The AI-native operating system that replaces operational headcount.

59 seconds, no voice-over. Every figure in it is the same figure on this page, read from the production API and the repository on 31 August 2026.

Software made your team faster. It never made them fewer.

Every tool you already pay for assumes a person is driving it. Add AI to that and you get a person driving it slightly faster. The headcount stays, because the work between the tools — deciding what happens next, chasing what did not happen, checking that it actually went out — was never in any tool to begin with.

Four layers, one engine.

Warroom is not a social scheduler with AI bolted on. It is an execution graph with a governance layer, and distribution is only one of the things it can reach. The same engine runs the routine operating work of a company — quotes, follow-ups, reconciliation — because that is what it was built to do first.

Not a flowchart. A tree that grows.

Every piece of work I do hangs off one tree: a goal at the top, decomposed into a handful of pipelines, decomposed again into steps. Each step is one of four kinds of node — and knowing which kind it is decides who does it, what it costs, and whether it can be trusted to run alone.

The workflow is executed by the system.
The human sets intent and approves.

Autonomy without an audit trail is unsellable to a business. Warroom was built approval-first because its first operator was personally liable for everything it published.

The model is not the moat. The rules are.

78 rules are live in production today, inherited across three layers. System rules are the operating principles every tenant gets. Industry rules are the ones that keep a business legal — medical aesthetics, food and beverage, and creator-economy compliance ship as separate rule packs. Personal rules are what a specific operator has decided, accumulated one decision at a time.

Rules are versioned with history, never overwritten. Every change records what changed, why, and who signed it off — so a rule can be rolled back, and so the accumulated judgement is an asset rather than a habit.

Your know-how becomes an asset you can carry.

The three rule layers are not only inherited downwards — the top two can be packaged and handed on. Once I had run creator-economy operations end to end, that configuration was itself a product: the pipelines, the thresholds, the rules that keep it legal, lifted out as a pack someone else can install. Thresholds arrive labelled assumed until their operation earns real ones, because the moat is in the numbers, not in the shape of the tree.

Installing a pack copies it. It does not subscribe to it — a seller revising their template must never reach into a line a buyer already has running.

I do not promise full automation.

There is no such thing as a fully automated AI tool, and chasing one is how you end up with a system quietly doing the wrong thing where nobody is watching. Warroom inverts it: the model's first job is to work out which parts it should take over and which parts genuinely need a person. What needs a person becomes a queue on my phone. I answer in one word — approve, revise, stop — and the day is fought.

Describe your process out loud. That is the build step.

You do not need to write code. You have a process already — so say it. Talk your workflow through and the system separates it out for you: which steps a model can take over, which are threshold checks against a standard, and which need you to nod. What comes back is a draft. You add and remove as the work actually runs, and from then on the pipeline grows with you.

In development. Listed here because this log records what is being built, not only what has shipped.

Publishing is irreversible. The system is built around that.

Missing a post costs you one post. Publishing twice costs you the account's credibility, and cannot be undone — it has already been pushed, already been seen. Every trade-off in the reliability contract falls the same way: rather miss and alert than repeat.

Not a mockup. The console, this morning.

Captured from the production system on 31 August 2026. The interface is in Traditional Chinese because its first operator is in Taipei; the English descriptions below each capture explain what is on screen.

The decision queue. Everything the system will not decide alone, ranked by leverage against the north-star metric, each with a deadline. Item 4 in this capture is the system's own card asking its operator to decide whether to apply to Hub71 before 21 August.

Messaging automations, measured end to end. Eight live flows. 2,276 triggers, 2,268 messages sent, 8 failures. Gated and ungated flows are counted separately, because a follow gate trades reach for list quality and you should be able to see the price you paid.

Per-post analytics with thresholds, not vanity. Every post is snapshotted every six hours and scored red / blue / green against engagement, save and share thresholds. The two outliers here — 55,675 and 30,040 reach — are what a single operator produced without an agency or ad spend.

The production board. Scripts carry their format rules with them. Status moves to-shoot → shot → shipped. The system knows what content exists before it promises to publish it.

Where it stands today

In continuous production since 17 Jul 2026 · 45 days, no rollback

Two revenue lines. Both bootstrapped.

It no longer needs me to run.

Warroom started as the system I built to run my own company without hiring the team I could not afford. That is no longer what it is. Every part of it that used to assume a single operator has been pulled apart into something a stranger can own.

Priced on judgement, not on account count.

Charging per connected account is the matrix-account curve — high volume, low care per post. Our customers run five to eight accounts and care about every one of them.

There are no tiers. We sell judgement — the refusal to publish a piece that fails the gate — and judgement does not come in a thirty-piece version and a four-hundred-piece version. Tiering would make you ask the wrong question: how much judgement do I need? The quotas exist, but as ceilings against abuse, set where almost nobody reaches them.

Card details are entered on the payment provider's page, never ours. Provisioning is automatic: pay, and the console is open with your key on the next screen. Fourteen-day trial on every tier — we do not do refunds, we do a trial.

Start with the workflow that costs you the most time.

Bring one process you currently run by hand. We map it into the tree, mark which steps need your judgement and which do not, and you see it execute before you decide anything else. No deck, no discovery phase.

Every plan starts with a 14-day trial and no card. Operators are onboarded by hand today — that is a deliberate constraint while the self-serve flow is finished, not a queue.

The roadmap, on the record.

A page that only lists what works reads like marketing. These four are in flight right now, with the same wording they carry in our own work orders — so you can hold us to the dates.

Still shipping. Every day since July.

This is the engineering record, not a highlight reel. Every entry is dated from the repository and the production database — the days it grew, the days it shipped, and the days it broke. The broken entries are here on purpose. A system you are going to run your business on should be judged on what happened when it went wrong, and anyone deciding whether to trust it deserves to see the same log I do.

This log is appended to as the system changes. Entries are never edited after the fact and never removed — including the ones that did not go well.

I was the market before I built the product.

Six years, four companies. Sole founder twice; co-founder and part-time CTO of a third; co-founder and principal of a fourth. Six years deploying enterprise systems and AI-enabled ERP — the operations Warroom automates are the ones I spent those years inside. Warroom was built end to end by one person.

The company runs with no fixed operations headcount: AI-executed workflows plus an on-demand bench of specialists in engineering, product, sales and advisory.