Salesforce Unveils Long-Horizon AI Agents for Businesses

Salesforce is taking its enterprise AI strategy a step further with a new generation of long-horizon AI agents designed to handle complex business tasks over extended periods rather than simply responding to one prompt at a time.

The company has announced a family of new “job-ready agents” for Agentforce, targeting areas such as sales, customer service, commerce, IT and HR. The goal is to make AI agents easier for businesses to deploy while giving them the ability to work toward larger objectives with less human intervention.

The move comes as companies continue experimenting with AI agents but face significant challenges turning pilots into reliable, production-ready systems.

Salesforce believes that purpose-built agents, combined with the right data, tools and controls, could help close that gap.

Salesforce Introduces Job-Ready AI Agents

The latest Agentforce expansion includes several specialized agents designed around specific enterprise workflows.

Among them is Piper, an inbound pipeline-generation agent designed to work across websites and inboxes. Its job is to identify, qualify and help convert potential leads.

Hunter, meanwhile, focuses on outbound sales. The agent works alongside sales representatives and can support the pipeline process from researching prospects through outreach.

For customer support, Salesforce is introducing Casey, a help agent designed to resolve customer issues across multiple communication channels, including voice, SMS, WhatsApp and web chat.

The company is also developing agents for other business functions, including IT and HR requests, as well as product comparison tasks.

Rather than asking companies to build every AI workflow from scratch, Salesforce is positioning these specialized agents as ready-made solutions that can be deployed for particular jobs.

Why Long-Horizon AI Agents Matter

Traditional AI assistants are generally designed around individual interactions. A user asks a question, the AI responds and the conversation moves on.

Long-horizon agents work differently.

Instead of completing a single action, these systems are designed to pursue a broader objective through multiple steps. They can maintain context, use different tools, respond to changing information and continue working until a task reaches its intended outcome.

That capability could be particularly important in enterprise environments.

For example, a sales task may require researching a prospect, identifying relevant information, preparing an outreach message, following up and updating the company’s records. Customer service may similarly require several interactions before an issue is completely resolved.

Salesforce’s strategy is to make AI agents capable of handling these longer workflows instead of limiting them to one-off responses.

Enterprise AI Adoption Still Faces Major Obstacles

Despite the growing interest in agentic AI, companies have struggled to move these systems from experimentation into everyday business operations.

Data cited in the announcement highlights the problem. Deloitte has reported that only a relatively small share of organizations have agents running in production, while many more remain in the pilot stage.

Gartner has also warned that a substantial portion of agentic AI projects could ultimately be cancelled because of rising costs, unclear business value or insufficient risk controls.

These challenges suggest that simply giving businesses access to powerful large language models isn’t enough.

Organizations also need reliable infrastructure, appropriate permissions, business context, tools, monitoring and mechanisms for handling failures.

That is where Salesforce believes its Agentforce platform can provide an advantage.

Are Large Language Models Enough?

According to Jayesh Govindarajan, Salesforce’s EVP of AI and head of engineering, one of the central lessons from enterprise AI adoption is that large language models alone are not sufficient.

An AI model may be capable of reasoning and generating responses, but an enterprise agent needs more than those abilities.

It needs access to the correct information, the ability to use business tools and controls that determine what actions it can take.

Salesforce therefore views the agent as a combination of the underlying model and the surrounding infrastructure.

This “harness” can provide context, access to tools and controls that allow an AI system to operate within a business environment.

The approach is particularly relevant for long-running tasks, where an agent may need to make decisions repeatedly over an extended workflow.

Early Results From Salesforce’s AI Agents

Salesforce says its existing agents are already producing measurable results.

According to company-provided figures, its Piper pipeline agent has surfaced $82.5 million in leads for Salesforce since launching in April.

The company also says its website has handled more than 1 million agent conversations.

Salesforce has highlighted several customer examples as evidence of how long-horizon agents can perform in real business environments.

Travel platform Engine reportedly had 50% of chat inquiries fully resolved by its help agent, Eva.

At travel and spend-management company Perk, Salesforce says the outbound sales agent Hunter was responsible for building 60% of the sales pipeline.

Autism Queensland, meanwhile, reportedly had 70% of its administrative requests resolved by an employee service agent.

These examples illustrate the type of workflow Salesforce wants Agentforce to automate: repetitive, multi-step processes where employees currently have to spend time following up, collecting information and moving tasks through different stages.

