14:25Mansel Scheffel · Tutorial
I Finally Got My Team to Use Claude Code (without forcing them)
A 14-minute consulting framework for getting any team to adopt AI without mandates, arguments, or forced rollouts.
June 10thThe argument in one line.
Before you build any AI automation on top of a client's systems, you need a machine-generated map of what data actually exists, where it lives, and whether it is safe to touch — because the client's mental model is almost always wrong.
Read if. Skip if.
READ IF YOU ARE…
- You are an AI consultant or agency owner who gets hired to build AI operating systems for clients.
- You use Claude Code or Claude Cowork and want to automate the discovery phase of an engagement.
- You have built automations on top of systems you did not fully understand and paid for it later.
- You want a defensible scope document backed by evidence rather than what the client described in a call.
SKIP IF…
- You are building AI tools for a single business you own and know intimately — the discovery overhead is not worth it.
- You do not do client-facing consulting work.
The full version, fast.
Clients almost never know what data they actually have, where it lives, or whether it is clean enough to automate. Running a Claude data map skill before any build gives you a machine-generated inventory of every connected SaaS system — schemas, record counts, PII flags, security misconfigurations, and dormant infrastructure. The output is an interactive HTML report you can walk through with the client. The central argument: discovery is not optional overhead; it is what separates consultants who build the right thing from those who automate a trash pile.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 02:30
01 · Why you map data before you build
Three-part discovery framework: Audit, Pod Map, Data Map. The trash-shack metaphor: clients think their data looks like a clean house; it almost always looks like a shack under a bridge. Lead with evidence, not opinion.
02:30 – 04:24
02 · Connecting systems via read-only MCP
Credential checklist slide. How to add MCP connectors in Cowork — read-only only, no write access. Custom connector path for tools without native support shown with Fathom and Beehiiv as examples.
04:24 – 06:35
03 · What the map actually unlocks
Illustrated slide with six outputs: build order, source of truth, security scope, data intake, context quality, roadmap and quote. The data-to-context chain diagram connects the map to a refined AI OS.
06:35 – 08:20
04 · Running the skill and viewing the output
Skill finishes in Cowork. HTML output opened — interactive source list with side panel per system. HeyReach LinkedIn data shown in detail: 1,086 enriched leads, one real campaign, small test runs.
08:20 – 10:50
05 · Security flags, real findings, build plan
Supabase panel: RLS disabled on all 6 live tables. Findings summary: lead data fragmented across Apollo and HeyReach, 6 of 7 Supabase projects inactive, Google Drive almost entirely screen recording exports. Open questions list for client confirmation before building.
Lines worth screenshotting.
- Clients think their data looks like a tidy house. It almost always looks like a shack built under a bridge.
- The data map does not build the AI operating system. It tells you how to design it.
- Read-only MCP connectors mean Claude cannot do anything destructive — the only safe way to run discovery on a client's live systems.
- Six of seven Supabase projects being inactive is a common finding — clients do not know which infrastructure they are actually using.
- RLS disabled on all live database tables is a security finding that surfaces in minutes with a data map and would take days to find manually.
- Lead data fragmented across Apollo and HeyReach with no system of record is the kind of problem that breaks automations before they are even built.
- Discovery is not a phase you can skip to move faster — skipping it means building automations that break on data you did not know existed.
- The data-to-context chain starts with the map: you cannot refine AI skills with good context until you know where the data lives and whether it is clean.
- Google Drive's recent file volume being almost entirely screen-recording exports is a common finding — most knowledge is locked in unusable formats.
- An open questions list generated by the skill surfaces what needs client confirmation before you write a single line of automation.
- The same Claude skill that runs in Cowork runs identically in VS Code or Claude Code — the vendor does not matter, the connectors do.
- Charging for scope is only defensible when the scope comes from a map, not from what the client told you in a discovery call.
Map the data before you touch the systems.
WHAT TO LEARN
A machine-generated data map surfaces what clients do not know they have — and what they do not know is wrong — before a single automation is built.
- Clients almost never have an accurate picture of what data lives in their systems, which systems are actively used, or whether their databases have basic security controls enabled.
- Read-only MCP connectors let you audit a client's entire stack without any risk of modification — if you cannot write, you cannot break anything.
- A data map produces six decision inputs: build order, source of truth, security scope, data intake priorities, context quality assessment, and the basis for a quoted roadmap.
- Dormant infrastructure is the norm — six of seven database projects inactive, cloud storage full of raw screen recordings rather than organized knowledge.
- Row Level Security disabled on a live database is a finding you can surface in minutes with a structured scan; finding it after building automations on top of it is a much bigger problem.
- Lead data duplicated across two outreach tools with no system of record breaks automations before they run — a data map catches this in the discovery phase, not in production.
