Amjad Masad, founder and CEO of Replit, posted the other day:
“I thought I hated sales culture. By the end of this year, more than half my company will be salespeople.”
He goes on: for most of the last decade he was a typical tech guy. Build a good product, put it on the internet, let it speak for itself. For years the entire marketing department was his Twitter account. Sales was what companies did when the product couldn’t carry itself.
Then the demand showed up. Companies started emailing them saying their employees were already using it, how do we buy it properly. On Replit’s side, one guy was handling those conversations on top of three other jobs. Four reps, and a market that wanted more than four reps could carry.
What changed his mind wasn’t a book or a mentor. It was watching the dinners work.
I wrote a version of this five years ago called “Eventually … Almost Everyone Has a Sales Team.” The list then was Twilio, Slack, Box, Stripe, Monday.com, Calendly, Canva. The pattern hasn’t changed. What changed is the number on the odometer when it happens, and how much of the P&L it eventually takes.
Seven things to take away:
- The trigger is never “the product stopped working.” It is almost always the opposite: demand arriving faster than a self-serve motion can absorb it.
- The first sales team never announces itself. It arrives as account management, or support, or one person doing it on top of three other jobs.
- The threshold moved. In the 2015-2022 era, PLG companies added real sales somewhere around $30M to $50M ARR. In the AI era it’s landing closer to $100M to $250M.
- Sales and marketing ends up at 23% to 44% of revenue at the public leaders. It is the largest line item in B2B software, larger than R&D at most of them.
- The company most famous for not having a sales team is the one growing its sales spend fastest right now.
- The AI-native leaders did not skip sales. Anthropic added 140+ salespeople in 18 months and now posts more sales roles than research roles.
- Waiting has a real cost, and the bill arrives all at once.
#1. The threshold ladder, then and now
“Everyone eventually sells” isn’t the useful part. The number where it stopped being optional is.
The pre-AI cohort:
- Monday.com stayed mostly self-service to roughly $30M-$40M ARR. After that, sales and success drove both faster growth and much higher NRR on the way to $240M and beyond.
- Slack was at roughly $30M ARR in February 2015 with no sales team at all, and staffed a serious one crossing $50M ARR.
- Calendly added its first real enterprise team around $50M ARR.
- Box now does roughly 90% of revenue through the sales team, from 0-1% as a pure freemium product.
- Stripe was self-serve and very SMB for years before building a full traditional enterprise team.
- Canva held out the longest, close to $1B ARR.
The AI cohort:
- Gamma reached $100M ARR with no sales team, and its CEO now calls that a mistake.
- Replit went from roughly $2.8M to $150M in annualized revenue in under a year, then raised $400M at a $9B valuation in March 2026 with the stated plan of international expansion and building out go-to-market. Now more than half the company will be salespeople.
So the threshold roughly tripled. The wall moved from $30M-$50M to somewhere north of $100M, and it is still a wall.
That’s good news and a trap at the same time. The demand you’re absorbing when you finally staff up has been piling up for a year.
2. The first sales team always shows up wearing a different title
In February 2015, at the first SaaStr Annual, I got to run a compare-and-contrast session with Stewart Butterfield of Slack and David Sacks, fresh off Yammer. Slack was around $30M ARR and hadn’t hired a single sales rep. The full session is here.
I asked Stewart the question everyone was asking: can you get to $100M without a sales team?
His answer is the part worth re-reading eleven years later. He said nobody at Slack had a title that said sales. They had account managers, and there was no outbound. Those account managers talked almost exclusively to people who had already decided to buy but worked at companies where purchasing isn’t something you can just do, where there’s a vendor review process and a security policy analyst and in-house counsel who wants to mark up the terms of service. He described the job as “midwifing the sale,” as opposed to being a salesperson. No commissions. He believed Slack could go without commissions and without outbound forever.
Now read Amjad’s post again. Companies emailing to ask how to buy it properly. One guy handling those conversations on top of three other jobs.
