# My AI disclosure By Warren Pridgeon | Published 2026-09-23 | Category: Tech Canonical: https://warrenpridgeon.co.za/blog/my-ai-disclosure --- There's currently a divide in the land of the online space. On one side you have the AI Zealots proclaiming "AI does EVERYTHING for me, I've got agents on my PC doing everything and I just check in with them in the morning before I head out to the pool to sip macchiatos and catch some rays". On the polar opposite you have people very much against AI usage, often proclaimed loudly on social media. Somewhere in the middle, where people are just getting on with getting on with it, most people probably exist. Sure, some of you might not use AI at all and some use it a lot... but you probably aren't beating your chest while proclaiming how amazing or abhorrent AI is. A while back I put out a post [who is writing your story](https://warrenpridgeon.co.za/blog/who-is-writing-your-story). Fair's fair, here's how AI is involved in writing mine. And at the end, the same post the way Claude would have written it, so you can see exactly why I rewrite. ## Answer the question before you ask it If you get halfway through this and think "does this clown even write code?!" you'd be asking the right question (and if you've read any of my other blogs you'd know I am basically a clown). The real answer is "Yep, I still write code... a lot of it but I have some help too". The lines of code per hour have always been the most replaceable part of what I do. This point is going to come back around in a bit. If you're here for one specific bit, skip to: [how I use it in the work](#part-one-how-i-use-it-in-the-work) ยท [how I use it in my life](#part-two-how-i-use-it-in-my-life) ยท [how Claude would have written this post](#how-claude-would-have-written-this-post) --- # Part one: how I use it in the work ## Does AI write my code? Some of it, yes. Am I out here going "magic token machine, give me a system and don't make mistakes"? No. Here's what it actually does in my working day: - **Reading what's already there.** Documenting systems that arrive with no manual and one person who "knows how it works" (who resigned), and comparing an in-progress new build against the legacy system it's replacing so nothing gets missed. - **First-draft specs and process flows.** I type out the architecture, the security policies, the actors, the objects, the process flows, the data models. It turns that into a document. This used to take me a week or two to do properly. Now it's a few hours. - **The plumbing.** Basic CRUD, scaffolding, mapping. Required, boring, and worth nothing on its own. - **Finding what's wrong.** Debugging weird scenarios, spotting shortcomings in functions, security and performance, and finding the dead code nobody has called since 2019. Hours of breakpoints, gone. - **"Method A vs Method B" research** when I'm weighing up how to build a new feature. How much of the code is AI involved in? Honestly there's no percentage I can give you. Lines of code has always been a BAD metric for just about anything. If a system has a whole load of customised lookups with screens for each one and tables and and and... that's a LOT of boilerplate/standard CRUD stuff that would be VERY inefficient to hand code... and wouldn't drive much value because it would largely be used during commissioning the system and once every 6 to 12 months to adjust a value in a dropdown. The 100 lines of code I hand wrote for the "business logic engine" are FAR more valuable than the 5000 lines of boilerplate that Claude delivered. The system as a whole delivers huge time saving and value to the customer, even if "Warren only wrote a small number of the total lines of code"... Technically true, but entirely meaningless (unless you're paying per line of code, which is not how I bill). Those 100 lines are the make or break, not the 5000 lines of boilerplate. Choreographing the dance that is software development isn't free either. At this point I see AI as a very enthusiastic junior with all the textbooks lying in a pile in front of it. It will happily disappear down a rabbit hole faster than you can say "I'm late, I'm late, for a very important date", and it'll be pouring tea at the Mad Hatter's table before you've noticed it left. A real chunk of my day is catching it: **Warren:** "EXCUSE ME?! Do NOT do that. What in the what what is going on here?!" (Usually with more colourful language.) At the end of the day I am responsible for what lands up in front of the client, not Claude or Copilot or Codex. I'm the one answering the phone (or the WhatsApp) when something doesn't work as expected. The enthusiastic junior doesn't get the call, Warren does. ## The hour it spent building from a document it never opened Here's the short version, the long one will get its own post. I once spent an hour arguing with Copilot over missing fields on a client's onboarding PDF, only for it to finally admit "AKTUWALLY I can't parse a PDF file". It had read the filename, looked at the headings in my HTML file, and hallucinated a plausible KYC form... while confidently answering me the whole time. *APPLY HEAD TO DESK HARD AND FAST* If I hadn't known that document inside out, those fake fields would have shipped. Keep that in mind for everything that follows. No verify, no ship. ## So what am I actually selling? A couple of weeks back I wrote on OrganiCode's blog about [what happens when everyone in your business is building software and nobody owns the result](https://organicode.co.za/blog/everyone-is-building-software). The same kind of thing exists here but aimed at me. Since I went self-employed ten years ago (two of them in a partnership, eight of them absolutely yolo solo) I haven't been purely "a developer". I've been the wearer of multiple hats, the chief cook and bottle washer if you will. I'm the business analyst, system architect, database admin, front end dev, backend dev, client liaison, accounts department and everything in between. AI has taken a bite out of a few of those hats. I still wear them all. Here's what my job actually consists of now, with the machine's part marked: - **Asking the "dumb" questions nobody else asked.** A couple of years into a client relationship, I finally asked "what do you actually DO with this data?" and that one question [saved thousands of hours](https://warrenpridgeon.co.za/blog/the-ugly-macro-that-saved-thousands-of-hours). AI's part: none. It wasn't in the room and it doesn't know the client. - **Working around a constraint instead of quoting an elaborate fix.** two servers that literally couldn't talk to each other. No integration project, no rebuild, a workaround that shipped. AI's part: none. That was a call about what the client could afford and what would actually get used. - **Knowing what should NOT be built.** [The village post](https://organicode.co.za/blog/everyone-is-building-software) (the OrganiCode blog post mentioned just above here) is entirely this. AI's part: I'll sometimes ask it to argue the other side. It is never the one saying "don't". - **Quoting and drawing up dev plans.** This used to take me a week to do properly: screen mockups, process flows, the lot. AI's part: substantial. This is where the single biggest time saving in my business currently sits. I describe, it drafts, I fix. - **Retiring old systems without breaking the business.** Parallel running, comparing the new build against the old one until you're sure. AI's part: it's pretty good at the comparison and documenting what isn't in the new build yet. Deciding when to cut over, and carrying the can if it's wrong, is mine. - **Owning the outcome, not the fault.** I'm the guy on the phone to Vodacom when their DNS breaks things for clients across the country, when most places go "server's up, your connection isn't our problem" *click*. AI's part: none. It can't phone Vodacom yet. (Nobody can, really. I try anyway.) - **Being there when it breaks.** [1am, eyes burning, system on fire](https://warrenpridgeon.co.za/blog/small-shops-win). AI's part: it's a decent debugging partner at 1am. It is not the one who's awake. - **Building it once and licensing it cheaply instead of quoting bespoke.** [A chiropractor needed forms digitised mid-COVID](https://warrenpridgeon.co.za/blog/eight-years). Building it just for them wasn't going to be worth it for them, so I built it as a product and licensed it at a reasonable monthly cost. AI's part: zero in that decision. Some in the code since. - **Still being there in year eight.