Start Here: teaching notes
Side-by-side notes for 00-praxis-start-here.html (13 slides). The deck is what the room looks at; this is what you say, where you stop, and how you make each idea land for the person in front of you. This is the prelude to the whole library: the idea, the proof, and how they find their own use case. Written from the deck against the Praxis teaching library. Method: teaching-notes-method.md.
How to use these notes
- These are guidance, not a script. Teach the person, not the slides.
- Each teaching slide has a Pause, then ask block. Put the question, then stop. Do not fill the silence. If they freeze, one line: "Take your time, type anything when you're ready, including 'I don't know'." Then wait again.
- By background gives you a line per archetype (from
tutor/scenarios.md). Reach for the one that matches whoever is in the room and ignore the rest. They are a palette, not a checklist to read out. - The proof slides (6 to 10) are real systems. Tell them as real: two are Tutto's own, one is a client's. Conviction comes from "I run my business on this," not from a feature list.
- Hold the voice (
train-the-trainer.mdPart 4): no "great question", no emoji, one term only, Claude Code. Underclaim. Cost is "about twenty dollars a month" and stop. - Run the intake five questions first (
train-the-trainer.md) so you know which background you're teaching to before slide 1.
The systems I build, and how you build them too
Set the frame: this is a short tour in three parts, not a lecture. The idea, the proof, then their turn. The watching is the lesson; you will keep handing it back to them.
"This is a quick tour in three parts: the idea behind how I build, real systems already running, then how you'd find yours. Don't try to remember it all. The point today is to see that it's real and that it's within your reach."
The intake answer they gave you. Name the one repetitive task they said eats their week, and promise you'll point straight at it on the use-case slide near the end.
Three things, then the rest of the library
Show the shape so nobody feels lost: the idea, the proof, your turn. Orient, do not teach.
"Three parts. First the idea, in four quick pictures. Then the proof, four real systems that I or my clients run every day. Then your turn, how you find the one that's yours. If you lose the thread, this screen is the map back."
Twenty seconds, then move. Don't read the list aloud.
Software is exact, AI is likely, pair them
Software is deterministic, the same result every time. AI is probabilistic, it produces what's likely to help. Claude Code is the pairing: you talk in plain English, it writes the code that runs the same way every time.
"Two different things here. Normal software does exactly what you tell it, the same way every time. AI doesn't follow rules, it produces what's likely to be useful. Claude Code is the marriage of the two: you talk to it in plain English, and it writes the code that does the work, the same way every time. Think of it as an Excel formula that now lives in your folder." (tutor/framings.md, Excel formula; Claude as translator.)
- "Where in your work do you need the exact same result every time, and where do you actually want judgement?"
- "What's something you do the same way every week that could become a formula?"
(wait)
- Investigator
- the report structure is the formula; the reading of the case is the judgement.
- Analyst
- the calculation is deterministic; choosing the analysis is probabilistic.
- Designer
- export sizes are a formula; the concept is judgement.
- Knowledge worker
- the repeated process is the formula; the exceptions are yours.
- Reporter
- the monthly format is the formula; the commentary is judgement.
- Content
- the house-style rules are the formula; the idea is judgement.
- Form-filler
- the template mapping is the formula; the odd case is judgement.
- Outreach
- the sequence is the formula; the personal note is judgement.
- Exec
- the dashboard is the formula; the decision is judgement.
"I'm not technical" / "too old" (method/10-threshold.md). The six-year-old framing: the unlock is asking obvious questions constantly, not knowing the answers already.
Every system I build is the same shape
One architecture under everything: a folder with the rules in plain English (CLAUDE.md), packaged know-how (a skill), a script that does the repeating, and your files. You don't put your files into an app, you put Claude into your files. The folder is the memory.
"Everything I build looks the same underneath. A folder on your machine. The rules written in plain English, the know-how packaged up, a script that does the repeating, and your own files. You don't put your files into an app, you put Claude into your files. The folder is the project, and the folder is the memory. There's no website to go down and no portal to break." (patterns.md, folder = memory; no web portal to break.)
- "What lives in scattered places right now that would be calmer as one folder?"
