Salesforce, HubSpot, Anthropic, and the fight over where your work actually happens. Every number below is labeled official or third party.
If you only read one box, read this one.
The CRM is being demoted. For twenty five years it was a place you went. Now it is becoming a thing your assistant reads. Every move in the last three weeks is a different company answering the same question: if the human never opens the app again, who gets paid, and for what?
The clean test is her own account. Of everything added to Olga's CRM in 2026, six percent was typed in by a human sitting in the CRM. Ninety four percent arrived another way, and every record added since late July came in through the API. She has not opened the app in months and it is still the tidiest record she has. That is the demotion, already done, in a business with no Salesforce budget.
That is why it feels like a land grab. It is. But the prize is not the CRM. The prize is the window the human is looking at when the work happens. Salesforce decided it cannot win that window, so it is selling itself into every window instead. HubSpot's co-founder went the other way and started building the small version from scratch. Jack Dorsey built a new window. And the money underneath all of it is the same problem: software has been priced per human seat, and agents do not need seats.
Dated, so you can say it out loud without hedging.
This is the frame that makes the mess legible. Everything else on this page hangs off it.
The CRM screen is dying. The CRM as a system of record is not. Everyone in this market is racing to become the layer that agents have to call, even if the user never opens their product again.
Headless is the admission. When a company this size tells you that you may never open its app again, it is conceding layer one and defending layer two. It is not giving up the ledger. That is the entire game, and the reason four unrelated product launches in six weeks are actually one thesis attacked from different altitudes.
And it is also a meter grab. If the agent does the clicking, per seat is the wrong meter. That is why Agentforce is priced per action, why HubSpot moved agents to outcome pricing, why Zendesk charges on resolution, and why YouSpot is one dollar plus pay as you go. Every one of these companies is changing what it counts at the same time it changes what it sells.
Put that next to Benioff's "the UI is the AI" and Amodei's "the systems where much of the world's commercial activity happens." Four executives at three companies describing different floors of the same building. They are not confused, and neither is the market. It has just never had to price a building like this before.
They are not the same move. This is the part that was confusing.
Claudeforce is not Salesforce picking Anthropic. Salesforce did the identical integration with OpenAI eight months earlier and shipped Agentforce Sales inside ChatGPT. It is now inside both windows.
The word doing the work in the press release is governed. Salesforce is not fighting to keep you in its app. It is fighting to make sure that when an assistant answers a question about your pipeline, the answer comes out of Salesforce, with Salesforce's permission rules attached, and Salesforce bills for the action. Benioff said the quiet part in his own quote: "the UI is the AI."
YouSpot is not a spinout and Dharmesh has not left. The site says made by HubSpot Labs, his newsletter says it is part of HubSpot Next, and he is still co-founder and CTO. Separate product, separate domain, separate billing, one dollar a month against a twenty five dollar list price.
That is the more interesting version of the story. HubSpot is deliberately trading near term growth for AI adoption, and the stock sits roughly half its 52 week high. So the company is paying its own co-founder to build the thing that could eat the bottom of its market, rather than watch Attio, Clay or folk do it from outside. Clay raised again in July 2026 at a reported 6.56 billion valuation. third party
His framing in the welcome email is the whole product in one line: "It's not a CRM with AI inside; it's AI with a CRM inside."
Salesforce made Claude the default model in Slack, powering Slackbot and a Claude Tag for team decisions. Block released Buzz to put humans and agents in one open source workspace. OpenClaw does the same thing for one person at a desk.
Three very different companies, one premise. The place where work happens is a chat room that agents can enter. The CRM, the doc, the ticket, the repo all become things the agent reaches into from the chat room. That is the actual land grab, and the CRM story is a symptom of it.
Both companies gave up on being the window. They disagree completely about what to sell instead.
Read those next to Benioff and the difference is clean. Salesforce picked partners and is selling the trust layer between the agent and the data. HubSpot refused to pick and is selling the pipe. Salesforce's bet pays off if enterprises will not let agents touch a system of record without governance. HubSpot's bet pays off if the assistant people use keeps changing and nobody wants to be locked to one.
Five things in three months. That is the whole list, so nothing was missed.
Short answer: it is not a two horse race, and HubSpot is not second.
IDC's worldwide CRM applications share for calendar 2025, published in 2026, is the last real scoreboard. Salesforce 20.0 percent, its thirteenth straight year at number one. Oracle 4.1. Microsoft 4.0. Adobe about 3.4. SAP about 3.1. Everyone else, 65.4 percent.
HubSpot is not in that top five. So the Salesforce versus HubSpot framing everyone uses, including this show, is a story about mindshare and about who owns small and mid sized business, not about market share. By revenue they are not close. Salesforce guided to 46.1 to 46.4 billion for the year. HubSpot guided to about 3.68 billion. Salesforce is roughly twelve times the size.
