Writing

Methods, written out in full

Everything here is a method you can run yourself without hiring anyone. We publish the working version rather than a teaser of it — if the method is sound, describing it plainly costs us nothing, and it is the only honest way to show what we would do inside your operation before you let us near it.

Eight pieces. Each one is a procedure, not an opinion.

Method · 2026-08-07

How to document a process people will actually follow

Most process documentation is written by asking someone to describe what they do, which is the defect no template fixes — a description is what someone remembers, a record is what happened. Six steps, written from the record, plus the handover test that is the only one that means anything.

Read it →

Method · 2026-08-07

What an AI agent should record every time it acts

A boundary you cannot verify after the fact is an intention, not a boundary. Six fields per action — including the one almost nobody writes, which is the only one that answers "was it allowed to do that?" — why the record must be append-only, and the four things to actually review each week.

Read it →

Method · 2026-08-07

How to decide what to automate first

The process that annoys people most is rarely the one that pays back first. Four properties decide it — how much time it really consumes, whether the rule can be written down, what breaks if it goes wrong, and whether you would know. A scoring pass you can run on your own operation in an afternoon.

Read it →

Method · 2026-08-07

Why your SOPs are out of date before anyone reads them

Procedure folders rot for structural reasons, not because anyone is undisciplined: a fact recorded in two places with nothing forcing the two to agree will diverge. Four causes, a five-sample test for how bad yours is, and what actually closes the loop.

Read it →

Method · 2026-08-07

Why your operations documentation doesn't match how the work actually happens

Every operation has four separate accounts of itself: what people say they do, what they are observed doing, what the written procedure claims, and what the systems recorded. They disagree, and the disagreements are where the money leaks. A four-source method for finding out which account is true — including the version you can run yourself.

Read it →

Method · 2026-08-07

What should an AI agent be allowed to do on its own?

The question that decides whether autonomous work is an asset or a liability. A concrete model for sorting every action into three buckets — do it, draft it for approval, never touch it — and for making the boundary auditable afterwards instead of implicit.

Read it →

Record · 2026-08-06

The hiring pipeline, running end to end

What the loop looks like once it is finished and running: roughly thirty owned steps, six ownership types, explicit stage gates, and an append-only record of every decision. A worked example rather than a description.

Read it →

Subscribe New pieces are published to an Atom feed — no email address required, no list to leave. Point any reader at https://mochlos.ai/feed.xml.

← Back to Mochlos

Bring us the operational problem you're carrying. We'll tell you directly whether we can solve it, and what we'd do first.

Talk to us