Free with a sign-in.
The best 10 min of 42
Marc Benioff Predicts Half of Conversations Will be With AI Agents Next Year
Marc Benioff says enterprises will pair AI agents with people, with agents handling half of conversations across key business functions.
The Logan Bartlett Show · Logan Bartlett
- Marc Benioff is cofounder and CEO of Salesforce.
The best minutes
- 2:19AI agents are filling Salesforce’s neglected sales pipeline3 min
- 12:56Benioff expects customer work to split evenly between agents and peoplePro pick 4 min
- 38:07Today’s LLMs are powerful but limited by finite data and algorithms3 min
Also worth watching
- 19:35Customers are asking for enterprise-wide agent pricing3 min
- 25:44Salesforce is testing pre-sale engineering to win deployments3 min
- 17:10Salesforce's AI and data line passed $1 billion3 min
- 8:11Salesforce is entering ITSM with a Slack-first agent model4 min
- 5:14Enterprise agents still need specialized application interfaces3 min
Skip the startup-investment stories, AI-company operating advice and career discussion before the final AI-limits debate.
Full analysis · 6 min read
Real Alpha · Nov 8
Marc Benioff says agents are headed for a 50/50 conversation split
- Benioff describes 1.5 million agent and human support conversations each over the next year, and a 50/50 direction elsewhere.
- Salesforce reports lower support staffing and about-equal CSAT, but omits cost per case and lead conversion data.
- Treat the split as a forecast to test, not a staffing plan; price integration and oversight too.
Free with an account
Read the rest of the analysis
7 sections, the 4 other voices it tests Benioff against, and what it means for you. Every episode we cut gets one.
Check your inbox
We sent a link to your address. Click it to finish signing up. Not there in a minute? Look in spam.
or
Signed up before? The same buttons sign you in. The episode, its moments and the transcript stay free.
Agents are reopening Salesforce’s unworked sales pipeline
Benioff says Salesforce had more than 100 million leads it could not call back over 26 years. The company lacked enough people to reach them.
Its agent now has conversations with more than 10,000 leads a week. Benioff says those conversations are turning into pipeline, while salespeople work alongside agents.
That is a capacity story, not evidence that agents replace sellers or close deals. For example, a lead revived after years without contact differs from a qualified buyer already in a sales process.
Benioff gives no response, qualification, conversion or deal-value rates for this pipeline. As Salesforce’s CEO, he is describing an internal test while advancing the company’s agent strategy.
“our pipelines have never been more full”
Our read
The backlog and weekly contact volume make a specific capacity case. Without conversion and revenue data, they do not establish financial return.
Support cuts accompany similar customer satisfaction
Benioff says Salesforce’s agent layer handled about 1.5 million customer conversations. Another 1.5 million went through human support in that period, with CSAT about the same.
Separately, he says he reduced support staffing from 9,000 heads to about 5,000. The episode does not tie that change directly to case costs or resolution time.
Those omissions matter because equal CSAT does not establish equal service economics. It also leaves unclear how many cases were escalated or differed in complexity.
Cal Newport said on Prof G Markets in October 2026 that enterprise AI investment is heavy, but returns remain mixed. He also identifies integration as a challenge.
Our read
Salesforce offers a meaningful operating signal, not a controlled test of savings. Buyers need service quality and per-case cost beside staffing counts.
Agents still depend on enterprise applications and data
Benioff rejects software’s obsolescence, saying enterprise work still needs specialized sales and support interfaces. He places those applications over data infrastructure that includes Data Cloud, MuleSoft, Tableau and Informatica.
That architecture is also a sales argument for Salesforce’s own stack. Its description shows what it sells, not that every buyer needs the same components.
For example, he describes Eaton’s field-service agent carrying customer and equipment context into the system of record for the next technician. The agent’s usefulness there depends on records and workflow carrying across visits, not just a language model answering questions.
Ari Weinstein on the a16z Podcast in October 2026 also expects agents to extend legacy software while new apps emerge. That comparison supports coexistence, not a guarantee that current interfaces keep their place.
Our read
The stronger case is for software changing shape, not vanishing. Integration and persisted context are central dependencies, but Salesforce’s preferred stack remains one vendor’s view.
