The revenue laboratory.

Every revenue org that has ever existed was run on art. Not because anyone chose art. Because no human, and no org chart full of humans, could hold the full context of a revenue motion at once. So we compressed: averages instead of segments, playbooks instead of evidence, intuition instead of observation. I found out what that costs when the system told me our best idea was wrong.

I am a software engineer. I have never had a sales job. Our flagship campaign targeted mid-stage SaaS founders, the ICP everyone told us to want. Across 637 people contacted it produced a 0.78 percent reply rate and zero meetings, while another segment, on the same product, the same sender, the same system, was replying at 31 percent. The blended average read a respectable 9.8 and hid the whole story. So we killed the flagship campaign, rewrote the ICP, and published the numbers anyway.

The blended average was the art. The segment split was the science. AI changes the one constraint that made the art necessary: an agent stack can hold the full context of a motion, every account, every touch, every promise and the outcome that followed, and act on all of it at once. Held context is what makes a science possible. If you run revenue or fund the people who do, the question in front of you isn't whether to adopt AI. It's whether you keep operating on the average that hides your 0.78, or run the org that finally holds enough context to find it.

I know because I am running it.

We started where the breakage is most visible: BD. We built the agent stack and ran it on ourselves first. Vruum is literally a tenant inside Vruum: the same schema, the same dashboards, the same rules as every customer. Running our own outbound takes me about an hour a day; the system has booked 28 meetings across 1,610 people contacted, and we closed our first paying customer. On LinkedIn, where the motion runs: 42.7 percent of our cold connection requests get accepted, against published industry averages in the mid-20s, and 29.4 percent of our follow-up messages get a reply.

All-time pattern, end-to-end, selling Vruum with Vruum:

Then operators started asking me to run theirs, and teams started asking to run it themselves. So we sell it both ways:

• Software-shaped contribution margins when your team runs it; managed engagements priced from $30K a year so the margin survives real operator hours.
• Self-serve ladder from $300 a month; managed is custom-priced.

BD is where the model proves fastest because the metrics are the cleanest: reply rate, meetings, and conversion to client are all measurable inside 30 days. We currently run the full top-of-funnel motion for clients: research and sourcing, outbound on email and LinkedIn, and demand gen via LinkedIn content and engagement-based prospect warming. Sales execution, CS and lifecycle, and the rest of the marketing stack are next. The same operator-plus-agent loop applied across every revenue function.

Two things about those numbers matter more than the numbers. First, they're audited: mid-quarter we found our analytics crediting organic connections to outbound, so we fixed the meter, republished the corrected truth, and said so, because a system that shows you what's working is worth nothing if you can't trust the meter. Second, they're checkable by anyone we sell to. The diagnostic that killed our own flagship campaign at 0.78 percent is the same one that runs on every tenant. Product-led growth has been the only go-to-market motion that operates as a fractal system: the same loop at every scale, the company living inside its own machine. This is the second one. The other signal: I am a software engineer who has never had a sales job. The numbers above are what the system produces, not what a senior salesperson produces.

The reason one operator and an agent stack can run this at all is not that the operator is exceptional. It is that the revenue org is broken in three specific places, and one shared limit kept it from adapting to any of them.

Three forces, one limit.

Three forces converged at the same moment.

The first is buyer fatigue. Cold email worked when it was rare. It stopped working when it became universal.

Inbox volume per knowledge worker is up roughly a third since 2015, but the share that is cold or promotional has grown far faster. By 2024, Gmail was routing roughly three-quarters of commercial mail away from the primary inbox, up from roughly half a few years earlier. The platform itself is filtering out volume that buyers will not tolerate. On LinkedIn, acceptance on generic cold connection requests has slid into the mid-20s across the large published datasets (opens in new tab), and cold InMail reply rates sit in the 6 to 10 percent band, a fraction of what the channel produced when it was novel. A competent SDR in 2019 could clear roughly 8.5 percent reply rates on cold email sequences. The same sender in 2026 averages 3.43 percent.

