Put your operations on rails

This is stage four of the five-stage founder journey — the Operate stage. With ten customers won in stage three, the work of the company starts repeating: onboarding, support, invoices, content, reporting, follow-ups. Operations is everything your company does repeatedly to stay alive, and the habits you set while it is small decide how expensive it is to run later. The lean version rests on three practices — keep fixed costs near zero, write down every recurring process, and move the written processes onto rails that run without you.

Keep fixed costs near zero

Every fixed cost shortens your runway and narrows your options, so default to not adding them. No office before customers demand one. Free tiers and monthly plans over annual contracts, because a young company's needs change faster than a contract term. Above all, be slow to hire, because salaries are the fixed cost that dominates every early budget — and much of the work that used to justify an early hire, the drafting and researching and reporting and following up, is now work agents cover with your review. A company whose costs are near zero can survive slow months, say no to bad customers, and change direction overnight. That freedom is worth protecting deliberately, because every cost you add trades a little of it away.

Write down every process the second time you do it

The second time you do anything — send an invoice, onboard a customer, publish an update, answer the same support question — write down how, in plain steps, before moving on. Not the first time, because a thing done once might never recur, and not the fifth time, because by then you are already improvising a slightly different version on each pass. The second time is the signal that a process exists.

Writing it down costs minutes and repays them every cycle, because an undocumented process exists only in your head, gets done differently each time, and can never be handed to anyone else, human or agent. Treat the growing pile of written processes as a real asset. It is the operating manual of your company, and companies that have one can delegate. Companies that do not cannot, no matter how much capacity they buy.

The written process becomes a workflow

Here is where this stage becomes concrete in Task Machine: a written-down process is most of a workflow already. The steps become nodes, and approval and verifier steps sit exactly where your check-the-result judgment belongs. The handover motion runs the same way every time:

1. List what repeated last month. Go through your calendar, your sent folder, and your task history, and write down every job that happened more than once. This list, not your instincts, decides what gets handed over first — the jobs that repeat most often and vary least are the best first candidates.

2. Install or build the matching playbook. For most recurring jobs a playbook already exists as a ready-made bundle of agents, workflows, and documents: "Weekly content pipeline" for the publishing cadence, "Support inbox triage & reply drafter" for the questions customers keep asking, "Invoicing & expense drafter" for the money paperwork, and "Bookkeeping & month-end close" for the monthly close. For processes that are genuinely yours alone, "SOP & runbook documenter" turns the steps you wrote down into a runbook an agent can follow, and the workflow builder lets you draw the process directly.

3. Run it supervised and correct heavily. Agents start supervised, so every draft and every consequential action waits in your inbox. The early runs will need real correction, and that is the point — each correction you make goes somewhere durable, into the agent's instructions, its memory, or the documents it reads, so the next run starts from what you fixed rather than from zero.

4. Add budgets and verifier checks. A budget caps what an agent or workflow can spend, in money or tokens, with alerts as spending crosses 80% and again at 100% — so a process on rails can never become an open tab. Verifier steps check work against your standards before it ever reaches your inbox, which means what does reach you is worth your attention.

5. Raise autonomy on the record of clean runs. When the run history shows approvals going through unchanged week after week, that record — not a feeling — is your cue to raise the agent's autonomy so routine steps proceed on their own and only the consequential ones interrupt you. Autonomy in Task Machine is earned on evidence and raised one process at a time, never granted globally on optimism.

Repeat that motion for one process at a time, and the shape of your week changes. The work still happens every day, but your part of it becomes the judgment: reading a draft, approving a send, answering an agent's question, redirecting a run that drifted.

This stage has a whole section behind it

Operating a company this way raises questions that deserve more than one page, and the Run a company with agents section is the deep dive behind this stage. It covers which work to hand to agents first and why some jobs are better first candidates than others, how to keep control without becoming the bottleneck every draft waits on, how autonomy levels work in practice as trust accumulates, how budgets keep spending honest, and how to decide when agent output can be trusted without reading every word. When a process resists the handover motion above, the answer is usually in one of those pages.

You are ready for the next stage when every recurring process in the company is written down, at least three of them run as workflows with approval steps in place, and your week is measured in inbox decisions rather than in hours of doing the work yourself.

Once operations run on rails, something subtle becomes true: the company works without consuming all of you, and the attention you freed is the raw material of growth. Stage five, scale what already works, is about spending that attention deliberately.