Benioff sees a 50/50 conversation direction
In support, he describes about 1.5 million agent and 1.5 million human conversations over the next year. That is the function he dates specifically.
He says a 50/50 balance is where sales, service, marketing, field service and employee collaboration are going. He gives no date for that wider direction.
That is a forecast about conversations, not half of all work, payroll or headcount. Conversation counts may conceal major differences in task complexity and value.
Benioff says agents need tone and behavior guardrails, plus immediate escalation to humans when needed. So humans remain part of the workflow, not merely deployment supervisors.
Danila Shtan on Unsupervised Learning in November 2026 says adoption depends on reliable agent scenarios and outcomes. That comparison suggests operators should measure completed work and escalation, not conversation totals alone.
“50% are with agents 50% are with humans”
Our read
Treat 50/50 as a testable direction, not a staffing forecast. Measure completion and exception handling alongside the agent-human mix.
The billion-dollar line bundles AI and data
Benioff says Salesforce’s AI and data product line is its fastest-growing and has passed $1 billion. He says it is moving toward $2 billion, linking growth to customers harmonizing data.
But this is a broad AI and data line, not standalone Agentforce revenue. Benioff says Agentforce might fit at some level, but not at that same classification.
The host, not Benioff, then cites 8,000 Agentforce deals and more than $100 million in ARR. Benioff says updated numbers will come at earnings.
This distinction matters because investors may treat the billion-dollar category as agent revenue. The episode does not isolate agent revenue or deployment returns.
Still, category growth is evidence buyers are spending on bundled AI and data products. It cannot show which component earns that spend or what customer economics follow.
Our read
The billion-dollar number supports commercial activity around data and AI, not the 50/50 forecast or Agentforce’s standalone returns. Boards should ask for workflow-specific revenue and cost.
Customers want an enterprise package before pricing settles
Benioff lists seat, consumption and conversation pricing, including flex credits. He says customers increasingly want one enterprise-wide agent agreement.
That is a reported customer request, not a settled Salesforce rate card. The episode supplies no price, contract terms or realized spend under these models.
A seat bill may not fit work completed by agents; usage bills could instead vary with demand. For example, a conversation bundle may simplify procurement while obscuring task difficulty.
Benioff also says adoption speed differs with company size, architecture and deployment needs. So one license does not mean one rollout pace or one workload.
Aghi Marietti on MTS in October 2026 says production AI adoption can arrive in step changes after months of implementation. That points buyers toward staged budgets, with cost assumptions revisited at each production gate.
Our read
The pricing problem is unresolved, and deployment is not uniform. Compare total cost against measured outcomes before treating an enterprise package as predictable.
Benioff says today’s models remain bounded
Benioff says current language models have relatively finite algorithms and data, and he does not call them AGI. He thinks GPT-5 looks more evolutionary than revolutionary compared with GPT-4.
He sees room for another set of models and capabilities. This supports his case for continued human judgment and enterprise applications, while his company sells applications meant to work with agents.
Yet a general limit claim does not identify which workflows fail today or when a new model might change the boundary.
For example, Salesforce’s support story already includes guardrails and escalation, regardless of whether models improve soon. The practical near-term case is narrower: connected records, defined workflows and accountable handoffs support bounded deployment.
Our read
Model limits justify caution about full autonomy today, not confidence that today’s application interfaces will endure. Test the workflow, not the AGI label.
What it means for you
- Track lead conversion. Compare agent-contacted leads with a human-handled group. Track response, qualification, conversion and revenue, not pipeline volume alone.
- Measure support economics. Track CSAT, resolution time, cost per resolved case and human escalation alongside agent conversations and staffing changes.
- Map workflow dependencies. Use the Eaton field-service example to map which records and decisions must carry across visits, technicians and agents.
- Model contract scenarios. Compare seat, consumption, conversation and enterprise-wide pricing against the same expected workload and measured outcomes.
What to watch
- Whether Salesforce reports support cost, resolution time and escalation alongside CSAT.
- Whether the 50/50 direction appears across distinct functions, and whether conversation counts track completed work.
- Whether Dreamforce’s October 14 to 16 customer examples disclose outcomes beyond deployment descriptions.