Volume used to be a moat: send more, get through more. That logic inverted around 2022, when every other vendor started sending the same email and the buyer's filter learned to discard the whole category at once. Gartner's 2024 fieldwork found 73 percent of B2B buyers now actively avoid suppliers who send irrelevant outreach, which is the part most operators miss: buyers aren't just ignoring saturation, they're penalizing the senders who cause it.

The same dynamic shows up at every layer of the funnel. Buying committees grew from 5.4 stakeholders a decade ago to 6–10 on standard B2B deals and 17+ on the largest, and 74 percent of those committees are in active internal conflict. Paid attribution broke in 2021 when iOS 14's ATT cut the data tracking the funnel depended on. Generative search began eroding organic traffic in 2024. Every revenue function is reaching the same buyer, and the buyer has stopped responding to the way the org is built to reach them. This is not fixable with copy. The channels are saturated.

The second is the cost of the unit. The fully-loaded SDR numbers above produce a cost-per-qualified-meeting in the $3,000 to $5,000 range and a cost-per-opportunity in the $15,000 to $25,000 range for outbound-sourced pipeline at typical conversion rates. For B2B businesses with sub-$50K ACV, CAC is broken before any other line item is added. At higher ACV the ratio still functions, but only for companies that can absorb the loss on outbound and recoup it through expansion or retention. Most cannot.

The SDR is the most measurable cost crisis. The same shape is breaking the rest of the revenue org. AE quota attainment collapsed: Salesforce's 2024 State of Sales reported 84 percent of reps missing quota (opens in new tab), RepVue's 2025 data shows 57 percent. CSM books stretched to 49 mid-touch and 144 low-touch accounts, with expansion buried under retention firefighting. Marketing CAC keeps climbing as paid channels saturate and demand gen produces leads sales rejects at rates nobody publishes. Every revenue unit is now too expensive for what it produces.

The third is AI, and most operators are not internalizing it fast enough. The volume work in outbound (research, list building, drafting, sequencing, reply triage, meeting prep) is template work. So is the equivalent volume in sales execution, customer success, demand gen, and the marketing stack underneath. Template work is the first thing agents commoditize, which is why a solo operator with a working stack now produces the throughput that used to require a small team, at a fraction of the cost, with better personalization than any of them because the agent actually reads more sources per account than a human would bother to.

And here the industry learned exactly the wrong lesson. The first wave of “AI SDR” tools used the new capability to make guessing cheaper: more sends, more sequences, more volume into the same fatigued inboxes, with no better idea of what was working. The result is public: churn rates of 50 to 70 percent inside a year (opens in new tab), and a buyer population that now penalizes the whole category on sight. Cheaper guessing is still guessing.

The first two forces alone would have made the existing model unsustainable. AI made an alternative possible at the same moment, and that's what breaks the old shape: the work now has somewhere to go.

When all three converge, the model cannot be patched. Hiring harder doesn't work, because every new hire inherits the same broken denominators. Adding another tool doesn't work, because the tools are already running 12-to-20 deep. Outsourcing to a managed BDR shop doesn't work either, because the shop is hiring against those same denominators and charging you margin on top. Every move that preserves the old shape just delays the math. And every one of them shares the same limit. The revenue org is shaped the way it is because no human can hold the context of a whole revenue motion: thousands of accounts, every conversation, every promise made and every outcome that followed. So the org partitioned it. Silos are context split across departments. Playbooks are context compressed into rules. The funnel is context reduced to stages, and the average is what's left when the detail won't fit in anyone's head. That's why the loop never closed: closing it means holding what you promised in acquisition against what happened after, across every account at once, and nobody could. Revenue ran as an art because art is what you practice when you can't observe the whole system. More heads partition the context further. More tools record it without holding it. That's the shared limit, and it's the thing that just moved.

What replaces it.

