Two CRM companies, one category, opposite physics: Attio, a London challenger that 4x'd ARR on a motion where the product does the selling, and Salesforce, a $41.5B incumbent whose "digital labor" story is real exactly where it cuts costs and absent exactly where revenue gets made.

What's actually AI-native, what's theater, where the investor story and the GTM motion diverge, and the three moves I'd hand Attio before it imports the disease it's disrupting.

In December 2024, Marc Benioff went on the 20VC podcast and said something CEOs usually save for the board room: "We're not adding any more software engineers next year because we have increased the productivity this year with Agentforce and with other AI technology that we're using for engineering teams by more than 30%." In the same interview, the other half of the trade: "We will probably add another 1,000 to 2,000 salespeople in the short term."

Nine months later he confirmed the support side of the ledger, on tape, in a sentence his comms team must replay at night: "I was able to rebalance my head count on my support. I've reduced it from 9,000 heads to about 5,000 because I need less heads."

Sit with the shape of that. At the company that coined "digital labor," AI has replaced the people who answer customers and frozen the people who build product. The one function AI apparently cannot touch is the one Salesforce is hiring thousands of humans into: selling the AI.

The number I keep staring at isn't revenue. It's sales and marketing expense: $3.77 billion in Q1 FY27, 34% of revenue, essentially flat as a percentage year over year.

Salesforce has spent eighteen months telling the street that AI agents do real work. If that were true of go-to-market work, this is the line that would compress. It hasn't moved. The digital labor revolution stops at the door of the sales department that's selling it.

And the growth arc underneath: +25% in FY22, +18% in FY23, +11% in FY24, +9% in FY25, +10% in FY26, with that last point flattered by an acquired company (Informatica contributed $399M of FY26 revenue). Attio's curve points the other way, off a base it won't disclose. That asymmetry, a private 4x against a public 9%, is the whole category fight in one sentence.

Six dimensions, loosely graded, plus a composite for how AI-native each company's own go-to-market actually is. Not the product demo. The motion.

The grades are the hook; the evidence is the rest of this teardown. Two rows deserve a flag before we walk them. Salesforce's in-product AI grade is higher than you'd expect from a 27-year-old company because its internal deployment is genuinely real (its own help portal runs at a claimed 75% AI resolution rate, and the 4,000-person support cut is the receipts). And Attio's outbound grade is lower than its branding suggests, because the most AI-native CRM company in the world is currently hiring a conventional SDR team. We'll get to both.

Attio's founding story is a teardown of its own. Around 2019, Nicolas Sharp and co-founder Alexander Christie decided to attack the largest category in B2B software armed mostly with the conviction that the incumbent product was beatable. Sharp's version, told to GV this March: "Alex went back to his desk, put on his headphones, and did all the onboarding for Salesforce. He looked at their API and assessed the overall user experience, and decided there was a lot that could be improved." And then, the line every founder will recognize: "I think we had the right amount of naïveté. If we had known more about what we were getting into, we might not have done it."

What they built in the following years (Sharp says the London team worked "almost completely undistracted, 16 hours a day" for three of them) is an argument about where AI value actually lives. Not in the chat box. In the data model. Sharp, in the GV interview that accompanied the Series B: "For AI to have the context it needs to really provide the value that it can, it needs much more context than legacy solutions are able to provide... The CRMs were historically a system for humans to transcribe things that happen in the real world. It doesn't make sense anymore. And so, all the way from the data stack right up to the UI, everything that came before has been invalidated by LLMs."

That architecture position produces a distribution position, and this is the part GTM people should study. Because the CRM populates itself from email and calendar data instead of demanding manual entry, the product onboards itself. Because it's self-serve with a free plan and no credit card, with AI features included from the free tier up, the product demos itself. One detailed teardown of the motion (Kasper Dudek, February 2026) put 75% of signups as organic, with near-zero outbound spend through the first 5,000 customers. The launch playbook was pure design-led challenger: a waitlisted private beta, build-in-public updates, then #1 on Product Hunt timed to the day after the Series A announcement. The aesthetic peer group Attio claims (Linear, Notion, Superhuman) is itself a targeting decision: it tells design-sensitive, AI-native teams this is yours, and the customer wall confirms who heard it.