- "If you wrote down the rules for how your work should be done, what would the first three be?"
(wait)
- Investigator
- the case folder holds the files, the house rules, and the report script.
- Analyst
- the folder holds the data, the method, and the output template.
- Designer
- the folder holds the brand rules, the assets, and the export script.
- Knowledge worker
- the folder is the project that finally remembers between sessions.
- Reporter
- the folder holds last month's format and this month's data.
- Content
- the folder holds the style guide and the drafts.
- Form-filler
- the folder holds the template, the rules, and the inputs.
- Outreach
- the folder holds the market knowledge and the sequences.
- Exec
- the folder is one place the whole operation can be asked about.
"my data" / "is it safe" (tutor/objections.md; method/10-threshold.md Stage C). Data-safety as control: there's a floor (dummy data, nothing installed) and a ceiling (runs on your own machine). They choose where they sit.
Don't ask for the answer, ask for the thing that makes it
Ask the AI for an answer and you get a good one, different each time. Ask it to build the script that produces the answer, and you own something that runs the same way for the next thousand. This one move is behind every system in the deck.
"Here's the move that changes everything. If you ask the AI for an answer, you get a good one, but a different one each time. If you ask it to build the thing that produces the answer, now you own something that runs the same way for the next thousand. We stop asking it for a fish and ask it for the rod."
- "What do you redo by hand every time that could be made once and then reused?"
- "If you had one button for one task this week, which task would it be?"
(wait)
- Investigator
- not "summarise this case", but "build the thing that summarises every case this way".
- Analyst
- not "analyse this month", but "build the analysis I run every month".
- Designer
- not "resize this", but "build the resizer for the whole set".
- Knowledge worker
- not "do this once", but "build the step I'll repeat".
- Reporter
- not "write the report", but "build the report generator".
- Content
- not "draft this post", but "build the drafter that holds our voice".
- Form-filler
- not "fill this form", but "build the filler for the batch".
- Outreach
- not "write this email", but "build the sequence engine".
- Exec
- not "give me the number", but "build the thing that gives me the number on demand".
"will it be reliable" / "good enough". Set the 80/20 ceiling now: you'll get most of the way and finish the last bit by hand, and the script repeats that gain every time.
Built, not bought, starting with my own company
The strongest proof is that I run Tutto on these: books, events, outreach, with the subscriptions cancelled. Built, not bought, and the asset stays yours, on your machine, no monthly fee. Then the same approach for clients.
"The strongest proof I can give you isn't a slide, it's that I run my own business on this. My accounting, my events, my outreach, all built this way, and the subscriptions they replaced are switched off. The asset is mine, on my machine, with no monthly fee. Let me show you three: two of mine, one we built for a client, and all three get better the more they're used."
- "Which of your tools is a monthly subscription you'd rather own outright?"
- "What would it mean to you if the system was yours, not rented?"
(wait)
- Investigator
- your case system as something you own, not a per-seat licence.
- Analyst
- your analysis stack on your machine, not a SaaS you feed data to.
- Designer
- your production pipeline owned, not rented per month.
- Knowledge worker
- the small tools you'd stop paying for if you owned them.
- Reporter
- your reporting, free of the dashboard subscription.
- Content
- your content engine, yours to keep.
- Form-filler
- your processing, off the metered platform.
- Outreach
- your outreach system, not a seat in someone else's CRM.
- Exec
- the line on the budget for tools you could own instead.
"but you're technical, I'm not". The honest line: the software is genuinely cheap and genuinely powerful; the scarce thing is a guide who has done it, and that's the role you're paying for. You learn by asking, not by coding.
My company's books and events, no subscription
A real owned system: an event or job becomes a quote, the quote becomes an invoice, it goes out through our own email service, and the books stay current. The accounting package is cancelled and the data stays with me. A whole back-office, built not bought.
"This is my own company's books. A job or an event becomes a quote, the quote becomes an invoice, and it goes out through our own email service, with the records kept up to date. I cancelled the accounting package. There's no portal to log into, no monthly fee, and the data sits with me. Built once, it runs every month."