- The discovery pass is what separates an engagement with a defensible scope from one that expands indefinitely because neither party knows what done means.
- Charging appropriately for AI consulting requires a map — you cannot scope what you cannot see.
Terms worth knowing.
- AIOS
- AI Operating System — the layered set of skills, context, and connectors built on top of Claude that runs a business's AI workflows. The data map is the prerequisite before building one.
- MCP connector
- Model Context Protocol connector — a plugin that lets Claude read from a specific SaaS tool without needing a custom API integration. Read-only connectors are used in discovery.
- Data map
- A machine-generated inventory of every connected system: what data it holds, how it is structured, what is active, what is dormant, and whether it has security issues.
- RLS (Row Level Security)
- A database access-control feature in Supabase and PostgreSQL that restricts which rows a user or service can read or write. Disabled RLS means any authenticated user can read every row.
- Pod Map
- The second of three discovery steps — maps the workflows, manual steps, and automation verdicts before the data map is run. Sits between the audit and the data map.
- Discovery pass
- The three-step pre-build process: Audit (stakeholder interviews, opportunity scoring) then Pod Map (workflow inventory) then Data Map (system inventory). The data map is one third of this pass.
- Cowork
- Claude's web-based collaborative environment where MCP connectors are added via a Customize tab and skills run interactively. Equivalent to Claude Projects with tool integrations.
Things they pointed at.
Lines you could clip.
04:15
“The map does not build the AIOS. It tells you how to design it.”
01:40
“A lot of people often think that their data looks like this merry little house that is already built, but in reality, it actually looks like some kind of trash shack that's been built under a bridge.”
02:00
“Everything that we do for our entire engagement is built off of evidence, it's built off of objectivity, and more importantly, it's built off of a plan.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphor
00:00One of the most important things you can do when you're building an AIOS for yourself or on your clients is to map the systems and what data lives inside those systems, because data forms the basis of everything that you're gonna be doing on your engagement. In this video, I'm gonna show you how to do that ridiculously fast using a Claude skill, and I'll also walk you through why we're doing what we're doing so that you actually learn something.
00:18Let's get into it. So first things first, I want you to run the data map skill and figure out what systems we actually have connected in here and then build me a report around that. So this thing's gonna run through it, and we'll take a look at what it does in just a little bit.
00:31For now, we need to flip across to our slides. So while that thing is cooking, I wanna talk a little bit about the theory behind why we're doing this. So when we have our AI operating system at the top, you can see we have various components that form part of this.
00:42We have the audit, and we have the pod mapping, and we have the data map. These are part of the discovery work that we want to do upfront, whether it's for ourselves or whether it's for our clients. We do this because it helps us uncover all of the specifics that we are going to need for our engagement to make sure that the client gets value, to make sure that you protect yourself with scope boundaries, but also to give you all the information that you need to actually build the system correctly with the right foundations in place.
01:06If you're wondering how to do an audit or what the hell this pod mapping thing is, don't worry. I've got separate videos that I'll link down below that have very deep dives into this exact process, and you really need to understand them because there is a lot of money and a lot of value in understanding them. But particularly for this video, we are focusing on data maps.
01:20So we're inventorying the data, the sources, the flow of the data, if there is currently any, and also the quality of the data inside there. We do all of this stuff first because without it, we are kind of .
01:30We don't really have a view as to what lives inside the systems that they're telling you that they use. A lot of people often think that their data looks like this this merry little house that is already built, but in reality, it actually looks like some kind of . And it's not their fault.
01:43They might not know any better, but why you as the consultant comes in in order to make this a pretty little place for them before you migrate them to an AI operating system. You don't want to because it's gonna complicate any automations that you build. But more importantly, it can also leak information that absolutely doesn't need to be in these systems in the first place.
02:00So our goal here with the order to the pod mapping and then finally this data mapping is that we lead with evidence. Everything that we do for our entire engagement is built off of evidence, it's built off of objectivity, and more importantly, it's built off of a plan.
02:13And this means that we're actually building off of a plan that makes sense and isn't just based on opinion. So our goal here is simple. As a part of our consultation, when we've gone through the audit with our client and asked them about the systems that they use and how they reach out to clients, whatever it is that they're doing, we're going to uncover the various SaaS and solutions that they use in order to achieve this goal.
02:31Once we've done that, it's very simple to get Claude to actually map this stuff for us. We're going to take all of their various systems over here, and we're just gonna head across to our co work session. You can use co work or you can use Claude code, whatever the client might be working in.
02:43Same thing applies for codex here. It's not like this is vendor locked or something like that. And you're gonna come on over to the customize tab over here because this is where we set up our connectors.
02:52The connectors are how we actually do the data mapping. So you see mine are already connected. For those of you who don't know, you would just click on plus and then browse connectors to find any ones that are available publicly.