The newest distribution surface is the most AI-native thing any CRM company has shipped: a hosted MCP server that gives Claude, ChatGPT, Cursor, or any agent read/write access to the workspace, on all plans. Most vendors treat AI access as an enterprise upsell. Attio treats it as a top-of-funnel surface, betting that when agents are doing GTM work, the CRM agents can actually operate becomes the default. The content motion runs the same logic at the human layer: GTM Atlas, an ungated operator library written with people from Lovable, Vercel, and Framer, aimed explicitly at "the first GTM hire who owns everything, the founder building revenue infrastructure before making a dedicated hire." That is not an audience description. That's the revealed ICP, in Attio's own words.

And the investors are underwriting precisely this loop. GV's stated reason for leading the round: "As customers discover Attio, they recommend it so frequently that the company has achieved 4x year-on-year growth at significant ARR." The product is the pipeline.

Salesforce's motion is the most successful field-sales machine ever assembled, and you can read its physics straight off the company's own surfaces.

Start with careers: a full SDR-to-enterprise-AE ladder, plus a new "Agentforce AE" overlay whose job description targets "CIOs, CTOs, and solution architects." Bloomberg reported the company hiring 1,000+ people specifically to sell Agentforce within weeks of launching it. Above the reps sits the events machine (Dreamforce as the launch vehicle, Agentforce World Tours through London and Boston this month, a branded beach at Cannes Lions), the Matthew McConaughey Super Bowl spots, and the partner layer: Accenture, Deloitte, Cognizant and an implementation economy of 30+ named partners who make money configuring what the AEs sold. The product is sold first and built second, by a supply chain of humans.

Now look at what the buyer signs. List pricing runs $25/user/month for Starter to $550/user/month for "Agentforce 1 Sales," every paid tier billed annually with no monthly option, and the real-world median contract lands at $75K a year (Vendr, across 2,216 deals). In August 2025, mid-AI-narrative, Salesforce raised list prices ~6% across Enterprise and Unlimited, its first broad increase in roughly seven years. Agentforce's own pricing has been repriced twice in 18 months: the launch model of $2 per conversation "sparked backlash for being unpredictable and expensive," got rebuilt around $0.10-per-action Flex Credits, and now hides inside the $550 bundle. Three pricing models in a year and a half is not iteration. It's a company that doesn't yet know what its AI is worth.

The customers, meanwhile, are describing a different product than the keynote. On G2, where Sales Cloud was literally renamed "Agentforce Sales (formerly Salesforce Sales Cloud)", the most-cited negatives across 25,825 reviews are "Learning Curve" (1,092 mentions), "Complexity" (682), and "Expensive" (612). A 2026 migration-practice writeup describes the end state precisely: "Instead of enabling teams, the CRM becomes something only a handful of specialists can manage." That is the admin-and-SI tax, and it's the exact wedge every lighter CRM swings at. Attio's flagship switcher story has Snackpass's COO saying the quiet part: "To achieve the detailed and automated set up we were looking for, we consistently had to involve engineering or third-party Salesforce consultants, which was expensive and inefficient."

And where reviews aren't gate-kept by an enterprise sales relationship, the score collapses entirely: on Trustpilot, Salesforce sits at 1.4 out of 5 across 619 reviews, one-star the dominant rating, with the AI-generated review summary opening on "outdated, incredibly basic, and not user-friendly."

My favorite detail is in the DNS. Salesforce's inbound mail runs behind Proofpoint, the enterprise email-security gateway that exists to keep exactly the kind of automated outbound its customers send out of exactly the kind of inboxes its customers target. Credit where due: the TXT records also show Pardot and Marketing Cloud verification keys, so unlike most vendors this size, Salesforce does eat its own cooking on marketing email. The stack reads like what it is: the world's biggest enterprise software company, buying enterprise software.