Send this to the COO and support leaders setting agent targets and service staffing.
Watch the minute where Benioff gives the 50/50 forecast
Written by Real Alpha from the episode's transcript and the other moments it links, then checked against the transcript. Quotes are the speaker's exact words.
Minute by minute
A Pro write-up, open to you here. Try Pro free for 7 days
A Pro write-up. The minute by minute is in Real Alpha Pro; the episode, its moments and the transcript stay free.
Read it with Pro, 7 days free- 0:34Salesforce applies agents across operations
Benioff says Salesforce handled about 1.5 million customer conversations through its AI service agents, while human support agents handled the same number, with similar CSAT scores. He says he reduced support staffing from 9,000 to about 5,000 and shifted capacity toward sales. Sales agents now contact more than 10,000 leads a week, and a website agent handles visitor questions.
- 4:56Enterprise apps retain specialized roles
Benioff rejects the idea that software and applications will simply disappear. He argues that specialized workflows still need powerful interfaces built on trusted data. He says Salesforce’s data and application layers support agents in Slack that handle renewals, support and employee requests.
- 8:11Enterprise agents need integrated systems
Benioff says Salesforce is entering IT service management with a Slack-first agentic product. In examples involving Eaton and PG&E, he says field workers use agents alongside mobile apps, with interactions stored in a system of record. He argues that enterprise use requires data, applications, agents and language models to work together.
- 12:56Agents need human oversight
Benioff says agents need defined tone, guardrails and rules for escalating difficult cases to people. He expects conversations in sales, service, marketing and field service to move toward a 50/50 split between agents and humans. He compares an agent’s handoff to a self-driving car asking a driver to take control.
- 17:33Salesforce data products are growing fast
Benioff says Salesforce’s AI and data product line is its fastest-growing ever, has passed $1 billion and is tracking toward $2 billion in revenue. He credits demand for bringing customer data together through products including Data Cloud and MuleSoft, as well as the planned Informatica acquisition.
- 19:35Agent pricing has several models
Benioff describes four approaches: per-seat pricing, consumption pricing, conversational pricing or flex credits, and a broader agentic enterprise agreement. He says customers want to buy more of their needs in a package than he expected as they expand agents across business functions.
- 23:16Adoption varies by business size
Benioff divides the market into five segments, from small businesses to government and very large enterprises, and says their products and adoption speeds differ. He predicts AI will enable a sharp increase in small businesses. As an enterprise example, he says Pfizer has deployed Agentforce to 20,000 sales professionals.
- 26:32Salesforce is testing forward deployment
Benioff says Salesforce is inspired by Palantir’s forward-deployed engineering approach, in which engineers build applications before a deal closes, and is testing it to speed customer acquisition. He also says Salesforce beat Palantir for a publicly announced U.S. Army deployment, citing lower prices and easier deployment as advantages.
- 28:35AI startups still have room to grow
Benioff says Salesforce Ventures has backed companies including Anthropic, Wiz, Snowflake and Databricks. He describes meeting two 18-year-old founders building with Anthropic tools and credits, and argues that Silicon Valley innovation remains active.
- 31:53Leadership and execution still matter
Benioff points to You.com and Artera as AI-first companies in his portfolio, citing search tools and cancer diagnostics as distinct examples. He says leaders need a clear outcome, strong values, quality teams, fast execution and revenue growth, especially in sensitive areas such as cancer care.
- 35:54Young workers can spot new opportunities
Benioff advises young professionals to create and deliver value rather than rely on a fixed degree path. He says AI-native interns and young founders can spot gaps in business processes and products, and argues that companies would lose insight and energy if they avoid hiring people fresh out of college.
- 37:44Current language models remain limited
Benioff argues that current language models are powerful but are not AGI because they rely on finite algorithms and datasets and cannot perform every function. He describes GPT-5 as more evolutionary than revolutionary compared with GPT-4. He cites the body’s production of immune cells near sick people as a capability he says current models do not reproduce.
Published Aug 29, 2025.
Transcript · 34 min read
The Logan Bartlett Show, hosted by Logan Bartlett, with Marc Benioff.