An agent stack holds what no org chart could. It reads every account, remembers every touch, and keeps the promise made in acquisition connected to the outcome delivered after, across the whole book at once, with no update meeting and no CRM anyone forgot to fill in. And once the context is held, science stops being a metaphor. You can run revenue the way a lab runs an experiment. Every belief about your motion, this segment buys, this message lands, this channel converts, is a hypothesis, stated where the evidence can test it. Every touch is a trial. Every reply, meeting, and deal lands against the segment that produced it, so no average can hide a failing cohort behind a working one. What survives the evidence scales. What fails dies before it burns two more quarters, the way our flagship campaign died at 637 contacts instead of 6,000. Today the system supplies the segment-level evidence and the operator renders the verdict. The version where the system designs and assigns its own experiments is being built. We'll publish it when it ships, the same way we published the meter fix.

Art is what you practice when you can't hold the context. Science is what becomes possible when you can.

The instrument that runs this loop is one operator paired with an agent stack. The person holds judgment, strategy, segment intuition, the relationship moments, and the edge cases the models still get wrong. The agents underneath do the volume: continuous research across thousands of accounts, drafts tuned to segment, channel orchestration across LinkedIn and email and reply, daily evaluation of what's working, reply triage and qualification and routing. It's the same job description an SDR manager used to hold, except the junior team is software now, the lab notebook writes itself, and the operator runs more of it than any human manager could. BD is just the loop we built first. The same shape applies to AE workflows, customer success, lifecycle, and the marketing stack underneath.

Calling this a productivity story misses what's actually changing. The unit of purchase is the thing being redefined. Buyers used to pay for software or for headcount. Now what they pay for is the outcome (meetings, pipeline, retention, expansion), and the team that produces the outcome is the firm's problem, not the buyer's. The firm delivers the work and owns the agents and operators that do it.

Sequoia named the category in March 2026: Services as Software (opens in new tab), autopilots that sell the work instead of the tool, with the labor budget as the addressable market. Vruum's managed door is the autopilot version; the same loop is sold as software your own team runs, which is how our first paying customer chose to run it. The comparables and the margin math live in the appendix.

What it actually looks like.

The unit is one operator running an agent stack across one or more revenue functions. The operator does the work that breaks if you hand it to a model: judgment calls on edge cases, the relationship moments where someone needs to actually sound like a person, the positioning calls where taste decides the deal. The agents do everything else, at a volume and consistency the old human team could never match. The parts of revenue that stay art stay human. What stops being art is the memory.

In BD, this is the version Vruum runs in production today. Continuous research across thousands of accounts. Hundreds of qualified outbound touches per day per channel, against a senior-SDR ceiling of roughly 95 activities total in an eight-hour shift (35 calls, 33 emails, 15 voicemails, 7 social), with most of those low-quality at the margin. Draft messages personalized at the level a senior SDR produces on their best day, generated for every contact instead of the top ten. Reply triage and routing inside the hour. Automatic segmentation by signal. Daily evaluation of what is working with same-day iteration on the segments and messages that are not. The operator approves the drafts that need a human eye, routes booked meetings to the client (the client takes them, not the operator), and tunes the segments that get tested. One operator runs up to 10 clients, roughly one hour per day per client.

In demand gen, Vruum ships the LinkedIn layer in production today. Content drafted by the agents, tuned to segment ICP and the operator's voice, posted to the operator's profile on cadence. Daily warming of named prospects: agents draft reactions and comments on the prospects' own posts and queue them for the operator's approval. Inbound engagement flows back into the BD pipeline as warming signal. The operator approves what goes out and holds the brand voice. Paid, organic search, attribution against pipeline outcomes, and the rest of the marketing stack come next.

In sales execution, the same loop is on the roadmap. The AE today spends about 30 percent of the week actually selling. The other 70 percent is admin, internal meetings, manual data entry, and prospect research. The agent stack absorbs that 70 percent: meeting prep written against fresh account context, follow-up sequences that hold deal state across weeks, deal-room curation, mutual action plans drafted from the last three calls, executive briefing notes, competitive intel pulled the morning of the meeting. The client runs the calls and makes the judgment calls that close the deal.

