Agents in the enterprise: Salesforce and DeepL see productivity move the needle as use cases grow for AI – News

If 2025 was the year of experimentation and pilots with agentic artificial intelligence, then 2026 is shaping up to be the year when agents find a permanent home in enterprise operations.

Use cases are scaling up rapidly, based on data delivered during briefings at the AI & Big Data Expo in London this week by companies such as Salesforce Inc. and DeepL SE. This momentum is being fueled by accelerated improvements in the operation of agents themselves, encompassing key metrics such as intelligence and observability. Productivity is proving to be a major attraction as enterprises deploy agents into infrastructure.

“To get buy-in we need to really look… at what the things are that move the needle,” said Paul O’Sullivan, senior vice president of solution engineering at Salesforce, during a conference panel session on Wednesday. “It’s really easy to link AI to a productivity gain. I think we’ve got a once-in-a-lifetime opportunity to reimagine how we run our businesses.”

Gaining an enterprise foothold

Attendees who flew through London’s Heathrow Airport to attend this week’s AI conference had an opportunity to experience how agents are beginning to transform the travel experience. The world’s fourth busiest airport uses a Salesforce-designed agent called “Hallie” to field visa questions and fill passengers in on wait times at security.

“They’ve shifted 70% of customer engagement into agents,” O’Sullivan told the gathering. “Heathrow can now offer a 24/7 service.”

Executives from the AI solutions provider DeepL provided further evidence that agents are gaining an enterprise foothold. Scott Ivell, vice president of product marketing for DeepL, told News that the German-based company has 2,000 customers globally deploying AI agents. They are being used for a wide range of purposes, encompassing report analysis, creating sales target lists and checking legal documents for key contract terms.

“We believe this is the way to free humans from repetitive work and also raise productivity,” said Ivell. “We want to bring forward real, usable enterprise use cases.”

Contextual understanding for agents

The expansion of enterprise use cases is being driven by a shorter ramp for implementation and strong interest in the productivity gains that AI agents can bring.

Executives from DeepL and Salesforce expressed a consensus opinion that agents can be implemented within a few days or a few weeks depending on the tasks involved and the availability of data. It is not a heavy infrastructure lift. Yet, like adding a new employee, it can take a little while for them to become a full contributor.

“It’s easy to get an agent up and running,” according to O’Sullivan. “How am I going to get this agent to be really effective? What I’m seeing is it takes a couple of weeks. Your agent is an extension of your workforce.”

There is another important factor driving enterprise adoption of AI agents. In a surprisingly short period of time, agents have upped their game. The more that agents can understand context, the better they will become as productive contributors in the workforce.

As agents capture inference data, this gets written to Salesforce Data 360, formerly the Data Cloud. Salesforce’s acquisition of Informatica last year has allowed it to combine both companies’ metadata with Data 360’s real-time context to replace agentic guessing with reasoning.

“We are starting to bring contextual intelligence to the agent,” said O’Sullivan, when asked by News if agents were becoming smarter. “What we are seeing is the agents intuitively now are drawing conclusions much quicker.”

Moving beyond the LLM

One key theme that emerged from dialogue at the Big Data & AI gathering revolved around enterprise adoption of large language models or LLMs.

Many enterprises have already embraced generative AI through LLMs and their capabilities to fill knowledge gaps. Yet LLM services also come with high operational costs, data privacy risks when using publicly available platforms, and the inherent complexity involved in fine-tuning large models to meet business requirements.

Agents are gaining an enterprise foothold Scott Ivell, vice president of product marketing for DeepL, said at the conference.

There is a perception in the enterprise world that, though LLMs are good enough to satisfy demand in the consumer arena, they may not always be the best option for enterprises. “Good enough is not good enough,” Ivell noted.

Salesforce has approached this issue by seeking greater control over how agentic AI operates in enterprise environments. AI needs to be connected to accurate data, business logic and governance to generate trusted, predictable outcomes that the LLMs’ raw intelligence can provide.

Salesforce has addressed this through Agent Script, a JavaScript Object Notation-based language that allows enterprises to define complex, deterministic workflows. This gives agent builders precise control through integration of deterministic rules with agentic reasoning.

“The LLM alone is not enough,” O’Sullivan explained. “AI has to be connected to a system.”

Focus on observability

Another key topic of discussion at the conference was observability. Once AI is connected into a trustworthy structure, enterprises need to monitor and fully understand the actions agents are taking. There is too much at stake for enterprises that manage trillions of dollars in transactions, such as the credit card powerhouse Visa Inc., which has committed to generative and agentic AI.

“If not tracked it should not be executed,” Konstantina Kapetanidi, vice president of global data solutions for Visa Europe, said during a presentation Thursday. “We are not operating anymore in black boxes. We’re here to give answers and to explain.”

In January, Salesforce released enhancements for MuleSoft Agent Fabric to provide a single control plane for AI agents, tools and metadata. Agentforce Observability also provides an ability for users to drill deep into every agent interaction. “Observability now is in the context of what agents and people you have running your business,” O’Sullivan said in an interview.

Companies such as Salesforce, DeepL and Visa are seeing firsthand how rapidly agentic AI is changing the enterprise landscape. The field is evolving quickly and the focus is now on how to deploy AI agents with trust and confidence that a business will measurably improve as result. There is also the likelihood at some point in the future that agents will become just digital furniture, an accepted part of infrastructure that no one really notices.

“We’re seeing traction in the main levers driving your business,” Ivell said. “Agents will come to the fore. Then they will disappear in the background.”

Photos: Mark Albertson/ News