The story behind Tantra.

A spreadsheet, a webinar, and a question that wouldn't go away.

I built my first revenue funnel by accident. A client needed leads in two weeks; we wired up a webinar, a follow-up sequence, and a seven-stage pipeline, and the thing worked. Then the second one worked. Then the third. By the tenth, I noticed something uncomfortable: every funnel I built was roughly the same shape. Different ad creative, different vertical, different tone of voice — but the math was identical, and the operational mistakes were identical, and the bottlenecks always landed in the same three places.

That was 2017. Over the next nine years, working under the GrowthClub banner, my team and I shipped funnels for B2B SaaS companies, marketplaces, ecommerce platforms, agencies, and a few category-creating startups in IoT and fintech. The work generated $500M+ in client revenue. Most of that revenue came from the same five mechanical things, repeated differently for each customer: a clean offer, a webinar or reverse-squeeze flow, a seven-step nurture, an SDR cadence with real personalization, and a CRM where nothing got dropped on the floor.

The question that wouldn't go away was: why am I a person doing this? Almost every action I took was something a sufficiently smart system could do faster, cheaper, and more consistently than I could on my best day. The bottleneck wasn't insight. The bottleneck was attention — somebody, somewhere, had to remember to send the third follow-up, to A/B test the headline, to flag the deal that had gone quiet for fourteen days. The attention was scarce; the work itself was not.

Revenue is the only metric that matters. Everything else is in service of revenue.

I picked revenue operations as the wedge for one stubborn reason: it's the part of a company where bad work is most expensive and good work compounds the fastest. A 2% lift in lead-to-call conversion isn't a 2% better quarter — it's, conservatively, a twenty-times-the-engineering-cost return over a year. And yet RevOps is, for most companies, a hand-stitched patchwork of HubSpot, a few Zapier flows, a Google Sheet someone is afraid to touch, and a lot of unspoken tribal knowledge living in one operator's head.

That's the gap. Founders pay tens of thousands of dollars a month for tools that don't talk to each other and headcount whose entire job is making them talk. Then those people leave, the knowledge walks out the door, and the next operator rebuilds it from scratch. It is the most expensive Groundhog Day in software.

Tantra exists to end that loop. The bet is that an autonomous system — eight agents on top of a single eleven-stage pipeline, governed by a manifesto and a budget — can replace the patchwork outright, not augment it. Not "AI-assisted HubSpot." A different shape of thing. The operator becomes a director: they brief the agents, they approve campaigns, they review the books on Friday. The hand-stitching is gone.

A new kind of operator.

Through 2023 and 2024 I tried every flavor of "AI for marketing" on the market. Most of it was lipstick: a chatbot in the corner of a dashboard, a button that wrote a worse subject line than I would. None of it was a real operator. None of it could be left alone for a weekend without burning down the pipeline.

What changed in late 2025 was tool-calling. Multi-turn, multi-tool agents that can actually do work — read a CRM, write a proposal, run an A/B test, escalate a stuck deal — became real and cheap enough to run for hours per task. The economics flipped: the marginal cost of a competent autonomous worker dropped below the marginal cost of a junior human doing the same thing, by roughly two orders of magnitude. That's not a feature. That's a different industry.

Tantra is what I built when I stopped waiting for someone else to build it. The platform's primary inference runs through Krutrim Cloud (an Indian provider, paid in INR, priced for ruthless cost discipline) with OpenRouter free models as a fallback. We measure spend per agent, per task, per turn. The average end-to-end cost of a Tantra-run campaign is a small multiple of the API spend — the rest of a normal RevOps budget (the headcount, the seat licenses, the integration glue) is simply gone.

Five claims we'd defend in a room of skeptics.

One. Revenue is the only metric that matters. Everything else — engagement, brand awareness, NPS, MQLs — is either a leading indicator of revenue or it's noise. We optimize against the funnel math and let the rest fall where it falls.

Two. Ship at 80% quality. The asymptotic 20% of polish is the most expensive 20% to deliver and almost never decides the deal. Every Tantra agent is biased toward shipping and instrumenting; we let the data tell us what's worth fixing.

Three. The funnel has math, and the math is knowable. There is no mystery. Every conversion rate has a distribution, every drop-off has a cause, and every cause has a fix that has worked somewhere else. The job of the system is to find the fix and apply it without waiting for permission.

Four. Autonomy must be governed. Eight agents running unsupervised would destroy a brand in a week. The manifesto, the budget envelope, the eleven-stage pipeline, the quality-control reviewer — these aren't decoration; they are how the autonomy doesn't eat the founder's reputation.

Five. Cost discipline is the moat. Anyone can spend $20K a month on GPT-4 and look smart. The interesting engineering is doing it for $200 a month and being right more often. We route every task to the cheapest model that will actually do the job, and the savings compound for the customer.

Toward a fully agent-run revenue stack.

The current Tantra build is V1: chat, funnel, CRM, and the agent pool, on a single eleven-stage pipeline, with eighteen scheduled jobs running the Karpathy auto-research loop in the background (observe KPI → hypothesize → experiment → measure → learn → repeat). It's enough to run a real RevOps stack for a real company today. We're using it to run our own.

V2 is about replacing the last manual surfaces — paid ad bidding, content programming, voice-channel follow-up, deeper integrations into Meta, LinkedIn, Twilio, Vercel, Google. V3 is the version a customer never has to log into: the system writes the weekly report, books the next meeting, and asks for a five-minute decision when one is genuinely required. The default state is the operator getting their week back.

The internal target is a million dollars in cumulative platform revenue by mid-2026. The external target is more interesting: we want to make it normal — boring, even — for a small team to run a sales motion that previously required twenty-five people. That's the company. That's the product. That's the whole game.