Many salespeople are already using AI:
Writing emails.
Organizing meetings.
Planning proposals.
Researching customer industries.
But every time they reach the most critical point, they get stuck.
No matter how smart Claude is,
it still doesn’t know:
Where your company’s deals currently stand.
What did the customer say last week?
How much pipeline is left this quarter?
Which deals are delayed?
Have quotes been revised?
Who is the account owner?
Before the next meeting:
What’s missing?
These answers are usually all hidden in:
Salesforce.
So the real sales workflow often becomes:
Ask AI first.
Then switch to Salesforce.
Look up accounts.
Check opportunities.
Review activities.
Reply on Slack.
Switch back to Claude.
And eventually piece together all the context yourself.
Salesforce and Anthropic are now ready to eliminate this back-and-forth.
The new solution is called:
Claudeforce.
What exactly is Claudeforce?
First, let’s clarify the names.
Claudeforce is the broader collaboration between Salesforce and Anthropic.
The first actual product released is called:
Salesforce in Claude.
It’s a Claude plugin that connects Salesforce’s
customer data,
pipeline,
workflows,
business logic,
and permissions
directly into Claude.
So, instead of:
exporting Salesforce data to Excel
and then feeding it to Claude,
Claude can, within authorized scopes, access
live Salesforce context.
The first phase focuses on sales
The initial version of Salesforce in Claude targets
salespeople.
Salesforce has prepared
37 pre-built Sales Skills.
These aren’t just generic chat prompts,
but functions designed specifically for real
sales workflows, such as:
Daily briefings,
pipeline reviews,
meeting preparation,
deal health reviews,
forecast narratives,
account planning,
activity logging,
CRM hygiene,
and win/loss reviews.
The easiest way to use it: ask "What should I handle first today?" in the morning
Many salespeople start their day not by making calls,
but by figuring out what’s happening today.
For example:
Which customers replied yesterday?
Which deals are about to expire?
Who hasn’t confirmed yet?
Which opportunity amounts changed?
Which deals are at risk?
Previously, this required clicking through multiple tabs.
The Salesforce demo for Claudeforce shows you can simply ask Claude,
"What should I pay attention to today?"
AI can organize:
live pipeline data,
highlight changes,
identify risks,
and recommend what to prioritize.
The biggest difference from general "help me plan my day" chatbots is the data source
Normal chatbots might say,
"Handle the most important client first today."
There’s nothing wrong with that,
but they don’t know
who the most important actually is.
Salesforce in Claude theoretically sees your
accounts,
opportunities,
pipeline,
activities,
and business context.
It’s not just guessing based on generic sales knowledge,
but grounding AI reasoning in
your company’s real sales state.
For example, when preparing to meet a client tomorrow
Previously you might:
search for the account,
check latest activities,
scroll through emails,
review opportunities,
discuss on Slack,
verify current quotes,
and then organize meeting notes.
You might lose 20 minutes before even meeting the customer.
Claudeforce aims to reduce all this to a single command
For example:
Prepare tomorrow's meeting for client ABC.
Tell me the deal status,
recent key interactions,
unresolved issues,
and the top three items to confirm in this meeting.
Claude then pulls from Salesforce,
Slack,
and other connected contexts
to create a comprehensive
meeting brief.
Note: this is not just a summary
The real value is that Claude understands
sales context.
For instance, in a generic meeting summary,
"The customer is still considering."
might just be a note.
Placed in CRM context,
it could mean:
the deal stage shouldn’t move forward,
forecast adjustments are needed,
and the next step needs follow-up.
This is where AI connecting with formal business systems makes a difference.
A second practical function: Pipeline Review
Sales managers often conduct weekly pipeline reviews to see:
Which deals might close soon?
Which are dragging?
Which deals lack next steps?
Which large deals have no recent activity?
Previously, they reviewed dashboards and then followed up manually.
Claudeforce wants Claude to understand pipeline status directly.
For example, ask:
Which three deals should I intervene in today?
AI uses deal status, customer interactions, and pipeline changes to shortlist candidates, then explains why.
This is more valuable than just listing the three biggest deals
Because the largest deals aren’t always the riskiest.
A $1 million opportunity might be progressing smoothly,
while a $300,000 deal might:
have had no next steps for three weeks,
decision makers absent,
and delivery deadlines approaching.
What sales managers really want to know is:
Where things are going off track.
The third feature: Account Plan
Suppose you just took over a large client.
You might spend half a day learning about the company’s history, orders, contacts, recent opportunities, lost deals, and current services, then prepare an account plan.