The break is structural. The support cut proves Benioff's AI works where work is a cost. The seller hiring proves it doesn't yet work where work is revenue. A motion that needs 34% of revenue in sales and marketing, forever, to hold single-digit growth is a motion that decays without constant human fuel. Attio's entire bet is that the fuel is the product now. Salesforce's own org chart says the bet has a point.

This is the section a tools roundup never writes, because it means reading earnings calls and a 10-K. Four disconnects, each one a quote against a number.

Benioff, Q3 FY26: "We now have over 9,500 paid Agentforce deals." By Q4 the headline number was 29,000 deals closed since launch.

Notice the switch: at Q3, only 9,500 of 18,500 total deals were paid. And divide the disclosed $800M of Agentforce ARR by 29,000 deals and you get roughly $27,600 of ARR per deal, pilot money, at a company whose median contract is $75K. The agents are in the building. They are not yet paying the rent.

FY26 revenue grew 12%, of which 4 points were Informatica, the $8B data company Salesforce bought in 2025. Strip it and organic growth is ~8%. Q1 FY27, same story: 13% headline, $444M from Informatica, ~9% organic.

Sharpest of all: the segment Salesforce now calls "Agentforce Apps," which is the renamed bundle of its core clouds, grew 7% in constant currency. That is the pre-AI baseline. The AI era core grows at the legacy rate.

Salesforce stopped disclosing net revenue retention, the single number that would show whether existing customers are expanding or quietly shedding seats, and replaced it with cRPO growth.

Then in Q1 FY27 it collapsed the product-level revenue disclosure, retiring the separate Sales Cloud, Service Cloud, Marketing Cloud and Platform lines investors had tracked for a decade, into two buckets, one of which is named after the AI product. If agentic pricing were expanding wallets at existing accounts, NRR would be the proudest slide in the deck. Its absence is a disclosure decision, and disclosure decisions are information.

Q1 FY27 Salesforce returned $27.5 billion to shareholders in a single quarter, against $6.7 billion of operating cash flow, funded by issuing $24.8 billion of debt.

Noncurrent debt went from $10.4B to $39.3B in ninety days. EPS grew 50%, much of it share-count math. Starboard added a ~$400M position calling for "more operational discipline," and the street holds a ~$260 target on a ~$170 stock that was the Dow's worst of 2025. A company certain that AI re-accelerates its growth invests like it. A company hedging buys back its own stock with leverage.

The fair read isn't "Salesforce is dying." $41.5B of revenue, $14.4B of free cash flow and the deepest enterprise relationships in software don't die; they consolidate. The fair read is that the AI transformation is real as a cost story and unproven as a growth story, and management's own metric choices suggest they know which.

Attio is winning its segment, which is exactly why the useful analysis is where it's exposed. Three cracks, and they rhyme with each other.

"From zero to IPO." Its customer page shows Granola, Modal, Flatfile, Railway, Snackpass, Passionfroot, and a venture fund. Those are excellent logos and every one of them is a VC-backed startup. The only large-organization migration on record, a 500-plus-seat Salesforce replacement at Jellysmack, lives on an implementation partner's website, not Attio's. The G2 head-to-head makes the gap measurable: on "Meets Requirements," Salesforce scores 8.8 to Attio's 8.2, and Attio's own reviewers name the holes: "It is not intended for outbound. And it doesn't have an integration with LinkedIn's inbox," reporting that one reviewer rates "shockingly bad," support that "has not always been that responsive." None of that matters at 5,000 startup customers. All of it matters at the first 5,000-seat deployment, and "4x ARR growth" off an undisclosed base (third-party estimates put it in the mid-teens of millions, unverified) leaves a very long road to the part of the market where Salesforce actually lives.

