Every quarter I get asked the same question by GTM leaders. Some version of: "Is sales tech actually still a thing? Or have we hit peak SDR-replacement-tool?" So I stopped guessing. I pulled the data on every Y Combinator-backed sales tech, CRM, and GTM company from S20 through F25, filtered for actual sales technology (not "we have an SDR who uses our healthcare product"), and built a real dataset. 139 companies. Five years. Every batch.

This is what I learned. Not the hot takes you've been seeing on LinkedIn. The actual patterns the data is pointing at.

And I'll tell you up front: two of my four going-in hypotheses were wrong. One was sharper than I realized. The fourth I'm still arguing with myself about. Let me walk you through all of it.

The Headline Numbers

Before we get into the meat, here's the basic shape of the YC sales tech universe over the last 14 batches:

The acceleration is obvious. The category essentially doubled between 2020 and 2023. ChatGPT launched in late 2022 and within twelve months we had a record year for sales tech launches. That's not a coincidence. That's an entire investor class deciding the GTM stack was about to be rebuilt.

But raw launches don't tell you what's working. Let's look at survival.

Now this is where it gets interesting. Pre-ChatGPT cohort (2018–2022): 18% inactive. Post-ChatGPT cohort (2023–2025): 6% inactive.

I know what you're thinking. The post-ChatGPT companies are younger. They haven't had time to die yet. Fair. That's a real caveat. But even adjusting for that, the data is telling us something specific: the wave of sales tech that's getting built today is being built on materially better primitives than the wave that came before it. Companies that launched in 2019–2022 promising "AI-powered" anything were promising something they couldn't actually deliver. They're now the inactive list.

Hypothesis 1: Sales Tech Is Cooling Off

Verdict — Wrong

I had this hypothesis going in because every other day someone tells me "AI SDRs are dead." And in narrow ways, that take has some merit (we'll get to it). But the overall sector? Not cooling. Accelerating.

Here's the part that doesn't fit the "AI SDRs ate the world" narrative: the median active YC sales tech company has 6 employees. The mean is 25 only because of a handful of breakouts like Apollo.io (800), Snapdocs (285), and CaptivateIQ (280) skewing it.

The actual shape of this category isn't a few massive AI agents replacing entire sales orgs. It's hundreds of 4-to-8-person teams trying to find a real wedge in a stack that's getting rebuilt. That's what an accelerating, healthy market looks like before consolidation. Not before death.

Hypothesis 2: Going Vertical Is Winning

Verdict — Sharper than I expected, with a twist

The vertical thesis is one of the few things in GTM tech that the data actually backs up cleanly. Let me show you the failure rates by category, because this chart changed how I think about what's defensible:

Three categories have a zero percent failure rate in my dataset: data and enrichment, RFP/proposals, and sales training & coaching. Vertical players are next at 12.5%. The horizontal "we're going to replace your whole stack" plays are the ones doing the dying.

If data is the new oil, the companies selling you "lists" of generic logos and undifferentiated TAM are offering barrels of sludge. The ones building structured, verified, vertical-aware data infrastructure — Cargo, Persana, Ciro, the LeadGenius core thesis — those are the picks-and-shovels businesses of this cycle. The data confirms it.

Here's the YC vertical roster, sorted by status:

Of 15 vertical-first companies, 13 are still active, one had a real exit (Bonfire), and only one is fully inactive (Caucus — and "AI CRM for government" was always going to be a long road, because the government moves at the speed of the government).

But here's the twist that doesn't show up in the static dataset and that I think matters more: founders are starting vertical and quietly pivoting horizontal. I see it constantly in the broader market. A founder launches "AI agent for construction sales." Six months later, their best customer turns out to be a marketing agency selling into construction. So they pivot. Now they're a horizontal LinkedIn automation tool competing with 200 other companies.

Why does this keep happening? Because founders consistently overestimate how ready a vertical is to adopt AI. The HVAC owner doesn't want an AI agent. He wants the phones to ring. Tech companies adopt AI in weeks. "Real" industries take years. And YC founders are wired for two-week iteration cycles, so when the vertical doesn't move fast enough, the temptation to chase tech-buyer revenue becomes overwhelming.

The vertical winners are the ones whose founders lived the industry pain before they wrote a line of code. And even those founders are mostly building boring software with AI features sprinkled on — not "agentic AI revolution." That's not a bug. That's the playbook.

