For years, extending what you could do with Salesforce meant staying inside Salesforce’s own screens. If your team lived in Slack, Claude or anywhere else, getting to that CRM data meant switching tabs first. This year, Salesforce built the fix, and called it AIforce.
AIforce is a live interface layer that brings the data, workflows, business logic, permissions and governance already running inside Salesforce out to wherever people and agents work. Salesforce’s own framing is simple: people no longer have to come to Salesforce to get work done, Salesforce comes to them.
What sits underneath AIforce
AIforce isn’t a standalone product so much as a new, fourth layer across an architecture Salesforce has spent years building. The other three:
- Data 360: the unified, harmonised context. Your data, metadata and memory, giving every agent something real to reason over.
- Customer 360: the application and semantic intelligence, the business logic, processes, permissions and actions agents need to get work done across sales, service, marketing and commerce.
- Agentforce: the digital workforce itself, ready-to-deploy agents plus a builder for making your own.
AIforce sits on top of all three, making them reachable from outside Salesforce entirely.
Three places you’ll meet it
AIforce launches with three integrations, and Salesforce says more are coming:
- Claudeforce ships with “Salesforce in Claude,” a prebuilt MCP server bringing Salesforce’s data straight into Claude, with 37 prebuilt sales skills covering everything from prospecting to pipeline hygiene. It’s been piloted by Deloitte, GitLab and Legora, and is now in open beta for all customers.
- Slackforce brings four things to Slack: Slackforce Surfaces (live, filterable interfaces built from Salesforce and Slack data), Slackbot (a personal assistant that can spot a quiet account, read the support cases behind it, reassign an owner and draft a win-back email, all from Slack), Slack CRM (updating Salesforce records from a Slack prompt), and Slack Code (a shared space for a team to build with an agent together).
- Agentforce Coworker is an AI teammate inside Salesforce’s own Lightning interface. Salesforce says 100,000 users activated it within its first 35 days.
All three run on a Headless Toolkit underneath, the same MCPs, APIs and plug-ins that let developers build their own custom experiences on top of Salesforce’s architecture.
Every request still runs on your existing permissions, and Salesforce is positioning this with Zero Data Retention: your business data answers the question in front of it, and isn’t kept by the model provider afterwards. That’s a strong story to bring to a security review.
A new portfolio of job-ready agents
Alongside AIforce, Salesforce showcased seven named, job-specific agents:
- Casey: customer service across voice, SMS, WhatsApp and web chat
- Paige: IT and HR requests
- Carter: shopper agent
- Hunter: outbound sales, still in pilot, general availability due November
- Marshall: supply chain and back-office orchestration
- Piper: inbound pipeline generation
- Fin: complex customer experience workflows, powered by Fin AI’s Operator and Fin Apex models
Piper and Fin trace back to earlier acquisition, Qualified and Fin, rather than being built from scratch for this launch, so if either name looks familiar, that’s why. Salesforce’s own pitch here isn’t ‘brand new technology’, it’s ‘stop building every agent from a blank canvas’: each comes with the skills, actions and data model the job requires, connected to whatever Customer 360 data and process you already have.
Early results are strong across the board, one standout being Anthropic itself now has 79% of its support conversations resolved autonomously by Fin.
Koa: a reasoning model trained on Salesforce’s own history, not the open web
Salesforce and NVIDIA also announced Koa, Salesforce’s first CRM-specific reasoning model, built by post-training NVIDIA’s open-weight Nemotron 3 Super on a synthetic dataset modelled on nearly 27 years of Salesforce’s own CRM deployments across more than 14 industries. No customer data went into training it. On Salesforce’s own CRM benchmark, Koa already matches or beats leading model performance with three times fewer errors on real CRM tasks like updating an opportunity or routing a case.
Salesforce controls the model weights and runs Koa entirely inside its own infrastructure, so customer data never crosses out during inference. As Benioff put it, the aim was to build “a reasoning engine that understands the structure of a deal, the lifecycle of a service case, and the workflows that vary across industries”. Koa is already live inside Salesforce’s own Slack usage, and moving into pilots with 1-800 Accountant, Baxter Credit Union, Engine, Formula 1, UChicago Medicine and Xero. General availability is expected in winter 2026.
Where this leaves organisations already on Salesforce
None of this delivers value on its own. AIforce, Koa and the new agent portfolio are genuinely capable, but they sit on top of the data quality, governance and configuration decisions your organisation has already made, or hasn’t. That’s where we come in.
As a Salesforce implementation partner, our job is turning this year’s announcements into a working plan: identifying which of the seven agents fit how you operate, configuring them against your existing data model, and getting the governance right before agents are handling live customer interactions, not after. We work with organisations to assess Agentic AI readiness, deploy with best-practice configuration, and optimise post-launch.
See you at World Tour London
We can’t wait to dig into these announcements at Agentforce World Tour London, back at Excel on 15th October. If you haven’t registered yet, then grab your spot here before it’s sold out! Come speak to our team of experts about what AIforce, Koa or the new agents could mean for your own Salesforce journey, or get in touch now and we’ll start the conversation before then.
We look forward to discussing the latest innovation with you at Agentforce World Tour London in a few weeks!
