Factory
Preamble
Factory is the garden’s venture engine: an evidence-gated system for discovering demand-backed software wedges, validating them, building products, distributing them, operating them, and learning from market outcomes without letting automation manufacture confidence faster than judgment can inspect it.
The Machine That Wants To Make Companies
Factory aims at automated software creation and scaling. The full version finds demand, validates a wedge, creates the product workspace, builds the software, launches it, watches the market, and feeds the outcome back into the next run.
That ambition puts pressure on every layer. A factory can multiply good judgment. It can also multiply the moment where an agent turns weak evidence into a plausible next step and keeps moving.
The current work starts with refusal. Before the Factory can run faster, it has to prove that bad data, thin briefs, stale projections, unsafe builds, and fake demand cannot pass as progress.
The Founder Moves Upstream
Most garden systems test one boundary at a time. Constitution asks what the machine may own. Coach asks whether advice can sharpen a decision without renting conviction. Decision Auditor asks whether a choice can survive evidence, alternatives, and risk.
Factory raises the stakes because the decision becomes a loop. If the loop works, the founder stops choosing every task by hand and starts designing the conditions under which tasks, ventures, and budgets move. That can be leverage. It can also become a laundering machine for responsibility: the founder writes a gate, a score, a prompt, or a budget ceiling, then watches the system produce the next thing.
The handoff changes the moral shape of the work. A bad single decision wastes a day. A bad autonomous loop can turn one weak premise into a product, a launch, a dashboard, and a postmortem before the founder notices the original lie.
What The Research Reveals
Research points at useful primitives. Polsia is the runtime reference: many venture instances, autonomous cycles, specialist agents, per-company infrastructure, dashboards, and portfolio-scale ambition. Cofounder is the memory and workflow reference: persistent company context, task state, agent orchestration, and business operations that survive beyond one chat.
The Harness Has To Come First
Factory needs a stack of harnesses. The evidence harness keeps source artifacts, claims, clusters, opportunities, and recommendations separate. The founder harness rejects unsupported briefs. The venture-instance harness gives each company local truth: work, evidence, docs, code, analytics, decisions, and postmortems.
The runtime harness owns queues, retries, logs, maturity states, approvals, pauses, replays, budgets, and kill controls. The build harness owns PRDs, specs, isolated work, tests, deployments, rollback, and observability. The market harness owns landing tests, outreach, waitlists, interviews, pricing asks, objections, and response writeback. Without those harnesses, the Factory becomes a machine for producing motion.
The absurdity is visible before the first product ships. I am building a factory to produce software businesses before I have proven the first app deserves to exist. The only defensible version of that move is a factory whose early work is mostly gates, receipts, quarantine, and brakes.
The Lifecycle Has Teeth
Discovery finds evidence-backed wedges. Validation turns them into pass, loop, or kill. Demand Seeding tests whether anyone cares before the product exists. Venture Workspace gives a validated wedge a local operating surface. Distribution selects channels and offers. Growth learns from revenue, retention, support pain, objections, and churn. Market Intelligence updates the constraints. Operations keeps the runtime honest about maturity, cost, and control.
Each branch can produce its own false positive. Discovery can produce rows. Validation can produce memos. Demand Seeding can produce attention. Venture Workspace can produce code. Distribution can produce launch activity. Growth can produce dashboards. Market Intelligence can produce links. Operations can produce completed tasks. Factory only progresses when one branch reduces uncertainty for the next.
The First Real Proof
The project has launched locally through vision, roadmap, branch futures, and active Discovery work. The architecture is a sapling until a venture completes the loop.
The Factory lands when one wedge moves from evidence to validated market probe to shipped product to outcome writeback, and the next run becomes cheaper, sharper, or easier to kill because of what the first run taught it. The machine is trying to automate venture creation. Its first job is making sure it cannot lie at scale.