Hypothesis 3: We're Heading Toward Fully Autonomous AI Agents

Verdict — Wrong, and the market has already told us so

This was the hypothesis I held most confidently going in. The trajectory seemed obvious: copilot → assistant → agent → autonomous worker. Every cohort, less human in the loop. By 2026 you'd just press a button and pipeline would happen.

The data — and especially what's happened to the loudest companies in this category over the last twelve months — is screaming the opposite.

Let's look at the YC AI SDR / outbound automation roster:

Twelve companies. Most of them tiny. Two of the largest (Warmly and Artisan) have quietly repositioned away from the "replace your SDR" pitch.

And outside YC, the loudest name in the category — 11x.ai, which raised $24M from Benchmark and $50M from a16z — became the cautionary tale of the entire AI SDR market this year. Investigations found 11x had been using "Contracted ARR" that counted three-month trial customers as if they were on annual plans. The company faced accusations of putting customer logos on its website without permission; ZoomInfo and Airtable publicly said they were not, in fact, customers. The CEO was reportedly fired amid the fallout.

And Artisan — the YC poster child for this category — raised about $47M total (including a $25M Series A in April 2025) and hit roughly $5M ARR with 250 customers. That's solid growth. But Artisan also reportedly delayed their broader "digital workers" roadmap, citing the need for "fundamental improvements" to their AI BDR. The pitch had outrun the product.

So why didn't this work? Here's the operator answer:

• Cold email is judged on outcome, not output. AI got really good at output (volume of personalized emails) almost overnight. It has not gotten meaningfully better at outcome (booked meetings with qualified buyers). The gap shows up in reply rates, in deliverability collapse, in prospects learning to spot the templates instantly.
• The buyer in B2B outbound is a defensive system, not a passive one. When the volume of AI-generated outreach hits a threshold, the inbox itself starts treating it as noise. We're past that threshold. By a lot.
• "Replace your SDR" is a hostile pitch to the person buying. The buyer is usually a VP of Sales whose entire org structure includes SDRs. You're asking them to fire their team. Most don't.

The companies actually scaling in this category have stopped pitching replacement and started pitching leverage: a tool for existing SDRs that automates the boring 80%, or a way to make outbound work for accounts that previously weren't worth a rep's time. That's a tool-for-humans pitch, not an autonomous-worker pitch.

Hypothesis 4: AI-Native CRMs Will Eat Salesforce

Verdict — Not in this decade. Maybe ever. But something more useful is happening.

The CRM category has 19 companies in my dataset. Roughly half showed up in the last three years. The pitch is some variant of: "Salesforce was built for managers who want reports. We're built for reps who hate data entry. And by the way, we have AI."

It's a great pitch. It also has a 21% failure rate, which is the second-highest in my data. Caucus, Xkit, Elucify, Sonnet — all gone.

The reason replacement is so hard isn't technological. It's organizational. When you replace a CRM, you're not replacing software. You're replacing years of workflows, integrations, sales ops rituals, manager dashboards, board reports, comp calcs, and institutional muscle memory. You don't rip that out for a prettier UI. You rip it out when there's a fundamentally different way of running revenue that the incumbent literally cannot accommodate.

So far, "AI-native" isn't that fundamentally different thing. It's a layer. A really useful layer. Just not a replacement.

The companies in this category that are actually winning are the ones that have stopped trying to replace Salesforce and started fixing the things Salesforce does badly:

The two oldest CRM companies on this list — Streak (2011) and Close (2013) — are also two of the healthiest. They survived Salesforce's heyday, the SaaS gold rush, the "death of CRM" thinkpieces of 2020, and they're still here. Both because they picked specific buyers (Gmail-native SMBs; SMB sales teams) and never tried to be everything to everyone.

Twenty is the open-source wildcard I'm watching. 44,800+ GitHub stars, modern UI, Y Combinator-backed. Whether that converts into a venture-scale business is genuinely uncertain — open-source CRM is a graveyard of well-loved projects that never made it to enterprise revenue. But if they convert even a fraction of self-hosters to a managed Pro tier, they have a path. And every $175/seat/month Salesforce charges is a margin Twenty can give back to its customers.

Where the Money Actually Is

Let me tell you what surprised me most when I sorted by failure rate. The categories getting all the attention — AI SDRs, AI-native CRM, "agentic everything" — are not the categories that are quietly succeeding.

The unglamorous categories are.

The Quiet Winners (0% Failure)

Sales training and coaching. Seven companies including Hyperbound, SilkChart, Trellus, Demodesk, Flockjay, Ambition. Plus Wingman, acquired by Clari. Zero outright failures (one acquisition counts as a win). Why this works: training is measurable, expensive to do manually, and scales by AI in a way that doesn't put the buyer on the wrong side of the table. When an AI roleplays a tough buyer, nobody is being deceived. When an AI reviews 200 calls and finds patterns, nobody is unhappy about it.