Salesforce in Claude can leverage existing company context
to generate an initial account overview.
But note this distinction
AI can
organize what’s currently in Salesforce.
That doesn’t mean Salesforce data itself
is complete or accurate.
For example, if sales reps haven’t updated contacts for six months,
AI only sees data from half a year ago.
So Claudeforce truly improves
the speed of accessing context,
not automatically making CRM data accurate.
The fourth feature: It can perform actions, not just read data
This is where Claudeforce truly differs from standard enterprise search tools.
Claude can:
update pipelines,
log activities,
modify CRM fields,
execute Salesforce workflows,
and even prepare external emails.
In other words, AI moves from
read
to
write.
But with write capability comes different risks
If AI misreads a deal amount,
it just answers incorrectly.
But if it actually changes the deal amount in CRM,
then the data is wrong.
If it follows up by sending an email to a client,
it extends into external action.
The maturity of enterprise AI depends not on whether AI can act,
but on
how far those actions are allowed to go before stopping for human confirmation.
Salesforce emphasizes one thing: existing permissions still apply
All answers and actions performed by Salesforce in Claude
still respect the original Salesforce
permissions and business rules.
So if an employee does not have access to an account,
connecting Claude won’t suddenly grant access.
This is far more important than building a separate AI permission system
Companies worry about years of carefully managed Salesforce permissions
being bypassed by a new AI permission layer,
causing conflicts.
Salesforce’s approach lets AI
directly follow existing enterprise rules.
For example, if Sales Rep A can only see Taiwanese clients
then Claude should only access the data Rep A is authorized for.
Sales managers who can see entire regions enable Claude to see more accordingly.
This means
identities and permissions follow the user.
However, “correct permissions” doesn’t mean data should be used carelessly
If a record is mistakenly set to be viewable by the entire company,
Claudeforce will still see it under existing permissions.
So
“AI not overstepping”
and
“correct company permission design”
are two separate matters.
The fifth feature: External emails can require human approval first
Salesforce states that on the write side,
companies can control
when Claude needs to ask for confirmation,
such as
sending emails outside the company.
It can prompt for you to check first,
or with enough trust in the company’s processes,
allow higher autonomy.
This is an important enterprise agent concept.
Don’t ask if AI can auto-send emails
The real question is:
Which types of messages, under what conditions, may be sent automatically?
For example:
Internal reminders might be low risk.
Customer thank-you emails might also have lower risk.
But quotes, refunds, contract terms, delivery commitments, and discounts are very different.
Agent automation shouldn’t be a simple ON/OFF switch
Different actions require different boundaries.
This is why Claudeforce is more valuable than generic "AI helps write emails" tools.
The sixth feature: Claude can generate interactive new work interfaces
Salesforce stresses that Salesforce in Claude is not just a traditional chat box.
Claude can create interactive views tailored to your work, such as displaying your
accounts,
pipeline,
and live data
organized into customized dashboards.
This reflects a bigger product shift.
Previously, enterprise software interfaces were fixed
Salesforce engineers designed
where buttons, tabs, and dashboards appear,
and users had to learn to navigate them.
Now AI interfaces will appear based on
what you want to accomplish.
The interface adapts to the task.
For example, in the morning you may want a daily pipeline overview,
in the afternoon a customer brief for meetings,
and month-end forecasts.
These don’t require the same dashboard.
AI can make the interface change dynamically with the work.
Salesforce even summarizes this as:
UI is the AI.
In other words, you no longer need to memorize where Salesforce menus are.
You just tell the AI what you want to achieve,
and it navigates the system to get it done.
This may be more important than how much stronger the Claude model becomes
Because what trips up employees most isn’t describing questions,
but finding where data is located:
which system, which page, which report, which filter.
If AI can bypass the UI,
enterprise software usage can be redesigned.
But Claudeforce isn’t yet available to everyone
This must be made clear.
As of now,
Salesforce in Claude is only open to selected pilot customers.
Salesforce plans an
open beta in September 2026.
So if a typical Salesforce user opens Claude now
and doesn’t see it,
it’s not a settings error,
but because it hasn’t been broadly released yet.
Also, the first phase covers only sales
The announced 37 skills mainly address sales.
Salesforce previews that later
service, marketing,
and commerce areas
will be added,
but they’re not ready now.
So today the actual product available is:
Salesforce in Claude for selected pilots
37 sales skills,
Salesforce context,
pipeline workflow,
and sales actions.
Additional modules for service, marketing, and commerce will follow later
Don’t confuse “coming soon”
with “available now.”
