At a YC dinner last week, a founder casually mentioned they’re not hiring sales anymore.
“So.. how do you do outbound?”
“14 AI agents.”
I thought they were messing with me. Then they showed me the dashboard. Agents handling lead research, outreach drafts, support feedback, onboarding. Multi-million dollar pipeline running through it. And the founders still closing every deal themselves.
That’s what stuck with me. Not the 14 agents. The fact that humans were still doing the parts that actually matter.
What I’ve Seen Agents Handle Well
There’s a term sales folks use: “non-revenue driving activities.” Filling the CRM. Looking up everyone who signed up and finding relevant context. Prepping for calls. Monitoring social. SEO research. Drafting articles.
The boring stuff. The stuff that eats your morning before you’ve talked to a single customer.
I think of it like the CEO-to-EA relationship. The EA won’t do the job of the CEO, but they’ll remove every part of the job that gets in the way. That’s what agents are doing now. Sales, support, marketing, basically any department.
The YC company I saw wasn’t replacing salespeople. They were replacing the hours of low-impact work that kept their founders away from actual selling. The agents prep. The humans close.
Where It Breaks
The danger is over-automation.
Would you task an AI agent to close an enterprise deal worth $1M/yr? No. But can it prep for your calls, research the prospect, fill your CRM, and draft a follow-up? Absolutely.
The failure modes I keep seeing:
Bad targeting. Agents are only as good as the inputs. Point them at a broad, unfocused market and you get spray-and-pray at scale. Targeted outreach wins every time, whether a human or an agent is doing the sending. You still need to break the market into tiny cohorts — by pain point, tools they use, or role. More context in, more personal outreach out.
Automating before you understand what works. This is the one I see most. A founder reads about AI agents, spins up a full automation stack on day one, and wonders why nothing converts. You have to send the emails yourself first. See what gets replies. Watch the metrics closely. You can’t automate judgment you haven’t built yet.
No humans in the loop. There’s a lot of hype around fully autonomous pipelines. But every good build I’ve seen has guardrails. People reviewing before anything goes out. Founders still making the high-stakes calls. You can’t replace salespeople outright. That’s not how this works.
People in the Loop, On Purpose
The YC company’s build wasn’t fully automated. On purpose. The goal isn’t to remove humans from the process — it’s to remove humans from the parts that don’t need them.
The build is more intensive than you’d think. Most of the good ones I’ve seen are internal, probably because they need so much customization to get dialed in. A generalized solution is harder to crack.
But if you can get it working? No brainer.
The Progression: Manual, Then Semi-Automated, Then Agent-First
Automation is great but not on day one. If I had no money, no revenue, no customers, I’d build a tight “Dream 10” list of best-fit customers. Names, titles, emails, LinkedIn, any useful intel. Then I’d do the outreach myself. Track the metrics closely. I aim for 20-40% open rate, around 5% reply rate, 1-5% demo rate. Don’t scale before you’ve nailed it.
Once the pitch is dialed in and you know what converts? Put it on autopilot. Hire someone to keep it rolling. Start chipping away at the non-revenue-driving work with agents.
Then — and only then — you get to “14 agents.” You earn the right to automate by doing the reps first.
This is the part that doesn’t fit into a tweet thread. The progression is sequential, not a choice. Manual teaches you what works. Semi-automated helps you scale it. Agent-first removes you from the parts that don’t need you.
Most founders who give up haven’t done enough reps for the math to kick in. 20 cold emails isn’t enough data to know anything. You need volume before automation has something real to work with.
The New Pattern
YC advice used to be “don’t hire until it hurts.”
The new pattern I’m seeing: don’t hire, automate. Automate first. Strip out the low-value work. Then when you do hire, the hire is productive from day one because the tedious stuff is already handled.
Cold outbound is still the most important skill — for attracting talent, warm investor intros, first customers, influencer partnerships. That hasn’t changed. What’s changed is how much of the surrounding work agents can handle.
But you still have to learn the skill first. Do the reps. Feel what works before you hand it off.
The founders at that dinner didn’t start with 14 agents. They started the same way everyone does — manually, painfully, one cold email at a time. They automated what they’d already proven by hand.
That’s the whole insight. Earned automation. Not a shortcut — a reward for doing the work first.
Still working through kinks on my own build. If you’re building something similar, DM me.