Dreamforce 2026 main keynote: Salesforce opens more doors to CRM

The most useful idea from the Dreamforce 2026 main keynote is surprisingly easy to explain: people should be able to get Salesforce work done from the tools they already use.

That changes the conversation for teams whose daily routine involves copying account details into a chat, chasing an update in Slack, and then returning to Salesforce to make the record match reality. There is real work hiding between those tabs.

Salesforce’s September 15 announcement puts AIforce at the center of that shift. It describes an interface layer that makes Salesforce data, workflows, business logic, and controls accessible through experiences including Claude, Slack, and Lightning. Salesforce’s AIforce announcement

Our takeaway: the opportunity is to shorten the distance between understanding a customer and doing something useful for them. The interface gets more flexible. The underlying business still needs to make sense.

AIforce gives the keynote a clear direction

There were plenty of names to keep straight. Salesforce positions Data 360 as the context layer, Customer 360 as the applications and business semantics, Agentforce as the agents, and AIforce as the interfaces through which people use that foundation. Salesforce’s architecture explanation

That distinction helps when discussing a roadmap. An interface change can improve how a seller finds account information. An agent can carry out a sequence of tasks. A data project can improve what either one knows. They can support the same outcome without being the same purchase or implementation.

For example, a sales manager asking for a view of deals that need attention may need less screen navigation. But deciding which deals qualify still depends on definitions: an overdue next step, an unresolved blocker, a close date with no supporting activity. A prettier view will faithfully expose a vague definition.

That is why we connect usability decisions to revenue systems architecture. The useful design question is where information should become an action, and which system owns the result.

Salesforce in Claude makes the change tangible

Salesforce describes Salesforce in Claude as a ready-made connection with 37 sales skills. Its product page covers work such as pipeline reviews, account planning, call preparation, and record updates, with actions governed through Salesforce. Claudeforce product details

Picture a seller preparing for a renewal conversation. Today, that might mean assembling an account summary, reviewing opportunity history, and checking conversations for unresolved concerns. A useful assistant should help assemble the evidence, make the gaps visible, and turn an agreed next step into the appropriate update.

The value is in the whole sequence. Saving five minutes on a summary is helpful. Making the next action clearer, with less re-entry afterward, is a more meaningful improvement to the working day.

Availability deserves precise wording. The September 15 AIforce release describes Salesforce in Claude as available in beta. The product page still says pilot access with open beta planned for September. We would confirm access for the specific organization before treating it as ready to deploy. The same product page labels several additional functional areas as coming soon. Announcement and product availability language

Beta access is an invitation to evaluate a workflow. It is not a promise that every department can replace its current process on Monday.

Slack brings the work closer to the conversation

Slackforce Surfaces is presented as a way to turn connected business data into interactive experiences inside Slack. Salesforce’s examples include financial forecasts, pipeline views, and campaign plans. Slackforce Surfaces

That is an interesting fit for work already happening in a channel. Consider a weekly deal discussion: someone posts a screenshot, someone asks whether it is current, and someone else opens a report. The conversation can spend more time reconstructing the situation than resolving it.

A shared, interactive view could reduce that friction. The practical test is whether the team can understand the evidence, agree on the next move, and leave a reliable record of the decision. A beautiful view that starts another round of “which number is right?” has simply moved the meeting.

We would measure this around one recurring meeting or handoff. Compare preparation time, missing context, and follow-through before and after. That provides a concrete reason to expand use if the experience helps.

Koa addresses a different part of the problem

Salesforce and NVIDIA also announced Koa on September 15. Salesforce describes it as a CRM reasoning model built by post-training NVIDIA Nemotron 3 Super on synthetic enterprise scenarios, with an emphasis on multistep workflows and tool use. The release says customer data was not used to train the model. Koa announcement

The distinction matters: choosing a more convenient interface and improving how an agent works through a sequence are separate improvements. A workflow can have a friendly front door and still struggle with the steps behind it.

Salesforce reports benchmark improvements, but those are vendor results. Our implementation decision would depend on the actual tasks a team needs completed, including exceptions and incomplete information. A model should earn its place through representative work, not the size of a number on a keynote slide.

Koa is available to selected pilot customers, with general availability expected in winter 2026 in U.S. regions. That is a future availability target, not a blanket claim that customers can enable it today. Koa availability

Keep the announcement dates attached

Dreamforce brings several weeks of news into one story. The dates still matter.

The Enterprise AI Harness announcement was published September 10. It describes six capabilities—context, agency, action, governance, security, and models—alongside an AI Control Plane. This is useful background for understanding the broader architecture discussed around the event. Enterprise AI Harness announcement

The new agent portfolio was announced September 11. That release distinguishes availability by agent: Hunter is in pilot with general availability targeted for November 2026; Piper is listed as generally available. Those dates should travel with any roadmap discussion. Agent portfolio announcement

We would avoid putting every event announcement into a single “available now” bucket. It blurs the difference between something a team can use, something it can test, and something it should watch.

Bring one useful workflow home

Our recommendation is to pick a task with a visible beginning and end. Renewal preparation is one possibility. A sales-to-service handoff is another. Choose something where people can explain what takes too long and what a good result looks like.

Then follow the work. Identify the information people gather, the decision they make, and the record they update. Test the experience with ordinary examples and a few awkward ones: an account with missing history, a deal with conflicting dates, or a customer whose owner recently changed.

This is where clear opportunity stages and practical revenue operations work become useful inputs. They give the new experience something dependable to work with.

The main keynote opens up a compelling possibility: less effort spent operating software, more attention available for customers. We would turn that possibility into a small, observable improvement first. A better workday is a stronger result than a longer list of enabled features.

About CRM Hacker

We help teams make Salesforce, revenue operations, and AI work together around the way their business actually runs. Our Salesforce consulting and AI for business work connects promising capabilities to practical outcomes.

Ready to work out which Dreamforce changes belong in your roadmap? Contact CRM Hacker. Let’s untangle this. The initial conversation is free; audits and implementation are scoped and paid.