Every few years Salesforce changes how people think about the product, and this year it happened at Dreamforce. The headline was AIforce, a new layer that lets people and AI agents use Salesforce without opening the Salesforce screen at all.
If your first reaction was "wait, what does that mean for my org," you are not alone. The announcement came with a lot of names. Claudeforce, Slackforce, Agentforce Coworker, Headless Toolkit, Data 360. It is easy to lose track of which one does what.
This article sorts it out. We will look at what AIforce is, how it connects to the parts of Salesforce you already know, what you can try right now, and what you should fix in your org before you switch anything on. At Concretio we spend our days inside Salesforce orgs, so the focus here is on practical steps rather than hype.
What AIforce actually is
Put simply, AIforce is an interface layer. Salesforce describes it as a live interface that brings the platform to wherever people and agents already work. The data, workflows, business logic, permissions and governance stay inside Salesforce. What changes is where you can reach them from.
For years the deal was simple. You wanted something from Salesforce, so you logged in, found the right app, clicked into a record, and read what was there. Even a well built Lightning page asks you to know where to look.
AIforce flips that. Instead of you going to the records, an agent reasons across the data and conversations in Salesforce and hands you the insight. Ask which deals are stuck, and you get an answer with context, not a list view you have to sort yourself. Ask it to update a record or start a workflow, and it can do that too, from Claude, Slack or another surface.
A useful way to picture it is a window. The house is the same house. AIforce just adds windows in more rooms.
A quick example of how it feels
Picture a sales manager on a Monday morning. Before AIforce, she opens Salesforce, filters opportunities, exports a list, and pings three reps on Slack asking for updates.
With Claudeforce, she asks Claude which deals have gone quiet in the last month and gets an answer built from live Salesforce data, with amounts and stages attached. She then asks it to update the next steps on those records. The reps see the same context in a Slack channel through Slackforce. No exports, no chasing.
That is the promise. Whether it works for you depends on how accurate those records are today.
Where AIforce sits in the Salesforce stack
AIforce does not replace anything Salesforce announced before. Salesforce calls the whole thing its Agentic Enterprise architecture, and AIforce is the newest floor on top. Here is how the pieces stack up:
- Data 360 brings together unified and federated data, along with metadata and memory, so an agent understands the customer and the business.
- Customer 360 carries the application and semantic intelligence, meaning the business logic, processes, permissions and actions agents need across sales, service, marketing and commerce.
- Agentforce is the digital workforce, with ready to deploy agents and a builder for your own.
- AIforce brings all of that foundation to any AI interface.
Notice the order. The bottom layers are about trust and context. The top layer is about reach. If the bottom is messy, the top will only spread the mess faster. Keep that in mind, because we come back to it in the preparation section.
There is a bigger story behind the launch too. Salesforce is positioning itself as a platform that goes beyond CRM into many enterprise workflows, and the idea that you might never log in to Salesforce directly to use its data and logic is a big part of that. It also leans on its community for proof of scale, pointing to 12 million apps built and 9.6 billion API calls every day. Whether you love that direction or worry about it, it is where things are heading.
The launches you can look at today
AIforce arrived with a few packaged experiences, so you do not have to build everything from scratch.
Claudeforce
Claudeforce puts Salesforce inside Claude. It combines a prebuilt Salesforce MCP server with 37 ready to use sales skills, covering work from prospecting to pipeline hygiene. Coverage of the launch says Claude Enterprise users can install an add on called Salesforce in Claude from AppExchange, and that it was in open beta at launch. Typical jobs include preparing for a meeting, checking where an opportunity stands, and updating records.
If you have never heard of MCP, it stands for Model Context Protocol. Think of it as a standard plug that lets an AI tool talk to another system. Salesforce ships the plug already wired, so you skip the custom integration work.
Slackforce
Slackforce turns Slack into what Salesforce calls a live, multiplayer interface for Salesforce. The word multiplayer matters. Sales work is rarely solo. A deal review in a Slack channel can now pull in live Salesforce context while the whole team is looking at it, instead of someone pasting screenshots of a dashboard.
Agentforce Coworker
Agentforce Coworker packages the same idea as an agent experience. Like the other two launches, it brings along the skills, dashboards and triggered workflows people already use, so the interface is new but the work is familiar.
