A persistent agent working in the background, on a schedule, without you starting it each time.
Everything through Level 4 still needs you to press go. Level 5 doesn’t: a loop runs on a schedule, checks something, does something, and reports back, whether or not you were watching. That’s the whole appeal, and it’s also the whole risk.
A loop that runs unattended and unreviewed doesn’t announce when it starts drifting. It just quietly does the wrong thing, competently, until someone notices. The fix isn’t more caution about turning loops on. It’s building the review rhythm before you do.
ANY TOOL
Pick one thing you already do repeatedly and set it up as a scheduled or triggered agent, not something new invented for the demo.
ANY TOOL
Build the review ritual before you turn it on. Decide when and how you’ll actually check its work. Put it on your own calendar, not just a hope.
ANY TOOL
Write down what “good” looks like for this loop, in one sentence, so you (or it) can actually tell when it’s drifting.
The same four pieces of work recur at every level - and they are your WILD Intelligence, working. The name on each card is the job; the capacity behind it is what the reps build. The framework itself ↗
Before anything runs on its own: what is this loop for, what does done look like, what would tell you it’s drifting? A loop without a charter optimizes for its own momentum. The charter is Wisdom, written down.
What it must never do without you. Where it must pause and ask. What it may read but never send. Intention at Level 5 is the boundary between what runs alone and what waits for your judgment - drawn before launch, not after the first surprise.
A running loop is a report you must actually read. This is Leadership as rhythm - showing up to what your systems bring you, deciding, correcting. Loops don’t fail loudly; they fail unread.
Every miss becomes a permanent fix - that’s how a loop compounds instead of decays. The correction ledger is Discovery’s bookkeeping: the loop learns because you noticed.
A prompt answers a question you have right now. A loop handles an obligation that keeps returning - the weekly report, the morning read of the day, the pipeline that needs an eye on it. Level 5 is learning to think in loops: the same job, on a rhythm, with memory of last time and boundaries on what it may do alone.
Every loop that holds has four parts: a charter (what it’s for), sources (what it reads), boundaries (what it never does without you), and a review it reports to. That last one is the part people skip. A loop isn’t finished when it runs; it’s finished when you’ve read what it did.
This is where our own WILD×AI Navigator lives: the persistent-context environment we run our work and life on. Its mornings start with a loop that reads the day, checks the systems, and queues what needs human eyes; a standing set of loops handles the rest, each stopping for the human wherever judgment lives. It is the reference implementation of this level - not a demo built for this page, the operating reality this site is written from. See the Navigator in the Library →
Designing a first loop that actually holds - and the environment it runs inside - is exactly the kind of build we do with people. Begin a conversation ↗
A loop is not a free-floating script; it runs inside something. Two pieces make that environment hold at small scale. First, a place for the context to live - not a real database yet, usually a well-organized tree of plain files, or a light store like SQLite when a loop needs to remember rows between runs. The pattern worth stealing is the world model: one living context file per client or project that every touchpoint feeds into, consulted before any meeting or deliverable. Second, somewhere for the schedule to fire that is not your laptop - a cheap always-on box or a scheduled cloud runner pulls the trigger, while the work still happens in your environment. That is the whole unattended-infrastructure story for a business under fifty people.
The Wisdom discipline comes first, or the leverage inverts: simplify and articulate the system before you wire a loop into it, because a loop pointed at a mess automates the mess competently, on a schedule, until someone notices. The safe first version is the honest one - read access to everything, write access to nothing - widened only as the loop earns trust from what you watch it do.
You’re ready for Level 6 when one loop isn’t enough anymore, and you’re finding yourself wanting several things running at once instead of managing them one at a time.