From the transcript The Logan Bartlett Show publishes with the episode.
there were more than a hundred million leads that we have not called back at sales force in the last 26 years because we have not had enough people. So we just couldn't call them back, but we have now an agentex sales just for our own company more than 10,000 leads a week right now having conversations, turning them into pipeline. Our pipelines have never been more full and a lot of it has to do with that we're feeding them with this agentex capability and so our sales people are also now part and parcel in partnership with these agents as well. Mark thank you for doing this. Oh yeah it's great to be with you Logan.
So I think it's been eight months or so since we had the last conversation. I think we're seeing a lot of at least I'm seeing a lot of end of software. This is the end of application software as we know it going on. I guess what's your perspective over the last eight months and some of the rhetoric about this that we know we're no longer going to see the commercial opportunity for software that we've seen in the past. Well I think there's two things you know you look at the last eight months I mean it's been eight of the most exciting months I think of my career. When you look at what's happened with applications and with software it's pretty awesome. If you remember our conversation was all about agent force and we were getting ready to play agent force on our support layer and now we have and I just tell you like we're customer zero for our new agentech service and support product that sales force. So we have now done about a million and a half conversations with customers and at the same time that's the agentech layer speaking to the customers. A million and a half conversations also happened through our support agents during that same period and the CSAT scores were about the same which was studying and also I was able to
rebalance my head kind of my support. I've reduced it from 9,000 heads to about 5,000 because I need less heads but there's also an omnichannel supervisor now that's kind of helping those agents and those humans work together and this is the most exciting thing that's happened you know in the last nine months for sales force and we also are deploying thousands of customers with the same vision. I want to dive into some of those things that you brought up in a bit but I guess I'm curious the agent force and you mentioned the the cost reduction or the ability to kind of manage costs
Read the whole transcript
“we have now an agentex sales that is calling back every person that contacts us in our company”
you've seen a bunch of technological trends I guess do you think public market investors or investors at large or sort of waiting to see this come to fruition either in the growth rate acceleration or profitability before they really lean in on the themes or what's been the
“there's no better interface that I've seen or I would use it on top of our data foundation”
“the system of record the agentic layer the LLM all have to be integrated together”
for example like last night I was shopping for ties so I was you know on an LLM and I'm looking for the right tie of this kind of thing it wasn't one of the two that you mentioned it was a different one and then it brings me in there it's kind of a super search you know it's like helping me search to find the website and then I'm getting now I'm on the website but I still had to go to the website to complete the transaction to identify you know who I was complete the commerce will that maybe get more refined it probably will but when it comes to the fundamental trust security and information sharing in an enterprise that is a different ballgame right there what is what is managing agents and managing workers in tail you guys have gone through this transition on the
“and a million and a half human conversations that are going forward over the next year how now bifurcated 50% are with agents 50% are with humans”
that kind of balance I don't know if you've reported since May on numbers related to this what I
“we're a billion dollars and we're fast tracking the two billion in revenue on that product”
“customers want kind of the ability to buy all of it is one big package way more aggressively than I realized”
customers show other customers how they're doing you mentioned OpenAI and Anthropic there and we touched on a little bit of the scaling elements of the model today I'm sure you get to see a lot of stuff in preview or get to have conversations that a lot of people aren't privy to do you think we're increasingly moving to like an absorption and digestion phase where we just try to take a advantage of all the value that we've already presented or those model companies have already presented for enterprises to take advantage of or do you think we're still on a fairly steep slope and it's a little bit of ways from the harvesting stage it depends what category of the market we're
talking about you know the market is really in five segments you have the small and medium business you know the you know call it a couple hundred employees you have the medium businesses call it a couple hundred to a few thousand employees you know you have the large businesses call them like four or five thousand employees then you of course have the very large fortune-wet in your businesses any of the government and these five segments of the market each need a slightly different version of the product they need something with a slightly UI they consume it a little differently and the speed of their deployment is different because their architectures are deployment are different and also AI is going to dramatically change you know small businesses there's going to be a radical explosion of the number of small businesses because AI is going to make entrepreneurs so much more you know capable and probably meaty business to and of course in large business like mine it's happening but you know not everyone is probably as committed to deploying the technology rapidly and at the bleeding edge as I am so those five segments will all adopt the technology slightly differently and that's my observation of technology over 40 years
