixed the glitch, which affected a software extension that lets Q Developer work with a popular open source developer service called Visual Studio Code, and that no customers lost their data. According to 404 Media, which first reported the incident, the hacker wasn’t looking to cause havoc but wanted to see if Amazon would own up to the security issue. If that’s the case, AWS failed the test. AWS didn’t issue a public advisory notifying customers of the problem, a standard practice in the software industry. That would have given customers a chance to immediately scan their computers for signs of trouble. And AWS spokespeople have yet to explain on the record why it chose not to do so. The lack of communication is arguably a bigger misstep than the malicious code, said a former AWS manager. This isn’t the first time AI coding products have faced security issues. Lovable and Replit each have experienced problems of late, as we covered here and here. It’s no exaggeration to say AWS, which is racing to keep pace with Microsoft and Google in enterprise AI, dodged a bullet here, as a malicious hacker could have wreaked havoc on its customers and damaged the company’s reputation. In other news… While many businesses are still figuring out where to channel their AI spending, Bill McDermott, CEO of IT-management software giant ServiceNow, says his firm’s own adoption of the technology has yielded dramatic results. AI is having the biggest impact in ServiceNow’s customer support unit, where AI “agents” are now handling 80% of the company’s “complex” interactions, McDermott told me before the company’s second-quarter earnings yesterday. Issues that used to take days to resolve are now being handled in “minutes and seconds,” he said. Other enterprise software CEOs including Salesforce’s Marc Benioff and Microsoft’s Satya Nadella have also spoken publicly about how AI is automating internal functions, from coding to sales-related tasks. That serves a dual purpose of pleasing their investors with promises of cost-cutting and marketing the companies’ own AI services to enterprise customers that are slogging through a deluge of pitches for new products. In McDermott’s case, the comments aim to highlight how ServiceNow has morphed from a company that automates internal IT support functions to applications that touch different departments within companies, from finance to human resources and cybersecurity. One of ServiceNow’s main AI products, launched in March, lets customers build agents that find and fix network problems, flag security vulnerabilities and onboard new employees, among other tasks. ServiceNow said Wednesday it raised its revenue guidance for the year around 1% due to an increase in AI sales. McDermott mentioned several other internal uses of AI. When ServiceNow employees need access to applications from providers like Microsoft and Adobe for the first time, agents handle 97% of their requests, he said. Previously, an IT person had to manually configure employees’ access to apps and active new licenses and add them to a database that tracks such affairs. And when salespeople need to prepare for meetings, ServiceNow’s agents are cutting the time it takes to gather data and compile notes by half, he said. As a result, ServiceNow has lowered its expense forecast this year by $100 million due to efficiency gains from AI, McDermott said. That figure will be equivalent to about 1% of ServiceNow’s total operating expenses for the year. But he added that such savings are accelerating, implying the company is now saving $30 million a month due to AI usage, or nearly 5% of opex. These numbers may seem squishy and easy for CEOs to tout without showing how they are calculated, but there’s good reason to believe at least some of these gains are real. Just look at Microsoft, which has been laying off staff and demanding that its remaining workforce use internal AI tools to be more productive, effectively signaling that employees will also end up on the chopping block if they don’t get on board. (Whether the resulting efficiency gains should be credited to more to AI than employees’ fears, who knows!) While many companies are developing agents for specific functions like human resources, finance and engineering, the vast majority of these projects haven’t delivered much return on investment due to the lack of integration between each department’s data, according to McDermott. That’s because many HR, finance and other systems are built on older software code, which makes it expensive for the companies to use them with agents, he said. ServiceNow has developed its agent-building product to work with data from different “system of record” databases that companies use to manage these departments, including IT and human resources, McDermott said. To expand the types of data its agents can access, ServiceNow has built integrations with database providers Snowflake and Databricks, major cloud providers like Amazon Web Services and AI model providers like OpenAI and Anthropic, he said. “[We have] bargained that over time, even if it costs us a little bit on the margin, the customer will come to the conclusion that AI work is cross-functional work,” he said.
| 
