The open approach behind AgentReady Base - the principles and steps, published free and in the open.
New here? What is AgentReady? · Take the AI readiness checklist →
If one of these is your company, the principle that answers it is named right beside it.
Everything tipped into one searchable archive, hoping the right thing surfaces - and when it doesn't, the system answers plausibly anyway instead of saying it doesn't know. Answered by principles 02 and 03: one source of truth, written plain on top and precise underneath, so an answer can be traced rather than guessed.
Every vendor bolts AI inside its own product, so five systems each hold a fifth of the picture and each assistant is confident in its fragment. Answered by steps 1 and 3: organize one home across the business, and set the rules for what each assistant may touch - so the picture is whole and governed.
One-off automations wired between systems that work until something changes upstream, break silently, and cost the most because nobody wrote down what was connected to what. Answered by principle 01 and step 5: if it isn't written down, it doesn't exist, and a living home kept current instead of frozen.
Most companies guard their method. We publish ours openly because the value was never in the secret; it's in doing it well. You can read this, follow it, and get somewhere on your own. Most businesses will still want help applying it because the hard part isn't knowing the steps but the careful work of doing them on real, messy files without breaking anything. Take what's useful. If you'd rather we did it with you, that's what AgentReady Base is. It also keeps us honest; a published method is one you can check.
AI assistants are finally good enough to help with real work. Two things stop them. First, your knowledge isn't written down, so there's nothing for them to work from. Second, you can't safely point AI at your files because of passwords, client data, and things it shouldn't see. The method fixes both in an order that keeps you in control: get the knowledge down, clean it, set the rules, and let AI in one careful step at a time.
If you remember nothing else, remember these.
Knowledge that lives only in someone's head is invisible to everyone else - and to AI. Get it out of heads and into one place.
One organized home for what the business knows; one place for each thing. Not three copies in three folders that quietly disagree.
Write so a newcomer reads the plain version up top, with the exact detail below - so the same document works for a person and for an AI.
Remove passwords, keys, and personal data before any AI sees a file - not after. This earns you the right to use AI on real work at all.
Start it read-only. Give it small jobs. Widen what it can touch only as it proves reliable - and keep the keys to anything that changes the live business.
Each statement carries a small tag - from a document, from a conversation, held only in memory, or confirmed by a named person on a date. So an answer can be traced, not guessed, and "uncertain" never quietly becomes "fact".
Documents say plainly what is unbuilt, unsigned, unverified, and unknown. A file that sounds more finished than reality makes people - and AI - act with false confidence.
Done in order, each step makes the next one safe. What this asks of your team: less than you'd think - a few guided sessions up front, then two small daily habits. The upkeep is light by design.
Take it away: download the one-page method (PDF) - no email needed, made to forward.
Set up one structured home for what your business knows - in the tools your team already uses (Microsoft 365, Google Drive, or plain shared folders). Not a perfect filing system; one obvious place, organized by the real areas of your business.
Get what your business knows into that one home from two sources. Some of it is already written but scattered across systems, folders and mailboxes - so it gets found, sorted, and reduced to one version everyone can rely on. The rest lives in people, and comes out through short guided sessions where they talk through how things actually work while it gets written up - you're capturing what they already know, not asking them to write essays. Some knowledge exists only as how someone does the job, captured later by watching them do it. The sessions ask little ongoing effort of your team; taking stock of what already exists needs someone to grant access and point at where things live.
Write the rules down, in plain language, area by area - and agree on them before anything is connected: what an assistant may touch, what it must never see, who does which kind of work and where, what's binding versus advisory, and what needs a person's sign-off no matter what.
Introduce an assistant at the lowest rung read-only. Give it a small job, check the result, and widen its reach only as it proves reliable, with each step written down. Early on it produces what you'll actually read: a short recurring summary that surfaces what changed, what's gone stale, and what needs attention. Two simple habits keep the base alive alongside it: an end-of-day "what changed?" and a one-line note whenever something changes directly in an outside system. The keys to anything that changes the live business stay with you.
The know-how that lives only in how someone works gets captured by watching them do it. Methods that aren't specific to one area move into a shared library, written once. And each new area you add is faster and cheaper than the last - the structure and those shared methods are reused, so the first area is the expensive one and every one after it costs less.
Open recipe; done-for-you kitchen.
The step people underestimate is the cleaning. Removing the obvious passwords is easy; the risk is what a quick search misses: a name and address buried in a paragraph, a credential that doesn't look like one. Do the cleaning with automated scans and a careful human read, on a machine you control, before any cloud AI sees the files. If you take one thing from this method to do properly, make it this one.
Last updated: 17 August 2026.
17 August 2026 - corrected the method against its source: restored the five-rung trust ladder with its promotion exam and demotion rule; capture now covers material that already exists as well as what is in people's heads; step 5 is now Grow (each new area cheaper than the last); added two principles (source-tags; write-the-truth); added the three common failure shapes; removed the claim that overstated how much of the method is published; structured the page for search and AI.
21 June 2026 - method first published free and in the open: five principles, five steps, the free/paid line, and the cleaning warning.