and so where do you think we are on that do you think we're at the absorption stage for for the mainstream enterprises or we're at the beginning of the beginning a lot of for big enterprises let's call them the big extra large enterprises you know they've a lot of them have tried things out some of the like in the dream force keynote Pfizer will be there they've deployed agent force to 20,000 you know their sales professionals they use our life sciences cloud which is now also a genetic that idea that we're able to kind of you know work with a company like that to kind of give them much more capability for their professionals that they're they're they're it's a force multiplier you know for their employees that is very powerful and for their customers as well and other customers need to see that in the pharmaceutical industry it's a great example you know it's a variation of companies who are willing to try technology to companies like Pfizer who are always the leaders in deploying that technology so you know I'm called the Albert Borla's CO Pfizer please come be part of my keynote that's very important for me because they're pioneers they're really doing an incredible job you know really delivering this next generation of AI one of the things
“our prices are so much better at lower cost and our technology is probably so much easier to use deploy”
pockets that you think are advantaged to be a startup these days versus in an incumbent in the in the market and you would say hey here's the opportunity for investment that might exist
outside of the core advantages that incumbency has well I mean sales force has been a very active investor you know we run our sales force ventures you know we own 1% of them thropic we just were part of selling whiz to Google you know which was a huge you know outcome for us I think more than a billion dollars and you know we tell we we took snow plate public that was more than a billion a billion and a half both of those I think were billion and a half I would say that I don't know what the IR exactly is in sales force ventures I think it's like 33% something like that when we're looking at these companies today and I just met with two entrepreneurs myself yesterday they're both 18 years old they're both like obidator they're both building on and thropic cloud they're both addicted to it because and thropic gave them 30 thousand dollars of free credits they're using their new Harmony API they're doing all of these things and I'm like whoa innovation is alive and well and Silicon Valley at every time you hear that's you know San Francisco is dead or something I'm like don't worry another gold rush is coming and then people never will go oh mark you don't know what you're talking about and oh yeah here we are it's a gold rush and
look at people are coming in from all over the world with their pans and their shovels and their picks and their got their blue jeans and their panting and I would say that it's a very exciting moment where you can see all these new technologies new ideas incredible innovation and things are moving fast and as evidence that I'm going to go back and watch what we said nine months ago and see how much of it has iterated evolved and changed and how much we've actually been able to get done because usually it's not these enterprise deployments are not done Logan in nine months you know it's like whoa I've got this oh yeah hey Logan I got this is my actual results here's the KPIs here we go and that's why I'm super motivated and excited I'm super energized and you know when you get to dream force this needs to be like a big message that I have which is hey you need to become an agentic enterprise here's the products every app is now agentic here's the data foundation here's all these pieces but I don't think customers care that much about what I have to say about the new products I think what they want to know is hey what did that person do what did FedEx did this and here's you know the CO FedEx explaining exactly what they
did here you know OpenAI did this and this is exactly how they did it and that I think is cool and that I think is another level of kind of capability that we have to be able to show and I'm going to pivot you know how I present my keynote based on that feeling I have which is that you know these customers like I'm built these big customers in Europe and they're like well I want to become an agentic enterprise these are customers that I've had for like one or two decades now I'm like hey we're going to move your whole infrastructure forward rebalance your head count the way I'm rebalancing mine change your business KPIs and you're going to come out the other side as an agentic enterprise. We kind of touched on it there and also when you were talking about the startups of the young kids going through why see these days but it does seem as a private investor there is a different rhythm or cadence that it seems the AI first companies are operating under and I think to some extent when your world around you is changing so quickly it forces an internal culture of change to be much more material in a meaningful way than like more traditional businesses I guess outside of adopting agent force in mass and you know all the good stuff that you guys