** [Every client I have came by word of mouth](https://warrenpridgeon.co.za/blog/eight-years), and many have commissioned me for more work. AI's part: none. You can't automate being there for eight years. That's a bunch of different things. The machine helps inside a couple of them, and does heavy lifting in a couple. There are many aspects to "business" which is outside the realm of machines. You might be running the numbers in your head... less typing, so the rate should come down. That's partially fair, and customers ARE getting more value for money out of it. But I've never billed a customer for lines of code. What I charge for is judgement, getting what you asked for, and someone being accountable when something goes wrong. An AI can't be held accountable... At BEST it will go "you are right to call me out for saying there are 2 r's in strawberry", and confidently reply directly afterwards "there are 2 r's in strawberry" again. It does not and can not care that it's month-end and forty people are waiting on a report that's gone wrong. That isn't the booby prize that comes along with the code. That *IS* the service and the code is the bit that comes along with it. ## Where the line is If I was the one paying me this is what I'd want to know. Here is the real "policy" instead of just going off of vibes. - **Nothing it produces goes out unverified.** When I used it to sort out some tax records (more on that below) it had to put the source file name next to every single line so I could check it. When it builds a form I make sure the form matches the original document. No verify no ship. - **It never gets the final word.** Decisions about what gets built and what doesn't are up to me. It doesn't decide what's right and wrong, because it will always go with the "textbook" answer which is often not what the client wants. - **The words are mine.** Every proposal, every mail, every one of these posts. It can check my spelling. It doesn't get to have my opinions, it simply cannot have the full context of "Warren" to come up with anecdotes and related stories. I ALWAYS put in similes, metaphors, analogies, song lyrics and general dumbassery. - **The relationship doesn't get automated.** I refuse to "automate" stuff like LinkedIn outreach and customer relationship management. I hate talking to chatbots because they often don't get me anywhere meaningful. At some point I'll probably use AI as a first step ticket logging tool that categorises the things but there will always be a "I need to chat to Warren now" ALWAYS. --- # Part two: how I use it in my life The first part was the "how it gets used in my job". The next bit is how I use it on myself. ## The taxes My taxes were behind by a few years. Life happens. (Yes, I'm putting that on the internet. [Broken windows](https://warrenpridgeon.co.za/blog/broken-windows), and all that. I've been fixing mine.) The problem wasn't really the taxes themselves... it was the huge pile of documents: bank statements, homeloan statements, insurance schedules, municipal accounts... years of them. In a drawer and in a bajillion mails. I gave the pile to an LLM and told it to consolidate everything into one schedule. That process involved things such as "rename all the municipal accounts to have the year, month, day as part of the file name. You can find the document date here in the document ..." because they all downloaded with the same name. Then "put this type of data here, this data here and next to EACH line place the source document name" so I could verify everything. Then I handed the consolidated data to a tax practitioner... a real human one, and paid them to do it properly. So the machine did the hunting, gathering and sorting. A professional did the submission. I did the deciding, guiding and paying. I didn't leave it up to the great AI in the sky. ## The health deep-dive I started out chatting to Copilot about my bloodwork, my gym sessions and my eating. It was epic and it really gassed me up "you are performing at an elite level in a caloric deficit and you have made a remarkable metabolic turnaround". But after a short while the conversation got long and it started doing the thing... it started inferring stuff, losing information, confidently making stuff up that wasn't even there to begin with. So I did the Warren thing and [built a platform](https://warrenpridgeon.co.za/blog/unpacking-flexhealth) that holds all the data and sends it, with a set of pre-analysis instructions, to the model every single time. It no longer guesses, because it no longer has to remember (the database does that). This platform has been helping me analyse, adjust and improve in my journey. I was analysing whether my session was effective (it told me to add more leg exercises... I did and I'm seeing individual muscles in my quads for the first time ever). I questioned it as to whether my current fuelling (how much I am eating) was sufficient while hitting 14 to 19 ton volume sessions over an hour and a half. "How are your energy levels? We can see by your sets and reps that you are only really running out of steam right at the last set. You're still good". When I added other data the confidence rose "You're consuming sufficient protein and the latest InBody scans show increasing muscle mass and decreasing body fat. Keep going". That magic token machine can be quite motivational! What it spits out goes to my doctor as things to discuss (it's specifically instructed to say "THIS IS NOT MEDICAL ADVICE NOR DIAGNOSIS"). I have a doctor who likes people doing their own homework. Find one of those. ## The writing Here's the lowdown on how these posts come to be. I'm not dropping the LLM a question like "hey I want a post about orange bicycles with red wheels and what they teach me about B2B marketing when you bunnyhop them". I type out my thoughts, I ramble a LOT. What the LLM is doing is "herding the cats". My head is basically a runaway train of ideas and these posts come out like we're building the track just in front of the train. I braindump all of it into a file and it goes something like this: **Claude:** "Paragraph one and paragraph two need something to link them. Consider writing something about A or B." And then I write that bridge. It checks facts, tone and flow. It then highlights any spelling or grammar errors. Sometimes I leave them in because it suits how I talk. I'm a walking Oxford dictionary and thesaurus but my grammar can be whack. We have the following conversation more often than you'd think: **Claude:** "This sentence is cumbersome and a bit awkward" **Warren:** "I know. So am I. Next." I'm not a fan of using AI for the full-on creative stuff. I have used it for images, because I'm not in a place where I can retain a graphic designer for blog cards and social posts, and I'm not going to pretend otherwise. I do create my own graphics sometimes, sometimes AI does it, for these posts I have templates for the social posts. In the same breath I'm building a game on the side. I could easily AI generate the characters in an afternoon. BUT I'm drawing them myself, and I'll animate them myself afterwards. Slower. Mine. I'm a musician, I'm working on a couple of covers and I'm also writing the soundtrack for the game. SUNO could help out and speed things up a lot... but I feel like creativity gets lost through that. ## The business lens I'm not gonna lie, being a one man army is hard work... I've been doing it for 8 years now. I've been using AI to evaluate my business from the outside, and trying to figure out how to grow. In fact it's how this very site came to be. Every second solo dev is a "generic dev shop" where I'm standing out and going "I am OrganiCode and OrganiCode is me... you're not talking to a faceless drone, you're talking to someone who cares". I've also used AI to compare OrganiCode to competitors and see what they do better vs what I do better. "Hey Claude, compare organicode.co.za to [competitor]. What am I doing well and what am I not?" I've done the same with specific products I'm building, against what's already in the market, to see what I've got going for me and where I'm losing. I see it as a power tool. A good one. It isn't a replacement for a relationship with my clients. ## How Claude would have written this post I told Claude to give this the full all-tropes-ai-treatment. It took off like an excited puppy after a tennis ball.

