Unpacking Salesforce's AgentForce announcement

I have to say, Salesforce's AgentForce announcement impressed me. However, tempering my enthusiasm, I must also acknowledge Salesforce's history of announcements that take time to materialize. That said, this launch feels different, and I want to dig into what I’ve learned and why it's significant.

Taking the offense - This announcement marks a return to Salesforce's roots, setting a bold vision and not being shy about expressing contrarian views. The company has set an ambitious goal to empower one billion agents by the end of 2025. In doing so, Salesforce has positioned its solution in stark contrast to what it now deems outdated copilots and chatbots.

Massive undertaking - Salesforce mobilized 4,000 developers in just the last couple of months to bring this vision to fruition. This rapid redeployment of resources underscores the company's commitment to this new direction and its ability to pivot quickly.

Setting realistic expectations - In an era where many tout autonomous agents as capable of taking over broad 'Jobs-to-be-done' (JTBD), Salesforce has set reasonable expectations on the types of tasks that can be accomplished. Such pragmatism is a smart move at a time when many businesses are struggling to turn AI pilots into production and scale them. Already, agents can deliver some quick wins.

The promise of low hallucination seems fair. Salesforce is anchoring its agents in its data and metadata, with possible actions well-bounded due to their transactional nature. Focusing the first wave of its agents within its 'walled garden' is prudent. However, most companies have data and systems outside of Salesforce. Furthermore, nearly all companies I know of struggle with CRM data quality issues.

A stronger platform story. Not long ago, Salesforce was a collection of clouds connected internally with Customer 360 and externally using MuleSoft. This announcement reinforces Data Cloud as the backbone. In recent quarterly earnings calls, Salesforce has shared impressive adoption numbers, with paid Data Cloud customers growing 130% year over year. The company predicts it will become its fastest product to reach $1B in revenue.

Models are taking a back seat. In just a few quarters, the focus of AI has shifted from being a model issue to a system-wide challenge. Agentforce is described as a three-legged system that connects to various models, selects the most appropriate one for the task at hand, integrates with multiple datasets, and interfaces with diverse systems of action and engagement.

What to expect next?

I’d like to lift the hood on the Atlas reasoning engine, though details have been kept close to the chest. From what I’ve gathered, Agentforce leverages technology from Airkit.ai, which it recently acquired. Additionally, it utilizes two newly announced AI models developed by Salesforce AI Research: xGen-Sales, a proprietary model designed to automate sales tasks, and xLAM, a new family of LAMs trained using APIGen, a proprietary pipeline for generating synthetic data.

I’d also like to see more customer stories that demonstrate how businesses are moving into production and overcoming operational challenges, particularly around data management. Data debt — especially poor CRM data quality — often goes underestimated, yet it can have significant downstream effects.

Finally, I’m curious about pricing, especially as businesses struggle to prove the ROI of their AI projects. Salesforce has introduced usage-based pricing, starting at $2 per conversation. For a customer service or sales rep handling 20 conversations a day, that translates to a hefty $9,000 per year before discounts.

In the upcoming grand battle for AI domination, Salesforce has scored a significant win, resetting market perception after the more tempered success of Einstein. The company has moved aggressively, not just against peer enterprise software providers, but also against hyperscalers, industry and vertical software vendors, and pure-play startups. Interestingly, Marc Benioff identified the "Do-It-Yourself" approach as Salesforce's key nemesis.

The AI agent market is red hot. In just the sales space, four significant funding events have occurred in the past few weeks: 11x.ai (AI SDR, BDR, and soon CSM) raised $50M; Bounti.ai (autonomous AI teammate) secured $16M; Landbase (AI-powered action model) raised $12.5M; and Artisan (AI BDR) raised $11.5M.

Agentforce's introduction marks an important milestone, setting the company apart with its bold yet pragmatic approach and platform-driven strategy.