While some vendors are working to ensure large language models become better at reasoning, other AI vendors are making them compatible in multiple languages.

Writer is a provider of a full-stack generative AI platform for enterprises.

While the vendor provides a generative AI platform that enterprises can use to build generative AI capabilities into their workflows, it also offers a family of LLMs: Palmyra. The models support text generation and translation in numerous languages, including Spanish, French, Hindi and Russian.

"Multilingual training data and models that can be as good in dozens of other languages as they are in English is something everybody should strive for," said Writer cofounder and CEO May Habib on a recent episode of Informa TechTarget's Targeting AI Podcast. Writer also uses large volumes of synthetic data to help build legal confidence in generative AI technology, Habib said.

Writer also publishes data on how its models score for bias and toxicity.

"We really want to make sure that we are compliant with folks' ESG [equity, sustainability and governance] guardrails and guidelines," Habib said.

Writer recently raised $200 million in series C funding, bringing its valuation to $1.9 billion.

Esther Shittu is a TechTarget Editorial news writer and podcast host covering artificial intelligence software and systems.

The contact center world is a difficult place, packed with frustration and stress.

Digital communications giant Cisco sees its mission as easing that experience for human contact center workers and the customers they deal with every day.

For that undertaking, the vendor has seized on generative AI and agentic AI as the vehicles to both automate and augment the work of humans, in essence, smartening up the traditional chatbots that have long helped companies interact with their customers.

"We're to see a lot more of what I call event-based communication, proactive communication outbound that we do particularly well, powered by AI," said Jay Patel, senior vice president and general manager for customer experience at Cisco Webex, on the Targeting AI podcast from Informa TechTarget. "And then the response path to that is we think there will be AI agents involved in some of the more simple use cases.

"For example, if you haven't paid a bill, they can obviously call you in the outbound call center, but probably a better way of doing it is probably to send you a message with a link to then basically make the payment," Patel continued.

Like many other big tech vendors, Cisco deploys large language models (LLMs) from a variety of specialist vendors, including OpenAI and Microsoft. It also uses open models from independent generative AI vendor Mistral, as well as its own AI technology developed in-house or acquired by acquisition.

"Fundamentally, what we are looking at is the idea of an AI engine for each use case, and within the AI engine you would have a particular LLM," Patel said.

Among the generative AI-powered tools Cisco has assembled are Webex AI Assistant and Agent Wellness, to tend to the psyches of busy contact center human workers.

"Customers call very frustrated; they may shout at somebody. And then if you've had a difficult call, the agent wellness feature will mean that the supervisor knows that this set of agents has had a set of difficult calls," Patel said. "Maybe they're the ones who need a break now. So, there are ways of improving employee experience inside the contact center that we think we can … use AI for."

Shaun Sutner is senior news director for Informa TechTarget's information management team, driving coverage of artificial intelligence, unified communications, analytics and data management technologies. He is a veteran journalist with more than 35 years of news experience. Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems. Together, they host the Targeting AI podcast.

Box has been in the AI game for a long time.

But when generative AI mushroomed into a transformative force in the tech world, the cloud content management vendor opted to turn to specialists in the new and fast-growing technology to power the arsenal of tools in its platform.

"We've been doing AI for many years. But the really cool thing that happened … AI got to the point where the generative AI models understood content," said Ben Kus, CTO at Box, on the Targeting AI podcast from Informa TechTarget. "For us, this whole generative AI revolution has been this great gift to everybody who deals with content. It's almost like having a very dedicated, very intelligent person who stands next to you, ready to do what you want."

When generative AI exploded with OpenAI's release of ChatGPT in November 2022, Box turned to OpenAI for its first batch of generative AI tools. Box CEO Aaron Levie had known OpenAI CEO and co-founder Sam Altman for many years.

However, when a passel of other independent generative AI vendors sprang up and the tech giants started releasing their own powerful large language models (LLMs) and multimodal models, Box decided to broaden its generative AI palette.

"Azure and OpenAI are partners of ours and we think they have great models, but we are not at all dedicated to any one model," Kus said. "In fact, at Box, one of our goals is to provide you with all of the major models that you might want."

These include generative AI models from Google, IBM, Anthropic and Amazon.

One example of how Box uses an outside model is Anthropic's 3.5 Sonnet LLM, which Kus called "one of the best models out there right now."

One application is at a financial firm that deals with long bond offerings. The company needs to analyze many of these complex financial vehicles to evaluate which bonds in which it wants to invest.