Flip the measure and it inverts. HubSpot had 306,446 customers as of June 30. Salesforce's own marketing number is 150,000 plus companies. HubSpot serves twice as many businesses on a twelfth of the revenue, which tells you exactly which one is in the room with a viewer of this show.
Microsoft sits at 4.0 percent of the CRM market and does not break out Dynamics 365 revenue at all. But Microsoft does not have to win CRM. It already owns the window: Outlook, Teams, Copilot, the desktop. And it is running the same play, shipping a service MCP server with 90 plus tools so agents can operate Dynamics from wherever the person already is.
That is the point worth making on air. If the fight is over where the human is looking, the company that already owns the screen starts ahead, and it is not Salesforce or HubSpot.
The other 65.4 percent is the real headline. Two thirds of the CRM market is not any of these names. Zoho claims over a million paying customers and 150 million users, though that is a company marketing figure with no audited revenue behind it. company claim Attio, Clay and folk are taking accounts that never needed the enterprise version. And YouSpot at one dollar a month is HubSpot deciding it would rather compete with them than lose to them.
Follow the pricing, not the press releases.
Seat based pricing is the whole problem. Enterprise software has been sold per human for two decades. If an agent does the work of the person, nobody buys that person a seat. That is the entire content of the February selloff, and every company here has already re priced around it.
That post is the death of the seat with a price tag on it. A dollar a lead. Fifty cents a resolved ticket. Nobody is buying a license for a person in that sentence.
So the fight is not really Salesforce against HubSpot. It is every incumbent trying to move its revenue from "how many people you employ" to "how much work got done" before the market prices them as a dying seat business.
All verbatim, all sourced. These are the on air quotes.
The pair worth putting side by side on air is Benioff and Dharmesh. One says the UI is the AI. The other says it is AI with a CRM inside. The biggest CRM company and the co-founder of the second biggest are both saying the app stopped being the product.
Closing prices, Thursday Aug 27 2026.
Salesforce's 52 week range is 146.32 to 269.11, so one earnings print took it from the bottom of its year to near the top. HubSpot's range is 169.63 to 525.51 and it is nowhere near the top. Same week, same sector, same AI story, two completely different verdicts.
The read: the market is no longer paying for growth. HubSpot grew revenue 20 percent and got punished. Salesforce grew 11 percent and got rewarded, because it showed Agentforce and Data 360 ARR near 3.9 billion. Growth is not the currency anymore. Proof that the AI line item is real is the currency.
The story ran as a miss. Rangan called it a decision, on the call, in advance.
The April changes were free trials on agents and outcome based pricing. Both extend the sales cycle on purpose. HubSpot then lowered its own guidance to match: full year revenue from 3,700 to 3,708 million down to 3,678 to 3,686 million, and net customer adds from a 9,000 to 10,000 range down to 5,000 to 6,000 a quarter.
That is the sharpest version of the two bets. Salesforce is defending the enterprise by becoming the governance layer under everyone's assistant. HubSpot is deliberately shrinking its own near term growth to move its customers onto AI pricing faster. One is protecting a moat. The other is paying to cross a bridge before the bridge is gone.
The 70 percent stat Olga remembered does not exist. Four that do are below.
It was hunted through Salesforce State of Sales, HubSpot, Gartner, Forrester, CSO Insights, Validity and Spotio. There is no primary study behind it. It circulates on vendor blogs as a rounding of a different number, the share of a rep's time spent not selling. Saying it on air is exactly the kind of clean sounding claim that gets caught.
One to leave alone: a widely repeated "68 percent of reps say note taking and data input are their most time consuming tasks" is attributed to Salesforce State of Sales 2026 but only appears on a vendor blog. It could not be confirmed in the report itself. Skip it.
Put those next to the six percent and the segment writes itself. Three quarters of companies cannot trust half their own CRM data. Two thirds of reps burn five hours a week typing into it anyway. Three percent have automated it. That is the gap every announcement in this deep dive is aimed at, and it is why the interface is the thing being fought over rather than the database.
Revenue share says Oracle and Microsoft. Customer count says something different, and it is messier than anyone admits.
The honest answer to "who is third." Nobody can prove it. After HubSpot, the largest audited sales CRM customer count is Freshworks at roughly 75,000. Pipedrive claims more but files nothing. Zoho is probably larger than both and refuses to separate CRM from everything else it sells. Every "top three CRM" list online is built on marketing claims stacked next to filings, which is why they never agree.
HubSpot's own research says 65 percent of sales professionals use a CRM. The other third run their pipeline on email and a spreadsheet. That is a bigger installed base than Freshworks, Pipedrive, Close and Copper put together, and it is the group every product in this deep dive is actually competing for.
