Top 10 takeaways from The Agents #006

SaaStr published the back end of their go-to-market agent stack on episode 006 of The Agents: commit counts, API stacks, monthly costs, live demos. Three humans running more than 20 AI agents is the kind of claim that usually dissolves when you look at the receipts. This time the receipts are in the episode. We pulled the 10 numbers worth sitting with.

1. $257 a month replaced an entire BI workflow

10K, SaaStr's AI VP of Marketing, runs at $257 a month. That is roughly $3,084 a year, under 3% of one loaded marketing-analyst salary. It started in January as a dashboard connected to Marketo and Salesforce. Four months later it owns the number, runs daily forecasting, tracks every campaign in real time, and pushes the top three marketing ideas every morning.

Price every agent against a loaded salary, not against a free tier. $257 against $100K-plus in salary and overhead is a category difference, not a discount.

2. 402,000 conversations in one event cycle

Amelia AI, running on Qualified, fielded 402,000 chat interactions across 2.25 million sessions on annual.com for a single event. Three people cannot physically run 402,000 conversations. Even a strong BDR handles a few dozen meaningful conversations a day. At that volume, the efficiency framing misses the point: with three humans, 402,000 conversations isn't expensive, it's impossible.

3. 614 meetings, roughly $52M of theoretical pipeline

Amelia AI booked 614 qualified meetings for SaaStr Annual. At an ~$85K average sponsorship, that's roughly $52M of theoretical pipeline from one agent. Not all of it closed. But against headcount: a strong BDR books 10 to 15 qualified meetings a month, so 614 in a single event cycle is on the order of 3 to 5 BDR-years of booking output, compressed, with near-zero complaints on the booking experience.

4. 1,000 commits in 120 days, and lines of code told them nothing

10K has roughly 1,000 commits across four months, about 7 to 8 a day. Annie, their event site turned agent, went from 18,000 to 45,000 lines of code in two weeks -- yes, some of it is slop. The output didn't care. SaaStr is improving one internal application daily, not shipping to a million users on a fragile base. Commit velocity and code volume are irrelevant metrics when you're judging internal agent performance.

5. The slide that costs $1,400 per deal

SaaStr marks up and discounts in a controlled band. The problem is what happens at the close: a rep who smells a deal slipping moves from a planned 20% off to 25, then 30, then 34. The data says the extra discount doesn't move the close rate. On a $10K ticket, sliding from 20% to 34% off is $1,400 of pure margin per deal with no measurable lift. Across 100 deals, that's $140K gone.

A 20% discount off a marked-up price lands differently than a 20% discount off the base price -- and Amelia AI applies the right discount inside hard rules without negotiating against itself under pressure. SaaStr frames it as real-time CPQ. Guardrail the discount band and you stop subsidizing the anxiety of the close.

6. $500K recovered from leads humans never touched

A-leads don't need an agent. A million-dollar inbound gets a reply from your least attentive rep in 60 seconds. The value is in the B-leads: real signal, real score, but never worth a human's time, so they decay in the database. SaaStr pointed Artisan at exactly those and it returned an extra $500K against a cost in the low thousands a month.

Agents on the work humans structurally skip produce incremental revenue that didn't exist before. Agents on the deals humans already chase produce marginal uplift.

7. 150 accounts managed at under 90 days old, with zero Salesforce data

QB, SaaStr's AI VP of Customer Success, manages 150 accounts and it's under 90 days old with no Salesforce integration yet. The point isn't that it's feature-complete. SaaStr's pattern across agents is to ship into production early, measure output, and layer in integrations as the agent proves itself. They don't wait for the perfect data environment.

8. Three humans as the operating constraint, not the ceiling

The 3-human number is the framing device for everything else in the episode. SaaStr isn't using agents to reduce headcount from 30 to 3. They built the operation at 3 from the start, with agents covering the surface area that headcount would otherwise fill. That's a different architecture from "we replaced some people with AI." The agent stack is the org chart, not a supplement to it.

The question worth asking isn't how to use AI to do what you're already doing faster. It's what work you would simply never do at your current scale that an agent makes possible. The SaaStr numbers are the answer to that question.

The pattern recurs across other operators. The same architecture appears in our breakdown of AI agent stacks at SaaStr, Owner, and Klaviyo, and the Aurasell CEO episode maps the GTM stack implications directly.

9. Daily forecasting is what makes the agents accountable

SaaStr runs daily forecasting through 10K specifically because they need to know what each agent produced that morning, not quarterly. The agent only proves its value if someone is watching the output. Without that measurement layer, you can't tell whether the agent is performing or just running.

10. Output is the only metric that survives contact with production

Across every agent in the episode, SaaStr's evaluation framework is the same: what did it produce? Not how it was built, how many commits it took, how clean the code is, or whether the integrations are complete. The 402,000 conversations happened. The 614 meetings were booked. The $500K came in. Those numbers don't care about the architecture decisions that generated them.

“Deploy agents where the work exceeds human capacity. That's where the numbers get interesting.”

The agents that perform are the ones with clear output metrics, hard guardrails, and deployment in work humans structurally can't or won't do. The ones that underperform are the ones layered on top of work humans were already doing fine.