That’s the same job. Slack in 2015 and Replit in 2026 hit the identical wall: enterprise procurement showing up on the back of bottom-up adoption. The only difference is that Slack called it account management and Replit is calling it sales.
Which is the first practical lesson. The first sales team never announces itself. It arrives as account management, or support, or customer success, or the founder’s inbox. If you’re waiting for a moment where you decide to start selling, you’ll miss it, because by then several people are already doing it part-time and badly.
Sacks’ side of that conversation is the second lesson. Yammer figured out roughly six to nine months in that it needed sales. Virality was a tremendous lead generation engine, and they never spent a dime on marketing, but it still took a salesperson to get deals over the top. Then, once the sales team existed, they reverse-engineered every other classic function of an enterprise software company: marketing to enable sales, customer success to drive renewals. A whole org built out of the decision to hire the first rep.
Two companies, the same shape of business, and a gap of nearly a decade between when each concluded it needed sellers. Yammer sold all-or-nothing company-wide networks, so somebody had to make an enterprise-wide decision, which pushed it into enterprise IT immediately. Slack sold to teams, so it could ride bottom-up adoption much further. Adobe ended up with fourteen separate paid Slack instances. The problem that eventually created isn’t acquisition, it’s consolidation, and consolidation is a conversation with a human.
That’s the real variable, and it’s the one to apply to your own company. Not whether you’ll need sales, but how far the unit of adoption lets you go before someone has to negotiate a company-wide contract.
Slack did go on to sell very hard. By $1B ARR it was a sales-led and product-led company, with the majority of revenue in enterprise deals. Stewart wasn’t wrong in 2015. He was right about the phase Slack was in, and that phase ended.
3. The AI cohort already made this call, on our own stage
Two SaaStr AI 2026 sessions answer this question from opposite directions.
Grant Lee, co-founder and CEO of Gamma, got to $100M ARR with 50 employees, 50 million users and 600,000 paying subscribers, most of it with no sales team. Then he said this on stage:
“We’ve always for better or worse been sort of reacting… I would advise maybe not do that.”
The specific version of that: they hired sales only when the inbound got embarrassing. Waiting was not the strategy. It was the absence of one.
Eleanor Dorfman, who runs commercial and industries sales at Anthropic, had the opposite problem. Her framing was that even if they had been ready to 3x or 4x or 5x the sales team, you cannot absorb that many bodies fast enough to deliver a positive customer experience. So in January 2026 they rebuilt the sales org around AI. Four months later, 54% of new enterprise logos came through the self-serve funnel. Real ACV, real terms of service, real invoicing.
That sounds like the anti-sales-team story until you look at what else Anthropic did. Kelly Loftus, who runs startup sales there, told an earlier SaaStr audience that she scaled that team from under 10 people to over 150 while the company went from 250 to 1,300 employees in 18 months. Her line on how they run it:
“We still don’t really have quotas. We have shadow targets.”
So the company most associated with self-serve enterprise AI added more than 140 salespeople in 18 months, and by late May 2026 had more open sales roles posted than open roles in AI research and engineering. More than 1,000 businesses now spend over $1M a year with Anthropic, roughly double the count two months earlier. A million-dollar contract does not close in a checkout flow.
Lovable is the same story earlier in the curve. It hit $400M ARR in February 2026 with 146 full-time employees, per CRO Ryan Meadows. Meadows himself is the tell: they recruited him out of Klaviyo in October 2025, at roughly $200M ARR, specifically to build the revenue engine. Elena Verna told the SaaStr AI stage that even approaching half a billion in revenue they were still on the product-market-fit treadmill. They have since opened GTM offices in San Francisco and Boston and posted roles for a Head of Sales North America and a Head of Channels and Partnerships.
None of these companies concluded they did not need sales. They decided when.
