AI might be gunning for your job are increasing.

Faced with pressure from stakeholders and competitors, many companies have pushed forward with slashing positions that they believe the tech is capable of handling. Last week, telecommunications firm Altice cut 1,000 employees, or 16% of its workforce, with AI creating redundancies among some jobs as the company cut costs.

Altice’s decision is far from singular. Major firms like Klarna, UPS, Duolingo and Cisco have used AI to replace workers. Though the tech comes with its fair share of risks, companies are leveraging it to reduce labor costs, increase productivity and garner returns on their investments in the tech.

Some may be feeling the squeeze of stakeholder scrutiny and competitive pressure, said Lisa Highfield, principal director of research and advisory services at HR research firm McLean & Co.

“When we see headlines such as 100 jobs laid off because of AI capabilities, competitors are bound to pay attention to that, and they don’t want to fall behind that curve,” Highfield said.

Questions remain, however, about whether AI is ready to become a member of the team rather than simply a tool. Highfield says it depends on the job and its specific tasks:

• Jobs involving “repetitive, low-complexity knowledge tasks,” such as answering frequently asked questions, summarizing, entering data and scheduling, will likely be the first to go. Assistant and clerk positions, for example, may have a “high exposure rate” to AI.
• Additionally, anything with “clear rules and delineated answers,” such as writing templates or coding, is also ripe for automation.
• Many positions may have bits and pieces of their jobs automated, but “not necessarily the entire job holistically,” said Highfield. “Job security around that is really dependent on how much of your tasks are correlated to AI.”

Enterprises seeking automation need to look within their specific industries and organizations to figure out where it’s a good fit, said Highfield. “We really need to understand how AI applies to our industry and the goals and tasks within an organization.”

Plus, many jobs themselves are in a period of change in the wake of AI. In the same way that lamplighters were replaced with electricians and mathematicians no longer need to do long division by hand, new opportunities will be revealed as others fade. “It’s not just cutting jobs, it’s what jobs are we creating,” said Highfield.

However, enterprises also need to understand the risks associated with making the leap into AI, said Highfield. Along with potential brand and reputation damage, enterprises open themselves up to regulatory scrutiny and declining workforce morale. The tech itself still has issues, too, including a propensity for data security slip-ups, hallucination and bias, especially as tasks become more complex.

Many companies may be better off starting with upskilling and augmentation of jobs, rather than replacing them with AI entirely, she said. “There’s a shift towards providing capabilities within the organization for our workers so that they can work with the technology.”

Wrangling AI agents can feel like herding cats.

While developers are keen to test a variety of platforms for building AI agents, doing so can become unwieldy for IT teams, leading to agent sprawl. The push to adopt agents has prompted some enterprises to jump in hastily, and amassing disparate agents can lead to a bevy of issues, said Amr Awadallah, CEO of Vectara.

“Because each one of these agents is being built in a different way, there is no single pane of glass, from a management perspective,” said Awadallah.

So how does agent sprawl happen in the first place? When developers test and build agents for multiple use cases on a number of different platforms – such as using OpenAI, Google, AWS or open-source options – these agents are then only controllable by the developers who built them:

• If one of those developers leaves the enterprise, “It’s hard to have developers inherit and maintain each other’s code,” Awadallah said.
• Moreover, it creates a situation in which IT teams are unable to govern these agents, he said, as they’re built on different platforms without one central system for observability. If something goes wrong, he said, a web of sprawled agents makes the problem that much harder to diagnose and fix.
• That lack of governance can create problems with agent accuracy, security and performance, he noted, as well as rack up costs.

Preventing this sprawl relies on deploying a uniform standard from the jump, said Awadallah. Picking one standard platform to build and deploy agents allows developers the freedom of creativity while giving IT teams the ability to manage them, he said. Additionally, to avoid being too locked in to one platform if it jacks up prices, figure out how much migration to another will cost if necessary, he added.

“If you’re building AI agents en masse … then spend the time and the exercise to come up with a common standard that all of your developers are building with, versus letting the developers go pick whatever they want to pick,” he said. “Otherwise, you’ll end up in a mess a year from now.”

If you’re already experiencing agent sprawl, now is the time to get a handle on it, Awadallah said. Pick a standard platform for building agents going forward, and start to untangle yourself from the “legacy” ones that were built on others, he said.

“It only gets worse over time,” he said. “Try to enforce the standard right away, so you don’t have to pay the cost of it later on.”

There are times when the cloud isn’t within reach.

Oracle might be trying to provide a stopgap: The company is seeking to patent a system for “cloud edge device virtualization” that allows cloud services to operate closer to the source of the data, rather than data needing to be transferred back to a centralized cloud platform.

“Current centralized cloud computing environments may not meet time-sensitivity requirements when streaming data due to the inherent latency of their wide-area network connections,” Oracle said in the filing. “Remotely generated data may need to be processed more quickly.”

Upon user request, Oracle’s tech creates a “virtual edge device” emulating a physical edge device that may be not easily accessible. That virtual device is used to run workloads on the edge, providing cloud-like functionality without needing to be near a data center or server farm.

This tech is particularly useful in situations involving large amounts of edge-computing devices, devices in remote areas with limited internet access, and applications where latency needs to be minimal.

Oracle’s tech highlights a growing need: As Internet-of-Things devices continue to be deployed and bolstered with AI, data processing and movement back and forth from centralized cloud environments can quickly gum up the works. Systems like Oracle’s could reduce friction in edge computing environments, allowing anything from smart home assistants to factory robots to leverage AI seamlessly.

Plus, while Oracle trails cloud hyperscalers like Amazon, Google and Microsoft, the company’s cloud offerings are gaining ground in the wake of AI demand: In its most recent earnings report, the company’s cloud earnings for the quarter totalled $11.7 billion, up 14% year-over-year.
According to CRN, the company holds 3% of total market share in the cloud services industry. But as AI continues to quickly take root in the physical world, tech like this could set it apart.