Read Attio's GTM job listings and you find an SDR Manager in San Francisco at $180-250K OTE, building a team the company itself describes as "nascent," an Account Executive req, a Sales Manager in London. This is the classic Series B playbook: take the GV money, build the SDR-to-AE ladder, go outbound. The self-aware touch is that the SDR Manager role requires three years of GTM engineering experience, and every GTM hire still closes with Sharp himself. But the tension is real: the company whose founder says "everything that came before has been invalidated by LLMs" is scaling pipeline with the most pre-LLM motion in software. Salesforce at least has the excuse of inertia. Attio is choosing this, now, with a blank sheet.

The anxiety is becoming part of the product experience.

Attio's own SPF record shows its marketing email routed through Mailchimp, while a G2 reviewer complains there's "no way to have the emails we send out from Mailchimp captured within the Attio CRM." The TXT records show Zapier verified, despite Attio shipping native automations. Small things, and exactly the kind of small things GTM engineers run dig to find. Meanwhile the pricing model carries a familiar seed: two credit currencies (seat credits and workspace credits), metered by tier, with paid top-ups. That's mild today. It's also the embryo of the billing anxiety that fills Salesforce's G2 page ("Expensive," 612 mentions) and that turned Agentforce's $2-per-conversation launch into a backlash. Consumption pricing resentment is the incumbent disease. Challengers are not immune; they're just early-stage.

Three moves, each chained to a finding above. If I had 30 minutes with Attio's GTM leadership, this goes on the table.

1. Build the migration proof surface, and put a seat count on it. Trigger: the only large Salesforce replacement on record (Jellysmack, 500+ seats) is published by a partner, not by Attio, while the company's own tagline promises "zero to IPO" and its G2 gap to Salesforce is "Meets Requirements," 8.2 vs 8.8. The move: a named "off Salesforce" program: migration tooling, a partner bench, and three case studies with seat counts, timelines, and what broke, published on attio.com. Why it moves the needle: the demand is already documented in Attio's own reviews ("We were previously using Salesforce, which came with high costs and a steeper learning curve"); what scale-up buyers lack is proof the jump is survivable. Requirements-gap objections die on evidence, not on features.

2. Make the new SDR team the most public AI-native outbound build in software. Trigger: the SDR org is "nascent" and being hired on the classic ladder, at the same company that ships an MCP server giving agents read/write on the CRM, and whose G2 critics say it's "not intended for outbound." The move: build outbound where agents run research, drafting, and logging through Attio's own MCP and humans only handle conversations, then publish the entire build, numbers included, as the flagship GTM Atlas series. Why: it converts a brand contradiction into the best demand-gen asset the company could own, and it answers the outbound objection with the company itself as the case study. Salesforce will spend $3.77B this quarter on the alternative approach.

3. Close the dogfooding gaps before a competitor writes the paragraph for you. Trigger: Mailchimp in Attio's own SPF record while a G2 reviewer asks for exactly that missing Mailchimp capture; Zapier verified against the domain despite native automations. The move: either move marketing sends onto Attio's own rails or ship the Mailchimp integration, and say so in the changelog either way. Why: Attio's revealed ICP is operators who audit tools for a living. For that buyer, "the AI CRM that compounds revenue around the clock" mailing its newsletter through a legacy ESP is a credibility leak, cheap to fix now and expensive the day a teardown like this one runs in someone else's newsletter.

Strip the brand names and this teardown is one finding: AI is currently better at deleting GTM cost than creating GTM demand, and the companies winning are the ones who structured their motion so demand creation doesn't need the humans in the first place. Salesforce proved the cost side with 4,000 support roles, then proved the demand side's limits by hiring 2,000 sellers and holding S&M at 34% of revenue. Attio proved the demand side can be structural (a product that onboards itself, content aimed at one specific operator, agents as a distribution surface), and is now testing whether that survives contact with Series B growth targets.

If you’re choosing what to copy:

Copy the architecture of the motion, not the headcount plan.

The cheapest pipeline is the kind your product and your content generate while you sleep, and the most expensive kind is the kind you have to re-buy every quarter.

One of these companies knows that and is drifting away from it. The other one's earnings calls are what drifting looks like, 20 years on.