Next-Gen Automation: The Outlook for Agents

Agents are everywhere. At least, that’s what advocates of AI would have you believe. It seems to be the only thing that big technology wants to talk about. Yet, agents are only a secondary concern for marketers and sales reps struggling to meet their targets in an unpredictable economy. Plenty of businesses are experimenting with AI, and some of those experiments have even reached production. However, even the most forward thinking managers are starting to question the wisdom of deploying Agents at the scale pitched by Marc Benioff and Sam Altman. A bit more openness about the current state of the technology would better prepare businesses for when agentic AI is finally production-ready.

Slow Progress

AI technology definitely has a bright future, but there are still many challenges to overcome before swarms of robots can automate the day-to-day operations of modern business. The capabilities of the technology are rapidly evolving, and accuracy is slowly improving. However, the technology still fails at many simple tasks. AI models only make sense if they can be trained easily and work effectively. They don’t need to be 100% accurate at everything. Humans never are. However, they do need to be good enough and cheap enough. At the moment, the technology is neither. It’s still too expensive, and the user experience simply isn’t good enough.

In the meantime, there are plenty of use cases for glorified chatbots. There’s a reason why search engines have become the latest technology to be disrupted. Generative AI has made natural language search a real alternative. Recent studies show widespread adoption of AI search engines for educational queries among Millennials and Gen Z. Sure, hallucinations are still a problem and the technology isn’t good enough for academic research. It works fine for everyday business use though, and Google is still available as a fallback. Not that Google is particularly reliable these days either. We’ve just become used to its quirks.

The trouble for investors is that Microsoft and Salesforce have bigger ambitions for Generative AI. They’re looking to automate the majority of business workflows, and then scoop up the cost savings through higher subscription and usage fees. That won’t work. Regardless of the reliability of technology, cost is still a significant barrier to AI deployments. Currently, the technology doesn’t unlock major new productivity gains, and it’s not cost effective enough to replace a human workforce. The success of Deepseek has shown another path is possible. A multitude of cheap models running locally is the way forward, with each model trained for a specific use case. However, the technology doesn’t allow for that at the moment.

Sales Agents

We’re already seeing that dynamic play out within the realm of sales. Last week saw two more glimpses into the transition that is to come. Both Microsoft and Salesforce announced major new AI capabilities intended to automate routine sales tasks using agents. Microsoft revealed a sales agent capable of carrying out the role of a typical SDR. That’s still a reactive capability, requiring human prompts before it can do anything. Agentforce has had a similar capability since its initial launch last year.

The next big breakthrough needs to be autonomous execution, where an agent can be triggered by a machine learning model. Salesforce finally announced that capability at their recent TDX event. The idea is that agents can be automatically triggered by data changes or by a traditional workflow. That allows agents to run autonomously when certain pre-defined conditions are met, such as when an account reaches a certain score threshold or when a lead hasn’t been followed up recently. The agent then uses its training and knowledge sources to choose the right action for each lead, based on the list of actions made available to it.

Autonomous Execution

Silicon Valley is starting to pivot towards that outcome across the board. Reports are beginning to emerge of potential new product announcements from OpenAI, including a set of role-specific models intended to have capabilities similar to those of different levels of knowledge workers. To begin with, pricing will be a major barrier to adoption. The rumours also indicate these models will be priced well above the expected salary for a human employee with an equivalent skill set. Although, as we’ve seen with deep research those costs will drop over time.

It will be a long time before AI is trusted to autonomously close enterprise opportunities. However, there is a real demand for automated buying further down the value chain. B2B buyers are used to online purchases now. Analyst research indicates that decision makers no longer want to deal with human sales reps unless absolutely necessary. The intelligent use of agents makes that possible, while preserving the human element for deals where it is really needed. We’ve already seen the same dynamic in customer service, where generative AI has elevated the quality of your typical chatbot. More progress is needed before the technology can see similar adoption across other areas of the business.