4. What it actually costs at scale
Here is the most recent reported quarter for four public leaders, straight from the filings. GAAP sales and marketing as a percentage of revenue:
The four split cleanly into two pairs, and the split is not about how good anyone is at selling.
- HubSpot and Salesforce sell seats to a business buyer. Growth comes from signing more customers and more seats inside them, and every one of those is a decision somebody has to be talked into. So sales and marketing runs 34% to 44% of revenue and costs roughly twice what R&D does. At HubSpot, 44 cents of every revenue dollar goes to acquisition and 25 cents goes to building the product.
- Datadog and Atlassian invert that, at least partially. Both are consumption or bottom-up land-and-expand businesses: a team adopts, usage grows, the bill grows with it, and a lot of next year’s revenue arrives without anyone selling it. Their S&M lines run 23% to 28%, and R&D is the bigger number at both, 42.6% at Datadog and 49.7% at Atlassian. Atlassian spends more than twice as much building as selling.
Atlassian is worth a beat of its own here, because it is the canonical “we don’t have salespeople” company and that was always partly a story about form rather than substance. For years after IPO they had almost no direct sales team but a very large channel of resellers selling into the enterprise, plus significant internal resources supporting that channel. Third-party selling instead of first-party. Our deep dive with former CRO Cameron Deatsch covers how that actually worked.
Two conclusions from that, and the second one matters more.
First, the pricing unit sets the sales-cost floor. If you charge per seat and the buyer is a manager with a budget, you are going to spend a third or more of revenue on go-to-market, and no amount of AI leverage changes the shape of that. If usage expands on its own, you get to spend it on product instead. That is a business-model outcome, not a discipline outcome.
Second, and this is the part that should stop the “we will never need sales” crowd: the low-S&M half of the table is where the spending is going up. HubSpot, Salesforce and Datadog are all trending down year over year. Atlassian, the lowest of the four and the most bottom-up of the four, is the only one trending up, and it grew go-to-market spend 35.8% against revenue growth of 26.0%. Twenty-five years in, at $6.6B in subscription ARR, in roughly 85% of the Fortune 500, and go-to-market is still growing nine points faster than revenue. Nobody at Atlassian decided sales culture was cool. They ran the math on unconverted demand sitting inside accounts they had already won.
The consumption model buys you a lower ceiling on sales spend. It does not exempt you from having a sales team, and past a certain point it starts pulling you back toward the middle.
The private side says the same thing. SaaS Capital’s 15th annual survey, completed in March 2026 across more than 1,000 private B2B companies, puts median spend at 15% of ARR on selling costs (up from 13% the prior year), 8% on marketing (unchanged), and 9% on support and customer success (up from 8%). R&D is 22%. Selling is the line that moved.
Equity-backed companies spend 70% more on sales and 100% more on marketing than bootstrapped ones at the same revenue. If you took the money, you bought the obligation to build the team.
5. On the “half the company” number
Worth being precise here, because Amjad’s number is striking and it isn’t the benchmark.
By headcount, go-to-market at most software companies is nowhere near half. Pave’s analysis of 2,914 companies with more than 50 employees puts the median at 10.6% of headcount in sales and 4.2% in marketing at 51-100 employees, moving to 15.4% and 2.6% at 3,001+. And their “sales” bucket is generous: it includes account executives, SDRs, account managers, customer success, sales engineering, sales ops, and enablement. Marketing shrinks as a share of the org as you scale. Sales grows. But the combined number lands closer to 18% than 50%.
What is close to half is the money, not the people.
So Replit at “more than half salespeople” is a deliberate, aggressive choice by a company staffing up fast off a small base, not the industry norm. That doesn’t make Amjad wrong. It makes him early, and expensive, on purpose.
6. AI is changing the shape of the team, not the size
The “AI replaces sales” argument has real teeth here, and it is narrower than people claim.