How does Claudeforce differ from Glean, introduced on August 13?
Both can access enterprise data,
but their focus differs.
Glean is more like a company-wide AI search portal
Its strength is gathering data from systems like
Drive, Slack, Microsoft 365, Salesforce, Jira, etc.,
into a unified enterprise search and work AI hub.
The core question is,
"Where is the company data?"
Claudeforce is closer to "formal work inside Salesforce"
It doesn’t just find customer data,
but also understands pipeline, accounts, business rules,
and executes CRM actions.
The core question is,
"After knowing the data, what should be done next with this deal?"
This means the two tools may not replace each other
Company data is often scattered,
so enterprise search tools are needed.
Sales processes formally run inside Salesforce require
deeper CRM agents.
The future may have multiple AI agents
integrating with different formal systems.
A seventh notable point: Claudeforce does not copy Salesforce data into Claude
Salesforce uses architectures like AIforce, Headless 360, MCP,
allowing Claude to access Salesforce data and capabilities
via enterprise-controlled gateways.
For most users, the key takeaway is:
AI uses Salesforce capabilities through company-managed interfaces.
It doesn’t require employees to
download CSV files every day
and upload them manually.
This is the difference between scalable enterprise AI and ad hoc demos
A company with 1,000 salespeople can’t have each person
setting up APIs, servers, authentication, prompts, and permissions themselves.
True enterprise products let an admin set it up once,
then the team uses it within existing governance frameworks.
This enables company-wide deployment.
Salesforce reveals they themselves were once doing manual integrations
Early on, their sales team and Anthropic built their own MCP connection, skills, and auth management.
Salesforce in Claude packages all this
as a plugin.
For enterprises, this productization is the real value,
not just "a technical possibility."
No need for each company to reinvent the wheel.
This issue closely mirrors the humanoid robots introduced earlier today
A demo showing it can be done
versus
daily utilization by enterprises
are completely different challenges.
A sales engineer might spend weeks to connect Claude and Salesforce
to demo the system,
but enabling 10,000 salespeople to start using it tomorrow
requires permission management, governance, deployment, support, and business rules.
These define an enterprise-grade product.
An eighth standout point: Salesforce itself is Customer Zero
This solution isn’t just for demos.
Their own sales team is participating in the pilot,
and Anthropic staff are involved in real work too.
This Customer Zero model means
the product is tested with
real pipeline,
real accounts,
and real deals.
However, official use cases can’t be directly taken as your ROI
Salesforce claims their sales team has strong demand for this tool
and that Slackbot and Claude-powered functions drive significant
annualized productivity gains.
These figures serve as internal reference
but don’t guarantee your company will save the same amount of time after adopting Claudeforce.
Actual results must be measured internally
For example, a 20-person sales team can track:
Meeting preparation time
Before: average 25 minutes
After: how much?
Weekly pipeline review time
How long did managers spend before?
CRM update delays
Are salespeople updating only once a week on Fridays?
If Claude updates smoothly, is data more timely?
Deal data errors
Will errors increase after AI actions?
External miscommitments
This should ideally stay near zero.
AI easily creates the illusion that more CRM updates mean better sales
Not necessarily.
If salespeople rarely updated CRM before,
and now Claude updates 100 activities daily,
CRM data looks great,
but customers may not buy more,
and revenue may not rise.
So don’t just measure how many AI actions occur,
but also watch sales cycle length, conversion, win rates, revenue, and customer relationships.
A ninth valuable feature: Deal Health Review
Many sales managers pay special attention here.
Because a deal sitting in the pipeline doesn’t guarantee it’s healthy.
AI can combine
recent activities,
deal stage,
next steps,
and customer interactions
to detect abnormalities.
For example:
An opportunity is still marked as under negotiation,
but no customer interaction for 21 days,
decision makers never involved,
close date changed thrice,
next steps missing.
AI can flag:
This deal might be at risk.
But “risk” is a judgment, not a fact
Managers shouldn’t immediately
cut such deals from forecasts
based solely on AI risk scores.
Better practice is
AI points out areas worth investigating,
and humans decide what to do.
This is similar to recommendation systems
The real value of AI
is not announcing truth,
but extracting a small set of items deserving attention from large data.
Sales managers used to review hundreds of opportunities.
If AI highlights 10 abnormalities first,
people focus on higher-value tasks.
The tenth feature: CRM Hygiene
Many Salesforce projects struggle not because the system is bad,
but because data isn’t updated.