- "Where does your quote to invoice to payment flow leak time or money today?"
- "What would you stop paying for if you owned this?"
(wait)
- Investigator
- case closed, invoice raised, sent, all from the case folder.
- Analyst
- billing reconciled from the same data you already hold.
- Designer
- project to quote to invoice without retyping anything.
- Knowledge worker
- the admin tail of every job handled once.
- Reporter
- the monthly billing run as a script, not an evening.
- Content
- retainer invoices out on time without the chase.
- Form-filler
- quotes generated to a fixed template, every time.
- Outreach
- won deal to quote to invoice in one move.
- Exec
- the back office running without a subscription stack.
"my accounts are too complex / regulated". Start with one slice, not the whole ledger; set the 80/20; and for sensitive data it can run locally (slide 11).
Outreach that knows each country
A sequence-based emailing and intelligence system with its own dashboard. The knowledge of how people communicate in different countries lives inside it as rules, so every message lands the right way. It tracks the pipeline and generates the reports and quotes to send straight to customers. Know-how captured as rules, run at scale.
"This one is outreach with a brain. It's a sequence-based emailing and intelligence system with its own dashboard. The knowledge of how to talk to people in different countries lives inside it as rules, so every message lands the way it should. It tracks the whole pipeline, and it generates the reports and quotes I send straight to customers."
- "What knowledge lives only in your head about how to handle a particular client or market?"
- "What would change if that knowledge ran at scale, the same way every time?"
(wait)
- Investigator
- the protocol for each jurisdiction held as rules, applied every time.
- Analyst
- the reporting cadence and house format, automated to each stakeholder.
- Designer
- the per-client brand rules applied without re-briefing.
- Knowledge worker
- the "how we do it for this account" captured once.
- Reporter
- the right report, in the right shape, to the right audience.
- Content
- tone tuned per channel and market, as rules not memory.
- Form-filler
- the per-region template chosen automatically.
- Outreach
- the system this slide is literally built for, sequences that respect each market.
- Exec
- a live dashboard of the pipeline, plus quotes out the door.
"AI emails sound generic". The rules here are yours and specific, not a generic model voice, and you still sign off the last ten percent.
One spec PDF in, seventy-eight assets out
A client build. An e-commerce platform sends a spec for 78 asset formats. A script turns that spec PDF into a reusable set of rules, like one big spreadsheet formula, then produces all 78. Each comes back editable, so a person finishes the last ten percent. The brittle manual part becomes a formula you run.
"A client gets a spec from an e-commerce platform, 78 different asset formats. We turned that spec PDF into a reusable set of rules, like one big spreadsheet formula, and now it produces all 78. They're not all perfect first time, so each comes back editable and a person finishes the last ten percent. The slow, brittle part, reading the spec and making every size by hand, became a formula they run."
- "What's a job where you produce many versions of the same thing to someone else's spec?"
- "Where would you happily take ninety percent done and finish the rest yourself?"
(wait)
- Investigator
- one instruction, every required document produced to format.
- Analyst
- one dataset, every chart and cut generated to spec.
- Designer
- one campaign, every size and placement rendered, ready to tweak.
- Knowledge worker
- one input, the whole batch of outputs in one pass.
- Reporter
- one month's data, every variant of the report at once.
- Content
- one piece, every channel's version drafted.
- Form-filler
- one record, all the downstream forms filled.
- Outreach
- one offer, every market's variant produced.
- Exec
- one brief, the full set of deliverables drafted for sign-off.
"ninety percent isn't good enough for my work". That's exactly why the output is editable, not final, and it improves with use, which is the next slide.
It gets better the more you use it
The return people miss. The user shows the system what it got wrong, the rules sharpen, and ten percent to fix becomes five, becomes one. A bought tool is as good as it will ever be on day one; a system you teach keeps improving, and the time saved grows.
"This is the part people miss. Not all 78 are perfect the first time, but the user shows the system what it got wrong, and the rules get better. Ten percent to fix becomes five, becomes one. A tool you buy is as good as it will ever be on day one. A system you teach keeps getting better, and the time you save grows every month." (patterns.md, the system improves through use.)