RFPs, proposals, and quoting. Eight companies. Two real acquisitions (Procoto, Bonfire — Bonfire had 95 employees at exit, real outcome). Active players include Inventive AI, Outlit, GovEagle, Powerdocs, GovernGPT, Veles. This is the most under-appreciated AI sales tech category in YC. RFPs are long, structured documents LLMs are genuinely good at. They're massive time sucks. Companies answer the same 200 questions across every RFP they touch. Existing solutions are 2014-era SharePoint. You don't need autonomy. You don't need agents. You need an LLM that knows your last 1,000 answers and drafts the first 80%. High willingness to pay. Drowning buyer. Solvable problem.

Data and enrichment. Zero failures. Period. The companies in this category — Cargo, Persana AI, Ciro, Fiber AI, Meticulate, Ponyrun, Origami, Anglera — are the picks-and-shovels of the AI GTM gold rush. They feed every other tool. (And yes, Clay is the elephant in the room here. Clay is not a YC company, founded 2017, separately backed, now at ~$100M ARR and a $3.1B valuation. But Clay's existence is the dominant force in this category, and the YC players are mostly positioned against or underneath Clay's footprint.)

The Quiet Unicorns

The three YC sales tech outcomes that quietly became massive businesses share a pattern. Let's look at them:

What do they share?

1. All founded before the AI hype cycle. 2012–2017. They had time to build deep workflow integration before the market got loud.
2. All sell a clear, measurable outcome. Faster mortgage closings. Accurate commission payouts. More leads in your funnel. Not "AI-powered" anything.
3. All are picks-and-shovels companies that added AI later, not AI-first companies looking for a workflow to attach to.
4. All survived at least one category transition. Apollo went from "ZoomInfo cheaper" → "GTM platform" → "AI-powered GTM." CaptivateIQ went from "Excel replacement" → "RevOps platform." Products kept changing. Customer relationships stayed.

If you're starting something in sales tech today, the lesson the data is screaming at you is this: AI isn't the moat. The workflow ownership, the data, the integration depth, the customer relationship — that's the moat. AI is the price of admission now, not the differentiator.

The Failures, In Detail

Twenty-three inactive companies. When I look at them as a group, the pattern that jumps out isn't bad ideas. It's bad timing, or bad wedges, or both.

If you sort these failures into buckets, you get:

• Pre-LLM AI plays (Flike, CoffeeAI, Buzzle, Echo, Sonnet, Criya) — built something that needed LLMs to actually work, launched 1–4 years too early
• Horizontal AI sales reps without a wedge (Penguin AI, Unhaze, Slik, Flyflow) — 2-to-4-person teams trying to take down Apollo
• Salesforce integration plays that Salesforce absorbed (Xkit, Elucify, Laserfocus) — moat evaporated when the incumbent shipped a native version
• Right idea, wrong buyer or wrong budget (PhoneSys, Compgun, Caucus, Protego, CoinTent)

Notice what's not on this list: nobody died because the technology didn't work. Everyone died because the market structure didn't.

The F25 Bets — What's Being Built Right Now

The newest companies in my dataset are from F25. The themes are revealing:

Three out of five F25 bets are entering already-crowded categories. That's a saturation signal, not a death signal — when a category attracts new entrants at this pace, it usually means the market is real, the unit economics work, or there's a perceived window before incumbents lock things down. Often all three.

If I had to bet on one from this cohort, it wouldn't be the loudest. It would be Aside. Sales engineers have completely different workflows from AEs. Tools built for AEs don't work for them. The buyer has budget and influence over deals. Small TAM, but the people inside that TAM care a lot.

What This Means for Operators

If you're a GTM leader trying to make sense of the noise — and that's most of you — here's what the data actually supports:

The Boring Truth

Most sales tech that wins is unglamorous. It compounds slowly. It's built by founders who would rather own a workflow than write a manifesto. It looks more like CaptivateIQ doing commissions than 11x trying to be your whole sales team.

The AI is the cherry. The work is the cake.

I'll keep tracking this. If you're building in this space and have a different read on any of it — or you want to argue about whether the 11x collapse is the canary or the floor — find me on LinkedIn. I genuinely want to hear it. The dataset is only useful if the people in the trenches are pressure-testing what I'm seeing.

— Derek Rahn · VP of Demand Gen, LeadGenius