Contacts leaving six months ago,
expired close dates,
blank next steps,
incorrect opportunity stages,
and duplicate accounts
all degrade forecasts and dashboards.
If Claude can daily remind users about
incomplete, outdated, conflicting, or duplicate data,
its true value may lie less in writing emails
and more in keeping the CRM clean,
which benefits all subsequent AI functions.
Because AI acts as an amplifier
Clean data amplifies insights,
bad data amplifies errors.
If your Salesforce is messy,
adding Claude won’t magically create a perfect CRM,
it just speeds up using that messy data.
So when first introducing Claudeforce, start with low-risk tasks
Such as:
1. Morning Brief
Just identify key deals for today,
without changing anything.
2. Meeting Preparation
Have AI organize customer context,
but don’t send any emails yet.
3. CRM Hygiene
Identify missing or outdated data and anomalies,
list issues instead of making mass changes directly.
These three are easy and verifiable starting points.
When the team trusts the system, gradually expand actions
Like logging activity,
updating confirmed next steps,
modifying simple fields,
and preparing follow-up drafts.
Only later consider higher autonomy.
Don’t delegate your entire pipeline to Claudeforce from day one
Every deal in the sales pipeline represents
revenue,
forecasts,
commissions,
and client commitments.
Jumping into full automation immediately is risky.
The rational enterprise agent rollout principle is:
Start narrow, then go deep.
Not the other way around.
How should a typical sales rep understand Claudeforce?
No need to think about MCP, Headless 360, or AIforce.
What matters daily is:
Previously, when asking Claude,
"How should I follow up this client?"
Claude could only answer based on the data you pasted in.
In Salesforce in Claude, it first knows
what’s really happening with that client now,
then suggests next steps.
This is the difference between "generic AI" and "work AI"
Generic AI
knows the world.
Work AI
also knows
your company.
Generic AI can talk about general sales best practices.
Work AI knows
the specific deal,
this client,
pricing,
recent interactions,
and applicable business rules.
The truly valuable enterprise agent connects both capabilities
Claude handles
reasoning,
while Salesforce holds
business truths,
permissions,
and workflows.
If AI only reasons without company data,
it’s just a smart outsider.
Conversely, Salesforce alone isn’t enough
It contains hundreds of thousands of records,
but users still have to click, search, and organize manually,
so data value is hard to fully realize.
Claudeforce aims to merge
AI intelligence
with
enterprise systems.
This will be the most important SaaS evolution in the future
Previously, people worried whether
Claude, ChatGPT, Gemini,
would replace Salesforce.
Now the emerging reality is:
AI doesn’t eliminate SaaS,
but becomes the business engine behind SaaS.
Users won’t necessarily open the original SaaS interface every day,
but customer data, business logic, permissions, workflows, and transaction records
still reside inside SaaS platforms.
AI just adds a
new interface layer.
This might be a major step forward for enterprise software.
Can anyone try it now?
Currently,
most people cannot.
Salesforce in Claude is limited to selected pilot customers.
Open beta is planned for
September 2026.
So the best approach now isn’t searching for install buttons,
but understanding what these new enterprise AIs change.
If your company already uses Salesforce, prepare three things first
1. Understand permissions clearly
Who can see which customers?
Which deals?
Which sensitive fields?
2. Improve CRM data quality
Fix expired, duplicate, blank, or incorrect fields.
3. Decide which actions always require confirmations
For example: external emails, discounts, quotes, close dates, critical stages, client commitments.
Don’t wait until the agent goes live
to figure this out.
The best agent guardrails aren’t post-incident prompt fixes,
but clearly knowing from the start:
What AI can suggest.
What AI can modify on its own.
What must be approved by a human.
Claudeforce’s true value ultimately depends on companies’ ability to define these layers.
One key takeaway:
Claudeforce is not:
just putting Salesforce into another chat box.
The real change is Claude starting to work within
official company data,
permissions,
and business rules.
You no longer need to
open ten CRM pages
to piece context for AI.
AI can directly
understand pipeline,
prepare meetings,
identify deal risks,
and update some CRM work.
However, this also means Claude goes beyond
simply giving advice,
and begins interacting with
your company’s real operational status.
At this stage, the key issues become not
"Is AI smart enough?"
but
Is the data accurate?
Are permissions reasonable?
Which actions can be automated?
When must humans stop and approve?
When enterprise AI moves from
answering questions
to
modifying CRM,
we enter the agent work era.
Today, let’s progress a little with AI.
Learn one AI skill each day.
Save a little time every day.
Improve a bit every day.
SasaDaily, growing with you.
