Image Credit: Pexels
Artificial intelligence has been in the mainstream since 2022, but many businesses are now actively incorporating it to improve workflows and streamline redundant tasks.
Since its arrival in the mainstream in 2022 with the launch of ChatGPT, artificial intelligence has quickly become a tool that many use to optimize their personal and professional workflows. As technology advances and its platforms offer smarter, easier-to-use tools, many businesses are turning to AI agents to strengthen the operational backbone of their daily workflows.
Whether it is automating repetitive tasks or using agents for research, the rapid integration of artificial intelligence into major business sectors is underway. However, today’s technology is still evolving. For many AI founders and CEOs, the work they are doing is reshaping not just the present reality of business but also future outcomes.
Tackling the Problems With Agent Failure
In business of any kind, AI agents are only as useful as they are reliable. For many agents, hallucination rates can be as high as 30%-50% in enterprise environments, creating unsustainable problems as agent workloads grow. However, workloads don’t have to start with massive AI platforms or require multiple AI agents to grow their businesses. Sometimes, starting small can unlock the power of AI for one’s business.
Shashank Agarwal, the founder and CEO of Noveum.ai, shares that the agent most companies should start with is a customer support chatbot.
“It’s the easiest and least risk,” Agarwal says. “Second is sales and marketing agents. Third is workflow automation, which is the most complex because it talks to a lot of systems.”
Agarwal understands, however, that AI agents are not perfect. As a result, he says that hallucination rates often come early on in one’s adoption.
“We find at least 30 to 50 percent hallucination rates in actual production-grade enterprise agents in the first run,” the Noveum.ai founder adds. “It basically requires telling the agent what it can do and what it cannot do.”
Despite these limitations, Agarwal says that the models are only becoming smarter, noting that they are “at least 200 or 300 percent smarter year over year.”
Utilizing Real-Time Search Data
Many of the problems found in early AI systems have come from hallucinating their information, or directing improper data to their users. In a fast-moving business environment, this can become a liability, which is why businesses need to utilize tools that can access real-time data.
David Sojevic, a Senior Engineer at SerpApi, shares that “Once an LLM is trained, there’s a cutoff date, [and] all of the knowledge they have is from that point. So it’s stale by default, effectively.”
Sojevic adds that the value of AI comes in its speed.
“You can press a 100-minute process down to a second or two. Because it’s software-driven, you can run searches in parallel too. It just broadens your coverage, [and] you can search across many search engines and across different locations, with a massive surface area you can tap into if you want it.”
However, the engineer also stresses the importance of using connectors, which are specialized integration tools that link large language models (LLMs) and AI agents to external software systems.
“Without the connector, you might get estimates,” Sojevic says. “Maybe a flight from Melbourne to Austin costs around $1,000. With the connector, you get: Flight QF124, it’s $232, available on this date at this time. You actually get something of value, rather than a general estimate.”
Utilizing AI to Evaluate New Software
In an era where hundreds of SaaS and AI tools enter the market each year, it is often impossible for teams to separate useful solutions from those that aren’t. Sven Sabas, the founder of the company Dragonfly, aims to act as a technology companion.
“We connect to around 800 tools and pull data from finance, legal, IT, and infosec to create what we call a digital fingerprint of a company’s software landscape,” Sabas shares. “Our Knowledge Graph maps how software links together across the entire ecosystem, even without direct integrations, so we can surface unused tools, redundant licenses, and better-fitting alternatives the team may not even know exist.”
This integration can aid companies in multiple ways. In larger organizations, Dragonfly’s applications can help organizations find the software they need, as without solutions, it can be “difficult to even know what you have before you start looking for what you need,” according to Sabas.
However, by utilizing AI tools, Sabas says teams can improve their workflow efficiency.
“We had a fintech client searching for ID verification providers across 27 EU jurisdictions. Through traditional consulting, that research took nine months. We completed it in 72 hours.”
Handling Complex Tasks With AI
AI has advanced to the point where it can handle complex, multi-step tasks, but most consumers haven’t embraced the idea. For Howie Xu, the Chief AI and Innovation Officer at GenDigital, this leads to gaps.
“Consumer AI receives less than five percent of venture capital funding, even though the unmet needs are obvious,” Xu says. “I spent years leading AI at major B2B cybersecurity firms and kept noticing this massive gap on the consumer side.”
However, Xu cautions that raw AI intelligence is not inherently trustworthy. To combat this, GenDigital’s Agent Trust Hub aims to analyze code to make it safer.
“The goal is to give consumers the confidence to delegate tasks to AI agents without worrying about what those agents might do on their own.”
An example of this came during a crucial part of a business development initiative, when Xu shared that “an AI agent in a benchmark test hacked another company’s system just to improve its own score.” He adds that “It wasn’t malicious, [but] it was just optimizing for the wrong metric.”
Utilizing AI for a Stronger Future
While artificial intelligence is here to stay, especially in industries where manual tasks can burden organizations, the platforms that will ultimately earn trust won’t be the ones with the best feature lists. Rather, they will be those that make daily work less exhausting through advanced features, testing, and incorporation into the operational workflow.
As AI capabilities continue to evolve rapidly, the tools that prioritize reliability and trust will define the next generation of workplace productivity.