03:01Otherwise, you can add a custom one. These are usually set up for providers that don't natively have them inside Cowork or Claude, and you can just grab them from the website. For instance, Fathom is one of them.
03:10Beehive is another one. There's a whole bunch of them that don't have native functionality. So you would just add that in here.
03:15Once we've done that, Claude can essentially access all of these systems over here. Important thing to know here is that we absolutely do not want to give Claude any right access at this point. All we want to do is be able to read from these apps to figure out what they have available inside these systems so that we can go out there and pull out the entire information schema so that we can map it as a part of our workflow.
03:34So I think for now, I've got more than enough connected, which means that we can head on back to here, and this thing is still working in the background. While it's running through our data map skill that I kicked off at the beginning of this, it will always ask you for specific things. Can it access this?
03:46Should it be allowed to do this? We'll go through the skill in just a little bit, but for now, we're just gonna hit yes on everything. Most importantly, I'm okay doing this because I've only given it read access.
03:54It can't possibly do something destructive here. The only thing you need to keep in mind is that you definitely wouldn't want to connect this to any type of system that has very sensitive information in it. That's pretty obvious at this point because anything sensitive would obviously be going through Anthropic servers.
04:07Even if you've turned off the training, it is still being processed by an AI model. So keep that in mind depending on how sensitive the data is that you have inside this environment. So you can see we've switched on MCP for all of this stuff.
04:17You can do the same thing from Versus Code or Claude Code and connect it via API for things where an MCP server absolutely didn't exist in any form. So our goal here would be to take all of this scattered information that people think they have or don't even know that they have, filter it through the skill that is currently running, and then it will create some kind of beautiful clarity for us where we have a direct view of the build order that we need to take for our automations alongside the information we gathered from the audit and the POG mapping.
04:42We also have a source of truth, so we no longer need to listen to opinion. We know exactly what data lives in what system and how it is structured. Structured.
04:49We We can can also also check the security scope. So for instance, if they set up a database, whether it's an Airtable or Superbase, whatever their database might be, we can do a security order at the same time over here to check if they've got row level security enabled and a whole bunch of other features protect any of the data stored in there.
05:03It also tells us what types of data we might want to bring in from these outside systems and store in our AI operating system so that we can refine the skills later on, perhaps add context to the system, things that help make our AI operating system way more valuable. It also helps the AI understand the quality of the data inside here.
05:19A lot of people store trash data that they aren't even aware of. So as a part of this, we can do a quality check. We can find out if things are actually worthwhile keeping, and if they aren't, we can remove them before we build any systems.
05:30That way, we have much more efficient systems that don't bring in all the bloat. And then finally, for you, for your scope setting, this will provide that very clear road map alongside the other two parts of the order that you already did so that you have a very cohesive map and you can charge accordingly for your engagements.
05:43And then just one final slide I promised. The reason that I'm bringing this up is because this is gonna be part of a video series that will be very important for you. It's a part of our context journey.
05:51Context forms the entirety of any system now because it's the information that your AI operating system or any AI model actually needs in order to function the way that you or your business might want it to. So our data map is where all of this starts. We need to understand where all of our data lives.
06:06But as we go through this little thing over here, it helps us get to our end goal, which is where we have this context system that can be refined on demand or even some of it automatically. If you wanna see how this works, I have a video that I posted last week, particularly into skill refinement using this methodology that I've outlined on the screen over here.
06:21And then these three other videos in the middle will be covered separately because they are in-depth topics that are worth discussing in their own right. Point is, what we're trying to get to here is a much better AI operating system, one that isn't built on hype, but actual data and knowledge from the people who build it and run it.
06:35Cool. So we're back in Cowork World over here, and you can see that it's been chugging through a few things. Its context window, it has a few of our apps that it's connected to via MCP, and it's currently probing meetings and scheduling connectors.
06:46So it's been through our CRM to check our outreach and things like that. It's probed our knowledge and documents, most likely from Notion and potentially even Slack, and it's gonna do that for everything that we've connected via MCP.
06:57And you can see here our skills in action. So it found a lot of dormant infrastructure. Six of seven super based projects are inactive.
07:04That's because I use them for development and don't really care about them. Several systems your skills reference aren't wired as connected in this session, and that's because I use a lot of them in my Versus Code environment. Point is this thing's doing exactly what it should be doing.
07:16It's understanding what is currently being used and what isn't so that later on when it builds the map for us, we have a much clearer picture of how to proceed. And this also uncovers things for people because they might think that these things are working or connected when they aren't actually working in any meaningful way.
07:29And there we go. This thing's finished. So it gives us a summary about what it found.
07:33It also gives us a complete markdown map on the right hand side here. But, realistically, we need something more visual, so we can just open this. And so if we take a look at our HTML file, you'll see that we have all of our sources listed over here, what's wide, what's not wide, and then we have a very deep understanding of what's actually inside here.