For a viewer of this show, that is the useful frame. The fight is not Salesforce versus HubSpot. It is structured software versus the inbox, and AI just gave the inbox a way to organize itself.
Four things to be careful with. The first one is the one worth saying on air.
Salesforce Ventures has held an Anthropic stake since 2023. Bloomberg valued it at about 5 billion dollars in June. The Q2 FY27 10-Q says the quarter's upward adjustments "include unrealized gains of $2.7 billion related to the Company's investment in Anthropic," with total net gains on strategic investments of 2,613 million dollars.
None of that appears in the earnings press release. It is in the 10-Q. And the same day, the company announced a deep partnership with the company it owns a piece of. Nothing here is improper, and Salesforce discloses all of it. But it explains the feeling: GAAP earnings per share rose 119 percent and non-GAAP, which excludes the investment gain, rose 103 percent. The operating quarter stands on its own. The headline number had help from a stake in the partner.
From Salesforce's own earnings release: "Effective Q2 FY27, Agentforce ARR includes our AI offerings, Slackbot and Headless 360." The headline 1.5 billion and the 240 percent growth are measured against a definition that widened this quarter. That is not a critic's claim. It is in the company's own filing. It does not make the number fake. It does make a clean year over year comparison impossible, and nobody on the coverage mentioned it.
The list going around says Figma down 76 percent and monday.com down 72. Checked against market data: monday.com is down about 45 percent over the year and Figma about 56. Both bad, neither what the list says. The 285 billion in 48 hours and the roughly 2 trillion since October 2025 are third party analyst compilations, not an official index figure. Say "reported" or do not say the number.
Select pilot customers now, open beta expected September, more skills late 2026. The 37 sales skills are a list, not a product anyone has run a quarter on.
Salesforce shipped the same integration into ChatGPT in December and January. It is inside both. Anyone framing this as an exclusive alliance did not check the last eight months.
Viewer value first. Her story is the proof, never the headline.
Cold open, ready to read: "In twenty four hours, Salesforce put its entire CRM inside Claude, and HubSpot's co-founder launched a one dollar AI native CRM for one person companies. That is not two product launches. That is the industry admitting the CRM app is optional."
Then the viewer's question, which is not "what did Salesforce announce." It is: is the software I pay for every month about to stop mattering, and what do I do about it.
Thirty seconds a layer. Brains: Anthropic, OpenAI, Microsoft. Rooms: Claude, Copilot, Slack, Buzz, OpenClaw. Ledgers: Salesforce, HubSpot, Dynamics, and now YouSpot at the bottom.
Then the one sentence that ties it: the fight is no longer who has the best CRM. It is who owns the customer memory and the right to act, when the interface might be Claude, ChatGPT, Slack, or an agent running on a Mac Mini in your kitchen.
Seats. Software is priced per human and agents do not need seats. Agentforce is now ten cents an action. Zendesk charges when the AI closes the ticket. HubSpot charges on outcome. That is the whole story of the year in enterprise software.
Benioff: the UI is the AI. Dharmesh: it's AI with a CRM inside. Read them back to back and let the audience notice they are the same sentence from opposite ends of the market.
The demotion already happened inside her own business, and it is measurable. Pulled live from her own CRM on Aug 27, stated as percentages so no account size is revealed: six percent of everything added this year was typed in by a human sitting in the CRM. Ninety four percent was not. And every single record added since late July arrived through the API.
The line that lands: the job her assistant used to do, opening the CRM and typing what happened, is now six percent of how her CRM gets filled. That is a stronger demo than anything Salesforce announced this week, because it already runs and it cost nothing.
Optional live layer: she is in the YouSpot private beta as of Wednesday night. Connect Gmail, LinkedIn and X, then ask "who do I know at Gusto?" If it is not solid by morning, cut it and keep the six percent.
Precision notes for air. Say percentages, never counts. The ninety four percent covers email logging, the browser extension and the API together, so it is "did not require a human to open the CRM and type," not "Claude did it." The API-only share is smaller, and it is the share that is growing.
The transferable move is not buying anything. It is the shift from storing records to storing context. Pick the one system your assistant should be able to read, connect it, and stop maintaining the rest by hand. That is the takeaway that survives whether Claudeforce ships or not.
Leave the tension open rather than resolving it. If this works, Salesforce and HubSpot get more valuable as invisible infrastructure. If agents get good enough to keep their own memory and act with light governance, the ledger gets commoditized and the 2026 drawdown was the market being early, not wrong. Wednesday's 22 percent day is the market choosing early. It has not chosen wrong.
Then one question: if your assistant can read your pipeline, your inbox and your calendar at once, what is the part of your job that is still yours? Then the ask, on its own card, last position, per the Ep52 lesson.
Primary first. Anything not on this list did not make it into the page.