Headless Toolkit
This one is for builders. Headless Toolkit gives a common way to make Salesforce available in new AI platforms as they appear. It uses Salesforce MCP, APIs and AI, so you can bring data, context, logic, workflows, identity and permissions into your own AI tool, app or interface. If you build products on top of Salesforce, this is the piece to read first.
Security and trust
Every serious buyer asks the same thing first. Where does my data go?
Salesforce answers in two parts. First, every request runs on your existing permissions and business rules, so an agent only sees what the person talking to it is allowed to see. If a sales rep cannot open a finance record today, the agent working for that rep should not be able to either. Second, Salesforce states a zero data retention approach, which means business data is used to answer the question but the model provider does not keep it.
Both points are good news, but they are promises about the platform. They are not a replacement for checking your own setup. Permissions only protect you if they are correct, and in many orgs they are not. Profiles that grew over ten years, permission sets nobody remembers creating, and sharing rules with exceptions on top of exceptions all become more visible when an agent starts answering questions across every object at once.
So treat AIforce as a reason to finally audit access. It is a lot cheaper to do that now than after an agent surprises someone with a record they did not know they could reach.
What it means for admins and developers
For admins
Your job shifts a little. Less time on page layouts, more time on data quality, permissions, and the descriptions that tell an agent what a field means. Field labels and help text suddenly matter because the agent reads them too. A field called Status__c with no description is a guess for a human and a bigger guess for an agent.
For developers
The interesting work moves toward actions and integrations. Agents need clean, well named actions to call and clear data to read. A plain example is the kind of query that sits behind a "which deals are going stale" question:
SELECT Id, Name, StageName, LastActivityDate, Amount
FROM Opportunity
WHERE IsClosed = false
AND LastActivityDate < LAST_N_DAYS:30
ORDER BY Amount DESC
LIMIT 20
If LastActivityDate is empty on half your records because activity capture was never set up, the answer will be wrong no matter how smart the model is. The query is simple. The data underneath it decides the result.
For marketing and service teams
The same logic applies. Salesforce says the platform intelligence covers sales, service, marketing and commerce, so pilots will not stay in sales for long. Start where your data is cleanest, not where the loudest team sits.
How to get ready
You do not need to rebuild anything. You do need a plan. Here is the order we would follow.
- Audit permissions first. List your profiles and permission sets, remove what nobody uses, and test with a real low access user. Since agents work within existing permissions, this is your main safety net.
- Clean the fields agents will read. Fix duplicates, fill the fields that matter for your pilot, and add descriptions to custom fields. Short, plain descriptions are enough.
- Pick one narrow use case. Meeting prep, pipeline hygiene or record updates are good starts. One team, one workflow, one clear measure of success.
- Set a baseline before you begin. Note how long the task takes today. Without that number you cannot say later whether AIforce helped.
- Run a small pilot. Give it to a handful of people who will complain honestly, and collect what breaks.
- Review and expand. Look at the results, fix the data problems you found, then add the next team.
A note on step three. The temptation is to try everything at once because the launch list is long. Resist it. Pilots that try to cover ten workflows usually prove nothing about any of them.
Mistakes we would avoid
- Switching on before cleaning up. Agents make bad data visible fast, and not in a comfortable way.
- Treating it as an IT project only. The people who do the work should help decide what the agent does.
- Skipping the licensing check. Confirm current packaging and costs with Salesforce before you commit budget.
- Expecting it to fix a broken process. If nobody agrees how a lead becomes an opportunity, an agent cannot settle that argument.
Where a partner helps
Some teams can run all of this alone. Many cannot, mostly because the audit and cleanup work takes more hours than anyone has spare. That is where an outside team makes sense.
Concretio is a Salesforce consulting and development company, and this kind of prep work, meaning org reviews, integrations and custom builds, is our daily job. If you want a second pair of eyes on your org before you try AIforce, get in touch with the Concretio team and tell us what you would like to pilot. A short review usually shows what is ready and what needs fixing first.
Final thoughts
AIforce is less about a new product and more about a change in habit. Salesforce is moving from a place you visit to a system that shows up where you already are, in Claude, in Slack and in your own tools.
That is a good change if your foundation is sound and a stressful one if it is not. The order is simple. Fix the data, check the permissions, start small, and measure. The flashy part comes after.
Have you tried Claudeforce or Slackforce yet? Tell me in the comments what worked and what did not.