Running an eight-figure B2B business with 1.2 humans and 20 AI agents

Executive overview

Most sales playbooks are broken — not because the underlying plays stopped working, but because the people executing them are being replaced faster than companies realise. Jason Lemkin replaced SaaStr's entire go-to-market team (8–10 humans) with one full-time AE, a part-time chief AI officer, and 20 specialised AI agents. Net revenue is roughly the same.

The core insight: AI already matches the median human sales rep — and the median rep is the problem, not the solution.

The transition works not because AI is exceptional, but because most human SDRs and BDRs were already mediocre, slow, and expensive. Agents work nights, weekends, and holidays; they don't quit on-site at your flagship event.

Why SaaStr replaced its sales team

• Two senior sellers quit during the 10,000-person annual event; Lemkin decided on the spot to stop hiring humans for sales
• A general-purpose Delphi agent (not trained for sales) had already closed a $70K sponsorship autonomously — proof of concept
• Junior SDRs at $150K/year were routinely ignorant of the product after 90 days; agents trained on best scripts outperform them
• AI is displacing the mid-pack and mediocre; the best humans retain value and gain superpowers

The agent stack SaaStr actually runs

• Delphi (digital Jason) — general-purpose clone ingesting 12 years of content; handles support, FAQs, event logistics; evolved into a sales closer
• Artisan — outbound SDR; ~60,000 emails sent, strong open rates; chosen because they offered the most deployment help
• Qualified — inbound qualification chatbot on saastr.com; qualifies prospects 24/7 and books meetings automatically; closed a sponsorship at 11 p.m. on a Saturday
• AgentForce (Salesforce) — re-engagement of leads sales had abandoned; achieved a 70% response rate on lapsed contacts humans wouldn't touch
• Momentum / Attention — rev-ops AI that auto-logs every human sales action into the CRM in real time; surfaced one rep who had done nothing for 30 days
• One human AE closes deals, negotiates pricing, and manages procurement
• Amelia (chief AI officer) spends ~20% of her time orchestrating and QA-ing all 20 agents (~10–15 hours/week)

How to train an agent that actually works

• Take the best email or script your best human ever wrote — that becomes the agent's base template
• "Ingest" means upload your website URL, wiki, training docs, and a few key PDFs; the agent processes them
• "Training" means answering the questions the agent generates until it stops making mistakes
• Expect 30 days of daily corrections (1–2 hours/day); by day 30 it is reliably good
• Agents A/B test variants automatically — give Claude three rewrites of your best email and let it iterate
• Light personalisation from CRM data (Salesforce, HubSpot) lifts response rates further
• Poorly trained agents from bad vendors explain why most 2024 AI SDR pilots failed — not the technology

Picking a vendor

• All leading AI GTM tools run on similar underlying LLMs; feature parity matters less than deployment support
• The deciding column in any vendor evaluation: who is your forward-deployed engineer (FDE) and will they do the work with you?
• Artisan and Qualified won SaaStr's business because they showed up and trained the agents themselves
• One major vendor demanded $100K upfront before helping; another feared bad PR; both were ruled out
• Entry price for a properly supported agent: ~$50–75K/year (some start at $100K); no effective $99/month products exist yet
• The best vendors sometimes turn away business if they cannot guarantee success — a positive signal

The orchestrator role: chief AI officer

• Running 20 agents is not passive — someone must segment the database, review outputs daily, fix hallucinations, and prevent agents from conflicting with each other
• This person needs to be quantitatively minded, comfortable with data routing, and ideally a nerd who came from marketing, product, or rev-ops — not traditional sales
• Promote internally; do not post a "GTM engineer" job ad expecting an experienced hire — the veterans do not exist yet
• Agents work around the clock; the orchestrator's job is never done and is genuinely exhausting

What is changing in sales — and what is not

Mostly or fully automated within 12 months:

• Cadence-based email SDRs running drip campaigns — no reason a trained agent cannot do this better
• BDRs qualifying inbound "contact me" leads — a bad customer experience that AI eliminates
• Basic support (already 50–80% AI across the industry)

Changing but not gone:

• Account executives: ~70% of AE jobs safe near-term; will decline to 40–50% as agents prove they can close lower-complexity deals
• Rev-ops tracking and reporting: fully automated via tools like Momentum/Attention

Unchanged or growing:

• In-person and field sales: all data shows higher close rates; AI has no answer here yet
• Enterprise hunting of named, high-value accounts (e.g., 50 target logos on a whiteboard): human judgment and relationships dominate
• Leadership, orchestration, and VP-level roles: still require humans; autonomous executives do not exist

The market context

• Traditional B2B had 3–5% of prospects in market at any given time; AI-category products are seeing north of 50% in market simultaneously
• This bifurcates GTM: hyper-growth AI companies have more leads than they can touch; legacy SaaS companies face evaporated demand
• Both ends benefit from agents — one for ruthless efficiency, the other for servicing massive inbound
• Buyers are getting fatigued: enterprise teams that deployed five agents in 2025 may refuse a sixth in 2026; close deals now while the window is open

Advice by role

For junior salespeople:

• Become the first person on the team to embrace whatever agent your company deploys
• Work alongside the agent, not against it; extra meetings it books are opportunities, not annoyances
• Transparency is now total — AI rev-ops logs everything; do not fight it

For VPs and sales managers:

• Buy an agent and deploy it yourself — do not delegate to an agency or a junior hire
• Go through ingestion, training, and orchestration yourself; the jargon is not hard once you do it
• Pick one painful problem, pick a vendor that will help, and get it into production; the second agent is much easier

For founders:

• Build a team of FDEs whose only job is making the agent work on day one; do not sell and disappear
• Do not build GTM agents internally unless you have a world-class go-to-market engineering team champing at the bit
• Run the incognito test: sign up for your own product with a fresh email, go through support, try to contact sales — then fix the part that makes you cry first

The future of the sales profession

• Email-based SDRs and inbound BDRs: effectively extinct within 12 months
• Senior reps managing agents instead of managing people: target $250K/year compensating one person who runs 10 agents
• Net headcount in GTM will still grow because AI-native companies are scaling so fast they need more humans even as productivity per head rises 5–10x
• The risk is not mass layoffs; it is that departing humans are not backfilled — and that people who do not adapt become unhireable
• "If you can close it on a text message, AI can close it" — the bar for human-only sales is narrowing to high-ACV, high-complexity, in-person enterprise