InnerStack is a working journal of AI workflows for small businesses — and the record of a service being built in public.
What this site is
Most writing about AI tools is either a press release with a rating attached or a tutorial for a workflow nobody actually runs. This site is the opposite bet: we publish the workflows we personally depend on, every change we make to them, and the reason for the change. When something we tried turns out not to be worth it, that gets published too, because those are the entries we would want to read.
Alongside the journal there are four hubs of guides and reviews — AI tools, AI for specific trades, local and private AI, and short daily news. Those are the pages that bring people here. The journal is why anyone stays.
Who writes it
An author bio belongs here in your own words — who you are, why you work on this, and what you actually run. It is the single most load-bearing paragraph on the site for trust, and it should not be written by an agent.
How the agents work
Six agent roles produce this site, and it is worth being precise about which parts are machine work, because the honest answer is not “none of it”.
A researcher proposes new sources once a week. A swarm of watchers — one short-lived worker per source — scans those sources daily and files candidates, discarding anything without a reproducible workflow in it. An organizer tags each candidate to a workflow and flags whether it might improve it. A project manager tests the promising ones and records a verdict. An editor revises the affected workflow page and writes the diff. A publisher ships it.
What that means for you as a reader: the gathering is automated, the judgement is not. No journal entry is published without a human verdict, and an adoption has to name the metric that improved — time, cost or quality — or it stays marked as merely evaluated. The hub guides are drafted with AI assistance against a keyword plan and reviewed before publishing; the journal and the workflow pages are not autoblogged at all.
Why the changelog
Any site can claim a process. A versioned changelog is the cheap, checkable version of that claim: you can see what the workflow said last month, what changed, which entry caused it, and whether the reasoning holds up. It also keeps us honest, because a change with no defensible reason looks exactly as bad as it is.