07:49So we could look at our sources on the left hand side. Let's take HayReach for instance over here. We can click on that, and we get a side panel view of what's actually happening inside HayReach.
07:58It gives us a summary of what it understands, LinkedIn outreach automation, one real campaign of substance, leads ad with ten eighty six finished, plus small test runs, basically anything that it carved out of the system so that we can get an understanding of what's in use, what's not in use, and then how we might wanna manipulate it as a part of other automations for this business that we're looking at.
08:17More importantly, like I said, it can do it for databases as well. So if we look at Superbase, we can see here I've only got one live active project, which is the LifeOSDB. This is storing a bunch of information.
08:27It tells me the different components of this database, what it's used for, and then it infers a few things that it might think are useful for us to know, such as the fact that we're storing billing information and things like that inside here. They could be PII that we need to address. Also important that RLS is disabled on all six tables, which is not very good because you obviously want that enabled for your row level security.
08:47It then also tells us paused projects that we have inside here. So, again, if somebody thought they had something enabled but they didn't, this would expose that. You get the point by now.
08:55This thing is mapping all of the data that it finds inside these systems so that you know exactly what you need to know as a part of your engagement. You can feed that back to the client. It will then also mark where it can anything that has PII in it, where there's financial information, which is content related, and which has actually any low value whatsoever inside here.
09:13At the end, it also gives you a summary of everything that it observed. It will also give you some recommendations around those findings that it had so that you can just get a really quick report and piggyback off of any of this information as fast as possible. You can see here a whole bunch of my personal information that, hopefully, my editor is going to blur, but we've got two separate estates.
09:30We've got my business AI OS, and then we've got my personal life OS, both running their separate aspects of whatever it is that I need them to do, and it gives us a little bit of information around the systems involved in making those things work. In the same sense, it would do that for your clients, and you can see here it pulled out the fact that our live databases are not using row level security.
09:48It also mentioned that lead data is fragmented with no system of record. The same prospect domain lives in Apollo and HayReach, and everything else paused. Obviously, these are from old dead POCs.
09:59It then lists anything that it couldn't do or anything that you should confirm with the client or yourself before you go ahead and build the system. That will be down here in the open questionnaire that for things that need to get drilled down into. But other than that, this can be manipulated into any form that you need.
10:12You can get it to audit every single file inside their Google Drive and build you a structure around what's in there, move their stuff around so that it's more organized. You can do the same thing for their databases. So once you gather this information, if something does need to be changed, you could, of course, plan that with Claude.
10:26And as a part of that, you could unleash that plan to build more structure before you go and get on to the automation. That's why it's so important to start with all these foundational audits and scans because they make the rest of your engagement so much easier. So hope this really quick video was helpful.
10:39If you have any comments, leave them down below, and I'll get back to you as soon as possible. Otherwise, check out the videos on the screen now. They'll definitely help you in your journey.
10:45Or you can check out my community where I'm helping people build things with AI every single day. I'll see you guys in
The Hook
The bait, then the rug-pull.
The data map skill is already running before the theory slides appear. The host kicks off the Claude command in the opening seconds, then pivots to explain why while the machine works — a structural choice that creates tension and earns attention for what would otherwise be a dry conceptual argument.
Named ideas worth stealing.
00:27model
Three-Part Discovery Pass
- Audit (stakeholder interviews, opportunity scoring, ROI)
- Pod Map (workflows, manual steps, automation verdicts)
- Data Map (inventory data, sources, flows, quality)
The three pre-build steps before an AI operating system engagement. The data map is step three and the subject of this video.
Steal forclient discovery proposals, engagement scoping documents
04:24list
Six Data Map Outputs
- Build order — what to build first
- Source of truth — what to trust
- Security scope — what to protect
- Data intake — what to capture later
- Context quality — what AI needs refined
- Roadmap and quote — what to charge
The six decision-enabling outputs that a data map produces before any automation is built.
Steal forclient-facing data map deliverable framing
05:45model
Data-to-Context Chain
- Data Map (where data lives)
- Data Intake (capture relevant data)
- Explicit Capture (structured context events)
- Context System (refined, on-demand or automated)
- Better AI OS (evidence-based)
The progression from raw system inventory to a refined AI operating system. The data map is the first required link.
Steal forexplaining why discovery is a prerequisite for AI implementation to skeptical clients
How they asked for the click.
VERBAL ASK
10:10next-video
“Check out the videos on the screen now. They'll definitely help you in your journey. Or you can check out my community where I'm helping people build things with AI every single day.”
Soft end-screen CTA linking to related series videos and the AI Native Skool community. No hard product push.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
Visual structure at a glance.