bring to bear up and down the stack are there are there things you've learned about how to change the speed of of moving a bigger organization that's gone through changes in the past but now has to operate at the speed of the competitors that were founded in the last couple years. Yeah I mean before talking to you I actually spoke you know to a company that I helped start in the AI world Richard Socher used to run Salesforce research you know has this incredible company called View.com and it's an AI first search company it's amazing it's scaling incredibly well it's just does incredibly accurate search based capability you probably have it on your phone if you don't it's pretty awesome they also provide an API approach which is widely consumed by some credible companies gives them a search infrastructure you know powerful and another example is I have another company called Artera which is an AI first company and that company is doing prostate cancer diagnostics got just got an FDA clearance it's about to do breast cancers do other ones and these are in my private venture portfolio you know which is time ventures so when I look at what Salesforce is doing what I'm looking I'm doing I would just say it's very similar you know
you have a great entrepreneur you know so you have Andre Esteva who's running Artera you have Richard Socher is running you.com they're visionaries they're experts by the way both of them use tour both of them are in Salesforce both of these to work together in the AI world they have a clear vision of what they're trying to accomplish one is doing you know next generation search through an API the other one is you know trying to provide democratizing high quality healthcare globally through AI it's leadership it's vision you know what do you really want what is your outcome where are you totally passionate and totally focused about what you're doing nothing is going to replace that and then the core values that they're building their companies on you know is it about trust is it about customer success is it about a vision and innovation I would say that when you're running something like how Andre's running which is you're in like the most sensitive aspects of somebody's life you know which is their cancer diagnosis I would say that that's something where all of a sudden you know you better have be completely buttoned up as a company and have the highest level of trust and capability and so your values matter so your vision matters your values
and then your execution matters the quality of your team matters how you are actually making things happen day-to-day matters and also are you rapidly resolving obstacles and what are your KPIs are you scaling your revenue both of those companies are scaling their revenue really aggressively like you need to scale your revenue and go because others will come for you well this has always been true and you know I've been in Silicon Valley my whole life I was born in San Francisco I grew up in the peninsula in Hillsborough my whole life has been in Silicon Valley and all the great entrepreneurs and all the great venture capitalists have proceeded us you know from the air of Arthur Rock and Sandy Robertson was one of my mentors you know to the great investors and entrepreneurs of today you know they have more things in common than different and one of them is you know you better go and you better go faster and you better go now I guess as we wrap one thing that I get asked a lot and I'm curious your perspective on this is like if you're a young person if you're 22 years old graduating college or whatever interested in entering a field in some way shape or form and AIs changing so much eight years ago I would have said just go get five years ago I would
say go get a CS degree and you know that'll benefit you no matter what you go do and now I'm not sure my council would be the would be the same anything you would give to someone that's maybe in the early part of their career and trying to figure out absolutely oh I told you I just met with this cool startup that's NYC with these two 18 year old founders you know that is amazing and and then I have two to interns from Stanford working in my office for the summer and hey create value create value and deliver deliver or something different and create something amazing and these kids I don't know you know these kids they're they're AI natives they're authentic natives they're not digital natives you know like they are they know intuitively what's possible and where this can go and then they're looking at my business process my product lines my capabilities and say and actually here's your gap and this is what can get filled in so that's very powerful and we need these people and I breathing these studies well we're not sure if we're going to hire these kids from college I mean what a mistake for companies not hiring people from college or bringing in new talents or bringing people in at that age I mean that is where the real value
and the insights and the energy is going to come from I guess one last one before we hop so you've been the CEO now sales force for 25 26 years you've seen every cycle that at least in my lifetime from dot com to mobile the cloud AI do you have a contrarium bet that maybe people look believe about AI today that we're going to look back on in five years and it's just going to be patently false
“we forget these large language models these are relatively finite algorithm sets with regular relatively finite data sets and those two things provide you some functionality but not all functionality”
I now that I'm invited I I was waiting to be invited I I'm happy to have you and it's going to be awesome you're going to love it and do you like Metallica or do you like Benson Boone it seems like it's maybe it's interesting I guess if those are the two extremes I would put myself in Metallica I thought yeah I thought so yeah well there's a whole group of folks who are Benson Boone people we're got both and baby Stevie Wonder might make an appearance you're going to do the Benson Boone outfit I hear big CEOs dress up I would not be oh and I will not be doing the back clip either but I do like the music but I of course a Lars I clip very close with as you know yeah Metallica does a lot with Salesforce so we love working with them now okay I will see you a dream force great to see you Mark it's great to see you again