AI Agents Are Becoming Better at Longer Tasks

Salesforce’s announcement comes as research suggests AI systems are becoming increasingly capable of completing longer tasks autonomously.

A study from METR found that the amount of time required for tasks AI agents can complete with a 50% reliability rate has been increasing rapidly.

This trend is important because the usefulness of an AI agent isn’t necessarily determined by how well it answers a single question.

The greater opportunity comes when an agent can take responsibility for a larger objective.

Instead of asking an employee to repeatedly prompt an AI system, a long-horizon agent could potentially continue working through a process while maintaining its state and responding to problems along the way.

That could significantly increase the potential productivity gains from enterprise AI.

Salesforce Wants Businesses to Start Small

Interestingly, Salesforce’s strategy isn’t based entirely on giving organizations completely autonomous AI systems.

The company is taking a narrow, task-specific approach with its new job-ready agents.

This could make adoption easier because businesses don’t need to design every workflow themselves.

Prebuilt agents can be configured around specific jobs, allowing organizations to start with relatively well-defined processes before expanding their use of AI.

Technology analyst Rebecca Wettemann argues that this approach could become increasingly important as businesses move beyond the initial excitement surrounding AI and become more cautious about the risks associated with deploying autonomous systems.

Ready-made agents can potentially reduce the expertise, cost and time required to build an agent from scratch.

Agentforce Also Gets an Agent Optimizer

Salesforce isn’t stopping at prebuilt agents.

The company has also announced Agent Optimizer, which is designed to work alongside customer-built agents and help improve their performance.

The concept is essentially an AI system helping organizations improve another AI system.

Agent Optimizer can identify where an agent fails and help determine what needs to be changed.

Salesforce describes this as moving toward a form of self-improvement for AI agents.

If successful, this could become an important part of managing large numbers of enterprise agents.

Rather than requiring developers to manually examine every failure and update every workflow, an optimization agent could potentially identify recurring problems and recommend or implement improvements.

Long-Horizon Agents Could Change Enterprise Automation

The broader significance of Salesforce’s announcement is the industry’s shift from conversational AI toward autonomous task completion.

Businesses don’t simply need AI that can answer questions. They increasingly want systems capable of completing work.

That distinction is crucial.

An employee asking an AI assistant for a sales prospect summary is useful. An AI agent that researches the prospect, prepares outreach, follows up, updates the CRM and continues working until the objective is completed could provide a much larger productivity benefit.

The challenge is making that autonomy reliable.

Long-horizon agents need to remember what they have already done, maintain state, recover from errors and understand when human intervention is necessary.

For highly regulated businesses, they also need strong controls around data access and decision-making.

Salesforce Faces Growing Competition

Salesforce’s move also places it in direct competition with other enterprise technology companies developing AI agents.

Platforms such as ServiceNow and other business software providers are similarly working to make agentic AI easier for organizations to deploy.

The competition is likely to focus not only on which company has the most powerful AI model, but also on which platform can provide the best combination of data, integrations, security, controls and ease of deployment.

In that environment, prebuilt agents could become an important differentiator.

Businesses may prefer an AI platform that allows them to deploy useful agents quickly rather than spending months designing complex systems internally.

The Future of Agentforce

Salesforce’s latest Agentforce announcement reflects a broader change in how businesses are thinking about generative AI.

The first phase of enterprise AI largely focused on chatbots, copilots and productivity assistants.

The next phase is increasingly about AI agents that can take responsibility for complete workflows.

Salesforce’s long-horizon approach attempts to address that opportunity by combining task-specific agents with the infrastructure needed to operate them at enterprise scale.

The company’s new agents for sales, customer service, IT, HR and commerce are designed to provide businesses with a starting point, while Agent Optimizer adds a mechanism for continuously improving agent performance.

Final Thoughts

Salesforce is betting that the future of enterprise AI will involve agents that do more than answer questions.

With its new family of job-ready Agentforce agents, the company wants businesses to delegate longer, multi-step workflows to AI while reducing the complexity of building and managing those systems.

The biggest challenge will be proving that these agents can deliver consistent results while remaining secure, controllable and economically worthwhile.

If Salesforce can make that balance work, long-horizon agents could become a significant part of enterprise software — turning AI from a tool employees consult into a system that can actually take ownership of defined business tasks.