Salesforce likes to take a bunch of related technologies and brand them under an aegis term. Currently Salesforce’s Ai offerings under the aegis of “Agentforce” are capable, but no more so than other agent frameworks. IMHO.
Under that aegis is their copilot like code writing helper for their proprietary language Apex, and their proprietary Js frontend technology Lightning web components. Its performance is ok, so long as you don’t expect it to reuse any of your existing code. Its performance can write tests, but don’t expect tests with mock data, for instance.
They also have “agent” like chat interfaces for customer service to end use chats. Again, performance depends on who implements it, but while functional, it’s not better than other tools.
What Salesforce is doing with ai, however is interesting. They’re implementing client facing Ai tools with a “trust layer” that ensures Salesforce’s customer’s data - including data on the customers customers is protected from leaking into a training g pool, and on the return side hedging against hallucinations. There’s more of course - using your own models, a whole host of prompt engineering and then some nice to haves like Ai calculated data fields. (Prompt response stored on records, recalculated when prompt inputs are changed…)
Their big bet, is twofold. 1: that the key differentiation in any agent system is not just the prompts and agent workflows, but the data available to ground the results. (Duh). They’re further betting that their enterprise customers that have terabytes of data on spending habits of b2b and increasingly b2c customers, will yield results that move the “deal” meter. 2: by giving away Ai certifications to anyone with a pulse in their Ai tools, their hoping the groundswell of “certified” people will give cXo’s an inflated sense of Salesforce’s market and mindshare leadership.
But Ai is only part of the picture at Salesforce. You should also be paying attention to their “DataCloud”. Data cloud is all the tech and tooling to enable a customer to become their own data broker. Take that ability to de-anonymize and coalesce disparate data streams into identities - and then feed that into your “Ai” agents and you might really have something. (Albeit probably a creepy something).
This might, for instance give you the ability to note that “bob” made capital expenditures in the fall as cmo of foo Inc, and that according to your linked in scraper he’s now at “bar Inc” so maybe you should reassign your sales exec from foo to bar, before q4…
You know, if you’ve thought to develop that agent.
I now work in the public sector, so I don’t know anyone developing “no seat in the rain” agents, but … I can envision how to do it with … well any of the agent frameworks I’ve used. The only difference is whether or not the customer is willing to pay Salesforce for easy access to their Salesforce data. Course, they got robust APIs so, pulling that data could just be step 1a in your agent workflow…
Happy to answer any questions