In CS and lifecycle, the same loop is on the roadmap. An enterprise CSM holds 22 to 50 accounts, and most of the week is reactive: onboarding, ticket escalation, fire drills. Health is gut feel above 200 accounts. The agent stack watches usage signals, support patterns, sponsor changes, contract dates, and product release fit across every account every day. Proactive outreach triggers on usage drift. Expansion timing surfaces from product behavior. Renewal forecasts run on real data. EBRs are drafted against the last quarter of customer activity instead of stale slides. The client runs the conversations that matter.

The pattern is the same in every revenue function: close the loop, put an operator on judgment and agents on volume, and let every outcome write back into the system so the motion learns. The org chart you're hiring against today isn't there because growth requires it. It's there because the old technology required it, and the technology moved.

What this is not.

This is not an AI SDR. The AI SDR wave used agents to industrialize guessing: volume without hypotheses. And volume without hypotheses is the disease, not the cure. A lab that sends less and learns more beats a cannon that sends more and learns nothing, every quarter, compounding.

This is not the end of your sales team either. Our first paying customer is an established consultancy whose own multi-rep team runs Vruum daily. They kept the reins on purpose, and their daily outbound habit is exactly why it works. What ends is the guessing: the part of the org chart that existed to produce volume nobody could evaluate.

This is not the elimination of humans from revenue. The operator role is real, demanding, and not commoditizable. The work that remains is the judgment-heavy, context-loaded, increasingly senior part, and there's more of it per firm than there used to be, not less. The model concentrates human work, it doesn't erase it. Klarna learned the hard version of this in 2024 when it replaced roughly 700 support agents with an OpenAI-built assistant, watched CSAT collapse on the emotionally charged and multi-step cases, and publicly walked the strategy back. The goal isn't no humans. It's humans doing higher-leverage work, supported by agents that handle the volume the humans never wanted to do anyway.

This is not a magic AI button. The agent stack is a real engineering artifact: research pipelines, evaluation frameworks, channel orchestration, reply analysis, prompt strategies tuned over months. We have iterated weekly for months to get to the rates above. Anyone selling you “agents do outbound” without showing you the eval framework and the segment-level data is selling you a 2024 demo with 2026 marketing. The 11x.ai story is the cautionary version (opens in new tab): $74M raised from a16z and Benchmark, ARR conflating trials with contracts, customer logos used without permission, 70 to 80 percent churn inside months. Demo-grade pipelines do not survive contact with real GTM.

This is not comfortable for the people whose jobs sit inside the old org chart. There are roughly 1 million sales reps of services in the US, and 36 percent of B2B companies cut SDR or BDR roles in 2025. Salesforce has cut 4,000 support roles citing Agentforce. Block cut 40 percent of its workforce in February 2026 (opens in new tab), with Jack Dorsey predicting most companies follow within a year. The new model creates higher-leverage operator roles, but they are senior, they are fewer, and the transition is a real labor question for the industry. We are not pretending otherwise.

The choice in front of you.

If you run revenue today, somewhere in your funnel there's a 0.78 hiding inside a 9.8: a segment, a message, a channel that has already failed, still consuming budget because the average looks fine. Nothing in the old game will find it. More heads inherit the same broken denominators, more tools add rows to a dashboard that can't say what worked, and the next quarter of guessing costs the same as this one. Meanwhile the closed-loop motion compounds, because every touch it spends buys evidence as well as pipeline.

And revenue is where this starts, not where it ends. The revenue org was never really a separate thing. It's the business, seen from the market's side. The aim of a business is profit and impact, and both run on the same context nobody could hold: what was promised, what was delivered, what it cost, and what came back. Revenue is just the first function measurable enough to run as a science. The same ceiling that lifted here lifts everywhere next.

The new game is the loop: state the hypothesis, run the test, read the verdict, keep what survives. One operator and an agent stack to run it, yours on our software or ours on your behalf, with every number on a meter you can audit. The guessing is what ends.