Emergence Capital’s survey of 560+ venture-backed B2B software companies found SDR and BDR teams took the sharpest hit of any sales function: 36% of companies cut SDR headcount over the prior 12 months, the highest of any role, and only 19% increased it.
The rest of the same data cuts the other way. Only 14% cut sales engineers. 28% grew account executive headcount. 34% grew professional services.
The cheapest, most automatable layer is getting compressed while the expensive, technical, relationship-carrying layers expand. The org isn’t shrinking. Prospecting and first touch are getting absorbed by software. Running a complex deal, passing a security review, designing an implementation, and saving a renewal are not.
Vercel put a number on that shift at SaaStr AI 2026. COO Jeanne DeWitt Grosser said their lead qualification agent took a 10-person function down to about one and a quarter people, running at roughly $5,000 a year in infrastructure and tokens, and that SDR quotas went up 30% that quarter. The 10 people moved into higher-value roles. The agent took the deterministic part of the job. Her larger read on where this leaves the function: go-to-market is moving closer to consulting than to selling.
Fewer, better sellers, as I’ve written before. Fewer is not zero, and better is not cheaper.
7. What “later” actually costs
The first hire gets harder, not easier. Recruiting a first real sales leader into a $200M ARR company with no sales culture, no comp philosophy, no forecast discipline, and a founder on record saying he hated sales culture is a much harder sell than recruiting one into a $5M company where the job is obviously to build from zero. The good ones will ask what happens the first time the forecast is wrong. You need an answer.
You compound the mess while you wait. Everything the sales org will eventually need (clean account data, a source of truth for who owns what, usage signal wired into the CRM, a real definition of an account versus a workspace) is cheaper to build at 100 customers than at 100,000. Companies that defer the sales build usually defer the sales infrastructure with it, then get to do both under time pressure.
You hand competitors accounts that had already picked you. Amjad names this one directly: the thought of handing a deal to a competitor is the worst feeling he knows. Every quarter of understaffed inbound is a quarter where your highest-intent cohort gets worse coverage than a competitor with reps who call back.
8. What Amjad got right that most technical founders get wrong
Two things worth copying.
The first: “What changed my mind wasn’t a book or a mentor. It was watching the dinners work.” Founders don’t get converted to sales by argument. Every technical founder who has been told to hire a sales team has heard the pitch and discounted it. What converts them is watching one deal get won that they know would otherwise have been lost. If you’re trying to get a skeptical founder or board to fund a sales build, stop making the case and go win one deal in a way that’s visibly attributable.
The second: “I’ll call whoever needs calling. I’ll get on a plane and because of that, we rarely lose.” Founder-led selling doesn’t scale, and it’s also the only reliable way to learn what does. The founders who build good sales orgs are almost always the ones who sold personally first, because they know what a real objection sounds like. The ones who build bad sales orgs hire a VP to run a motion they’ve never run and can’t evaluate.
The contrast he draws is the honest one. Consumer growth is weather. You ship, you test, two weeks later you might learn nothing. Sales is a contact sport where effort goes in and the outcome is mostly yours to decide. For a certain kind of founder, that trade is a relief once they see it.
How To Know When Your Turn Comes
- Plan for the sales org before you need it, and expect it to arrive later than it used to. If PLG is working, the wall is at a much bigger number than it was in 2019. That’s the good news and the trap.
- Watch for the specific signal, which is not a growth slowdown. It’s inbound asking how to buy properly, procurement emails, security questionnaires, and one person handling all of it on top of three other jobs.
- Build the data and systems layer early. The sales org gets built on top of whatever exists when you finally staff it.
- Budget for the mix shifting: fewer SDRs, more sales engineers and post-sale. Higher cost per head, not lower.
- Sell personally before you hire someone to do it. You cannot evaluate a VP Sales candidate on a motion you have never run.
- Expect sales and marketing to settle somewhere between a quarter and 45% of revenue. That isn’t a sign the product failed. It’s what it looks like when a product wins a market big enough to be worth covering.