"They use [the model] to extract key info. It takes the [job] of looking through these bonds. From hours or days to … hopefully, minutes," Kus said. "If the model is very good, it can give you very good answers. If it's not as smart, then it can be off a little bit. So, this particular company really wants to have the best models so they can get the best sort of use of this kind of AI."

Shaun Sutner is senior news director for Informa TechTarget's information management team, driving coverage of artificial intelligence, unified communications, analytics and data management technologies. He is a veteran journalist with more than 30 years of news experience. Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems.

This is the year of AI agents.

The last few months of 2024 brought much talk about and expectations for AI agents that can operate autonomously and semi-autonomously. Many vendors have capitalized on the enthusiasm to introduce new agentic products: Salesforce came out with Agentforce, and Microsoft introduced Copilot agents.

With 2025 here, questions about whether the momentum on agents will continue. Some see the agentic hype, and real progress, persisting this year.

Craig Le Clair, a Forrester Research analyst and author of the soon-to-be-published book Random Acts of Automation, is among those who think AI agents will continue to gain momentum in the new year.

"It's the biggest change toward AGI [artificial general intelligence] that I've seen," Le Clair said on the latest episode of Informa TechTarget's Targeting AI podcast, referring to the concept of AI that is as smart or smarter than human intelligence.

Enterprises will likely adjust the ways they use applications that use AI agents as copilots to augment humans, because many of those applications are not profitable, he said. However, AI agents will be the driving force in helping enterprises build platforms that use generative AI technology to spur business value, he said.

"When you really start to turn piles of data into conversations with people ... that's the opportunity for this," Le Clair said. "For an employee to have a conversation with standard operating procedures to get advice on what to do, or for standard operating procedures to be taken out of that PDF repository and actually put into a prompt and generate tasks that are then followed by an agent to get something done -- the potential is really there."

As with all new technology, AI agents involve a trust issue. Enterprises still do not trust the technology to be fully autonomous and perform tasks from start to finish all on its own, Le Clair said.

However, organizations can rely on AI agents to perform part of the work with the assistance of a human in the loop.

With the speed of the technology's maturation, progress toward fully autonomous agents by 2028 is likely, Le Clair predicted.

Esther Shittu is an Informa TechTarget news writer and podcast host covering artificial intelligence software and systems. Shaun Sutner is senior news director for Informa TechTarget's information management team, driving coverage of artificial intelligence, unified communications, analytics and data management technologies. Together, they host the Targeting AI podcast series.

The AI application startup, which was founded in 2016 and was valued at more than $2.1 billion in 2021, uses a reasoning engine to help employees search for information across the enterprise.

Since its inception, a key ingredient in the company's success has been AI and generative AI technology.

"We were the first company after Google to deploy BERT in production," said co-founder and president Varun Singh on the latest episode of Informa TechTarget's Targeting AI podcast.

BERT was Google's first model with bidirectional encoding that enabled computers to understand large text spans. It was pretrained, so Moveworks did not have to train it from the ground up. It also did not require a lot of data.

After using BERT to train its automation platform, Moveworks started using GPT-2 from OpenAI in 2020. This is two years before the mass popularization of the generative AI vendor's ChatGPT chatbot, mostly to generate synthetic data.

Singh added that he and his team had failed to realize right away that the model could also be used for reasoning tasks.

"It's not so much a mistake that was made or not, but it was just sort of as technology evolved, the moment a paradigm shift actually comes into full focus, you look back and you're like, 'We could have done that sooner because we had access to the models, but we didn't see how powerful they could be,'" he said.

Since the shift, Moveworks has evolved from a platform with a reasoning engine to a platform for building AI agents.

On Oct. 1, Moveworks launched Agentic Automation as part of its Creator Studio offering. The system enables developers to build AI agents.

Throughout the evolution of its business, Moveworks has differentiated itself with its use of AI technology, Singh said.

"Without AI, there's nothing Moveworks has to offer to the world," he said. "There's only value from Moveworks because of AI."

Esther Ajao is a TechTarget Editorial news writer and podcast host covering artificial intelligence software and systems. Shaun Sutner is senior news director for TechTarget Editorial's information management team, driving coverage of artificial intelligence, unified communications, analytics and data management technologies. Together, they host the Targeting AI podcast series.

When generative AI became the next big thing in tech, enterprise software giant Oracle bet heavily on a startup to provide it with foundation and large language models rather than scramble to develop its own.

