Drinking from the 2024 Salesforce Dreamforce (Agentforce) AI firehose.

As with most things in tech, AI announcements dominated Salesforce’s 2024 Dreamforce event in San Francisco, spanning new services (Agentforce), extensions into industry clouds, acquisitions (Tenyx, Own Company), data governance and compliance, and enhancements for Tableau and Mulesoft. While there were a sizeable number of announcements across almost all of Salesforce’s solutions, the primary focus was Agentforce.

Agentforce is an AI-powered platform, designed to deploy autonomous and assistive agents across various business functions, such as sales, service, marketing, finance, etc. Built on the Salesforce platform and integrated with the broader Salesforce Customer 360 ecosystem, Agentforce leverages artificial intelligence (AI), automation, and real-time data to help businesses streamline tasks, support routine activity, provide data and insights into decision making and improve customer experience.

‘Agents’ are the key to Agentforce, capable of performing a wide range of tasks, such as interacting with customers, handling routine service requests, automating workflows, and personalisation of customer interactions. These agents can work seamlessly within Salesforce apps like Slack, Tableau, and Marketing Cloud, as well as with third-party applications (through API integrations).

Salesforce states that Agentforce can integrate real-time data from Salesforce Data Cloud and other data lakes, providing more context aware solutions. There are 5 attributes that govern each agent:

1. Role. Thejob the agent performs.
2. Data. The dataand information the agent can access.
3. Actions.What capabilities the agent has.
4. Guardrails.What the agent is not permitted to do.
5. Channel. Themedium through which the agent communicates and works.

Additionally, Salesforce introduced a suite of data governance tools designed to help companies comply with new AI regulations and safeguard sensitive data, emphasizing its commitment to data security and management.

Image source: Salesforce

Agentforce’s presence permeated across all key Dreamforce announcements, and it shows the evolution of Salesforce’s investment in AI (which started back in 2014) from generative AI solutions to today’s (what Salesforce terms) ‘agents’, including:

• Tableau Einstein. Tableau Einstein integrates Agentforce for Tableau and Pulse, enabling real-time data analysis and insights, and leveraging AI to enhance data discovery and business context. By using secure connections to Data Cloud, Tableau Einstein ensures real-time, secure data integration across departments.
• Salesforce’s MuleSoft integration further allows businesses to connect CRM and external systems, enhancing AI agents' capacity to automate tasks and personalize customer experiences on a larger scale.
• Slack has also incorporated Agentforce, enabling teams to interact with data and perform tasks directly within Slack while collaborating with AI agents from partners like Adobe and Anthropic. In addition, Agentforce enhances Salesforce’s Marketing Cloud by automating campaign creation and personalization.
• Data Cloud introduces innovations like processing unstructured data,expanded connectivity with platforms like Stripe and Twilio, and strengthened data security with AI tagging and secure sharing options.
• Lastly, Salesforce's partnerships with Google, IBM, and NVIDIA further elevate AI agent innovations, improving real-time data workflows and potentially enhancing performance across Salesforce's ecosystem.
  

There’s a lot to process, however initial thoughts suggest there’s quite a bit to like and a few things to ponder.

Like:
Our TRA Asia Pacific Heatmap for AI investment in Asia Pacific and Japan shows 2 of the top 3 priorities for companies are sweet spots for the Agentforce services - #1 Sales & Marketing and #3 Customer Operations.

• With companies concerned about some of the unknown knowns of AI (licensing, costs, integration, security, GRC, etc), the positioning of Agentforce addresses most of these head on with pitches around business cases, security, GRC, integration and education. (We’ll come back to costs in a minute). Our own AI research amongst Asia Pacific organisations highlights these same concerns, especially around cost, security/GRC, the rate of change in LLM development and general AI solution complexity.
• The acquisition of Own Company, a data protection/management company, should absolutely strengthen security and governance capabilities and complement other aspects of Salesforce’s data protection and ethics/auditability of AI solutions. Tucked away amongst a plethora of announcements, this one is a clear standout.
• The 5 agent attributes are the precursors forspinning up an agent, contrasting with many of the generative AI tools being ‘on by default’, adding to concerns and complexity. Organisations have control and ability to move at their own pace, rather than reacting to ‘on by default’.
• There’s a pretty solid set of templatised use cases (~100) already available, addressing the typical ‘how and where do we start with AI?’ issue many organisations are confronted with as they start planning their AI strategy.
• There is a small but capable partner ecosystem available to organisations. It includes 4 content creation partners (eg. Box, Docusign, etc ), 5 sales/customer insights partners (Moody’s, Korn Ferry, etc) and 5 industry automation partners (Copada, OpenText, etc). For implementation, as of 18^thSeptember 2024, Salesforce states there are 44 SI partners available, mostly GSIs.
  Lo-code/No-code approach. Targeted squarely at automating front office agents and entire workflows, Agentforce is architected to work across any channels using Saleforce’s Datacloud as the core and will work with other data lakes as well...admittedly you might need some pretty solid partner support for that.
  Agent to human (A2H) hand-off is standard and Agentforce is capable of handling both structured and unstructured data.
  

Things to ponder

• No Data Cloud, no Agentforce. Unless all your data sits within Data Cloud, you’ll need to ingest from other data lakes. This, and Data Cloud’s pretty stringent data requirements, means careful, detailed planning will be necessary, and most probably partner support too.
  
• Pricing. We get it, it’s THE major Salesforce customer event of the year and the focus is on the ‘new’. Deep dives into pricing tend to take the gloss off launches however given the reliance on Data Cloud (where pricing can be complicated) and concerns about AI costs, licensing and usage, a little bit of clarity for Agentforce pricing would be helpful.
  
• Partner agents, that is, agents working with 3^rd parties in the ecosystem need to be built from scratch. If you don’t mind partnering with the global SIs, you should be fine. Otherwise, you may need to wait while local, specialised partners become available.
  
• As organisations become more familiar with Agentforce and deploy larger number of agents, multi-agent orchestration will become increasingly important. Right now, this is not yet supported within the Salesforce platform.
  
• Salesforce has long been a leader in responsible AI usage, ethics and related areas, and given the deep reach of Agentforce into all Salesforce solutions, greater clarity on the Agentforce ‘brain’, the Agentforce Reasoning Engine, that drives output is needed.