Athanor is an AI-native operating studio. Three modes from one vessel — ventures we own, founders we embed with, engagements we take by hand.
Most "AI products" are thin wrappers over a single model call. We don't build those. Every Athanor venture ships on top of deep agents — long-running, tool-using, stateful — running inside a hardened agent harness we maintain across the studio. Three engineering tenets, applied to every venture below.
Every venture is built on agents that hold long context, plan across hours and days, use real tools, and persist state across sessions. Single-shot prompt chains do not ship from this studio.
Tool registry, sandboxed execution, memory, evals, traces, and a typed orchestration layer — built once, hardened across ventures. Every venture inherits the harness on day one instead of rebuilding it from scratch.
Every product exposes itself to agents — MCP servers, tool-use APIs, machine-readable contracts — before it exposes itself to humans. Human UI is a second client of the same surface.
Most studios choose one posture: venture builder, embedded operator, or strategic advisor. ATHANOR holds all three because worthwhile company-building often needs more than one. The difference is clearest side by side.
Ventures we own and operate. Built inside the studio, financed off our own balance, released only when the cycle is complete.
Founders we embed with. We become a missing co-founder for 6-18 months, on the operating side, with skin attached to outcomes.
Selective engagements taken by hand. Strategy, instrumentation, and craft for teams already underway.
"Six ventures on our balance sheet. Let go only when they can stand without us."
The studio's first mode. We hold the cap table, finance the first build cycle, run the operation, and let go only when the venture can stand without us — when the temper holds.
4 GA / 2 incubating / ready autumn 2026
Continuous-context tooling for engineering teams running long-cycle systems. Live with 240 paying teams.
Operator-grade analytics for embedded marketplaces. Built inside an engagement, spun out as a venture.
Inference scheduling for industrial floor systems. Two pilot lines running, third entering Q3.
Treasury and capital-flow primitives for studio-style operating companies. Used internally; opened externally.
Long-running compute leases for research-stage teams. In thermal stability; graduation queued for autumn.
Field-grade measurement instruments for coastal and estuarine work. One partner deployment live.
Each Studio venture runs on the same treasury surface, audited monthly. Available to embedded partners on request.
"Eleven embedded teams. Four days a week each, for 6-18 months. Then we step out."
The studio's second mode. We become the missing operating co-founder: four days a week, equity-aligned, for a constrained window. Then we step out and the company keeps its shape.
11 embedded / max 14 / 3 graduated
We never run more than fourteen at a time. Below the line — the eleven currently in the vessel.
A named partner-company case — founder quote, role, outcome, dated — will sit here once attribution is cleared with the founder.
Draft fields: Name · Company · Engagement window · One outcome line · One direct quote.
"[Founder quote — one or two sentences, in their voice. The single most concrete thing the studio changed. No platitudes.]"
"Twenty-three engagements on file. Two live. Six- to twelve-week scopes, taken one at a time."
The studio's third mode. Focused interventions for teams already in motion: strategy, instrumentation, capital architecture, hiring, or editorial systems. Accepted one at a time.
23 historical / 2 live / accepting Q3
Re-pointed a stalled growth function around three measurable loops; instrumented before any people changes; team retained.
Built the operator-grade analytics layer that later spun out as Heliotrope. Engagement closed on time, equity option declined.
Designed the role, ran the loop, made the offer, sat the first month. The hire is now CTO.
Restructured a five-entity holding stack into one transparent treasury surface. Reused internally as Salt House.
Quarterly editorial cadence, voice, and publishing rails for an 18-person research org. In year two.
Restructure of a 60-person engineering org around shipping cycles instead of squads. Mid-engagement.
The operating model is simple. Studio is ownership. Partners is embedded operating capacity. Works is a fixed intervention.
No vague advisory. No endless retainers. No misaligned incentives.
We hold the cap table. We finance the first build cycle off our own balance. You join the venture; we don't bill it.
Four days a week, equity-aligned, for a constrained window. We become the missing operating co-founder. Then we step out.
A focused engagement: strategy, instrumentation, capital architecture, hiring, or editorial systems. Single-orchestrator delivery.
Every message lands in one shared inbox read by the studio. We answer within forty-eight hours, on Tuesdays and Fridays. No qualification gauntlets. No pitch theater. If you want to talk, write.
Two rhythms: quarterly editions and daily fragments. Editions are the long account. Fragments are notes from the floor. Both are open. Neither is gated.
On the inflection where a founder stops being the bottleneck, and what we changed at Ferrum, Lyra Labs, and Saffron Capital to get there. With three case readings.
Pricing rewrite for Quill. Three tiers became one: $48/seat/mo, no annual lock, no enterprise tier. Conversion this morning: 6 of 14 trials.
Ferrum closed its Series A - $14M, Index lead. Embedded since Jan 2024. We step out at month nine; the operating wiring stays in place.
Reread Christopher Alexander, Vol. II, pp. 412-430 - on patterns that survive the death of their authors. Reused for a hand-off doc this afternoon.
A founder asked when a venture is finished. We use three criteria: founder runway greater than 18 months, decision velocity flat for two quarters, no remaining open loops on the studio side.