That then-fledgling company was Cohere. Founded in 2019, the generative AI vendor raised $270 million in a Series C round, and its investors included Oracle, Nvidia, Salesforce Ventures, and some private equity firms. In July, Cohere raised another $500 million and reached a market valuation of $5.5 billion.

Cohere's open generative AI technology is now infused in many of Oracle's databases, a fixture among large enterprises. The tech giant has also tapped Cohere's powerful and scalable Control-R model for Oracle's popular vertical market applications, including those for finance, supply chain and human capital management.

But while Oracle has put Cohere at the center of its generative AI and agentic AI strategy, the tech giant is also working closely with Meta.

The social media colossus has gained a foothold in the enterprise AI market with its Llama family of open foundation models. Oracle is customizing Llama for its Oracle Cloud Infrastructure platform, along with Cohere's models.

"We have made a decision to really partner deeply around the foundation models," said Greg Pavlik, executive vice president, AI and data management services at Oracle Cloud Infrastructure, on the Targeting AI podcast from TechTarget Editorial.

"What we're looking for are companies that are experienced with creating high-quality generative AI models," he continued. "But more importantly … companies that are interested in enterprise and specifically business solutions."

Pavlik said Oracle values the open architecture of the models from both Cohere and Meta, which makes it easier for Oracle to customize and fine-tune them for enterprise applications.

"The advantage really of having a deep partnership is that we're able to sit down with the foundation model providers and look at the evolution of the models themselves, because they're not really static," he said. "A company will create a model and then they'll continually retrain it.

"We see our role as to come in and proxy for the enterprise user, proxy for a number of verticals," Pavlik continued. "And then try to move the state of the art in the technology base closer and closer to the kinds of patterns and the kinds of scenarios that are important for enterprise users."

Oracle also uses generative AI technology from other vendors and enables its customers to use other third-party models, he noted.

Shaun Sutner is senior news director for TechTarget Editorial's information management team, driving coverage of artificial intelligence, analytics and data management technologies. Esther Ajao is a TechTarget Editorial news writer and podcast host covering AI software and systems. Together, they host the Targeting AI podcast.

At the beginning of the wave of generative AI hype, many feared that generative models would replace the jobs of creatives like artists and photographers.

With generative AI models such as Dall-E and Midjourney seemingly creating unique works of art and images, some artists found themselves at a disadvantage. Some say the generative systems took their artwork, copied it and used it to produce their own images. In some cases, the generative systems allegedly outright stole the creative work.

Two years later, artists have to some extent been reassured by the support of stock vendors like Getty Images.

Instead of trailing behind generative AI tools such as Stable Diffusion, Getty created its own image-generating tool: Generative AI by Getty Images.

Compared with other image generators, Getty has taken great lengths to restrict its model through the data set. The stock photography company maintains what it calls a clean data set.

"A clean data set is really a training data set that a model is trained on that can lead to a commercially safe or responsible model," said Andrea Gagliano, senior director of AI and machine learning at Getty Images, on the latest episode of TechTarget Editorial's Targeting AI podcast.

Getty's clean data set does not contain brands or intellectual property products, Gagliano said. The model's data set also does not include images of well-known people or likenesses of celebrities like Taylor Swift or presidential candidates.

"We have taken the very cautious approach where our generator will not generate any known person or any celebrity," Gagliano said.

"It will not generate Donald Trump," she said, referring to the President-elect. "And it will not generate Kamala Harris," referring to the vice president and former presidential candidate.

"It has never seen a picture of Donald Trump," she continued. "The model has never seen a picture of Kamala Harris."

Gagliano added that removing this possibility also guards against those who want to misuse the technology to create deepfakes. Therefore, any generated output is labeled synthetic or AI-generated.

"We don't want any situation where we start to undermine the value of a real image," Gagliano said.

Finally, the data set that Getty uses produces images with licenses on them, ensuring that creators get compensated. Thus, a portion of every dollar made by Generative AI by Getty Images is given to the creator who contributed to the data set.

"The reason for that is the more unique imagery that we bring into the training data set, the more additive it is," Gagliano said.

Getty updated its generative AI tools Tuesday. The new capabilities include Product Placement, which lets users upload their own product images and generate backgrounds, and Reference Image, which enables users to upload sample images to guide the color and composition of the AI-generated output.

Esther Ajao is a TechTarget Editorial news writer and podcast host covering artificial intelligence software and systems. Shaun Sutner is senior news director for TechTarget Editorial's information management team, driving coverage of artificial intelligence, unified communications, analytics and data management technologies. Together, they host the Targeting AI podcast series.