- "What's something you'd be glad to teach once and never have to re-explain?"
- "What would a few percent less rework each month be worth to you over a year?"
(wait)
- Investigator
- each corrected report teaches the next one your standard.
- Analyst
- each flagged misread sharpens the method.
- Designer
- each tweak you make teaches the brand rules.
- Knowledge worker
- every fix you save makes the next run cleaner.
- Reporter
- each month's edits become next month's defaults.
- Content
- every correction trains the voice closer.
- Form-filler
- each caught edge case becomes a new rule.
- Outreach
- each reply you learn from tunes the sequence.
- Exec
- the system compounds while a bought tool stands still.
"is the learning curve worth it". Yes, because the returns compound: tell it at the end of a session what worked and what didn't, and next time it's better.
One build becomes many, and it can stay private
Two takeaways. Build it once for a single case and you've built it for every case shaped the same, ten or a hundred. And when the data is sensitive, the whole system can run on your own machine, where nothing leaves your network. Reuse, plus control.
"Two things to take with you. Build it once for a single case and you've built it for every case that looks the same, ten of them or a hundred. And when the data is sensitive, the whole thing can run on your own machine, where nothing leaves your network. Confidentiality stops being a promise and becomes a fact about where the data sits."
- "What do you do once and then immediately have to do again, slightly different?"
- "What data would you only trust if it never left your building?"
(wait)
- Investigator
- one case system covers the whole caseload, and runs offline for sensitive work.
- Analyst
- one analysis serves every period, on-premise where the data demands it.
- Designer
- one pipeline serves every client of the same type.
- Knowledge worker
- one tool covers the whole class of task.
- Reporter
- one generator serves every report like it.
- Content
- one engine serves the whole content line.
- Form-filler
- one filler serves the entire batch and beyond.
- Outreach
- one system serves every market.
- Exec
- one build becomes an operational capability, kept in-house.
"we're not allowed cloud AI" / "IT won't approve" (tutor/objections.md). The local-only ceiling answers it: nothing leaves the network, so it's a control question, not a blocker.
How to find the one that's yours
The test for a first use case: a task that touches several systems, comes out the same shape every time, and is costly when it's wrong. Score on all three and start there, this afternoon. The honest economics: about twenty dollars a month, ninety percent of the way then finish by hand, and the only scarce thing is a guide.
"You don't need a strategy, you need one task with three marks: it touches several systems, it comes out the same shape every time, and it's costly when it's wrong. Score on all three and that's where you start, and you can find one this afternoon. The honest economics: about twenty dollars a month, you'll get ninety percent of the way and finish the last ten by hand, and the only scarce thing is someone who has done it before, not the software." (tutor/framings.md, the 80/20; cost line verbatim.)
- "Run the three marks against your week. What scores on all three?"
- "What's the one you'd most want working by next week?"
(wait)
- Investigator
- case intake to report, several systems, same shape, costly if wrong. Textbook.
- Analyst
- the recurring analysis that pulls from several sources into one format.
- Designer
- the batch of assets produced to a fixed spec.
- Knowledge worker
- the cross-system task that starts from an email and touches many tools.
- Reporter
- the weekly or monthly report that's the same shape every time.
- Content
- the house-style production that's costly to get wrong.
- Form-filler
- the categorisation against a template, done at volume.
- Outreach
- the proposal or sequence generated to a known shape.
- Exec
- the operational report you need the same way, on demand.
"this sounds too good" / "what's the catch" (tutor/objections.md). The catch, plainly: the guide costs, the software doesn't. Say it straight.
Talk to it like a colleague, give it your context, build the thing that repeats
The through-line of the whole tour and the whole library: talk to it like a colleague, give it your context, build the thing that does the repeating. That's how every system they saw was made, and how they'll make theirs.
"If you remember one thing: talk to it like a colleague, give it your context, and build the thing that does the repeating. That's how every system you just saw was made, and it's how you'll make yours. The rest of the library goes one level deeper, one course at a time. You can start anywhere."
Return to the task they named at the start. Say which course in the library is the natural next step for it, and say the cost line once, "about twenty dollars a month," then stop.