โš  AI voice mode engaged. This box is a parody. Warren did not write this.

Let's be honest โ€” AI isn't just changing how we work. It's changing who we are when we work. ๐Ÿš€

Here's my confession: I use AI. Every. Single. Day. For code. For taxes. For my health. For this very post. And here's the kicker โ€” I'm not ashamed of it. Because AI isn't a replacement for expertise; it's a multiplier of it.

Let that sink in.

Did AI help me consolidate years of financial records? Yes. Did it transform my health journey from guesswork to data-driven clarity? Absolutely. Did it write my code? Some of it โ€” but here's what most people miss: it's not about the code. It's about the judgement. It's about the accountability. It's about being the one who picks up the phone at 1am. ๐Ÿ“ž

โš  Still the parody. Still not Warren. Keep scrolling for the real one.

Three things I've learned:

1. Transparency isn't optional โ€” it's the foundation of trust.
2. The tool doesn't define the craftsman โ€” the craftsman defines the tool.
3. The future belongs to those who adapt โ€” not those who resist.

The truth? Nobody is coming to take your job. But someone who knows what NOT to build โ€” who owns the outcome, who shows up when it breaks โ€” they're not going anywhere. That's not a threat. That's an opportunity. ๐Ÿ’ก

Read that again.

Agree? Disagree? Drop your thoughts below ๐Ÿ‘‡ and let's start a conversation. #AI #Leadership #Authenticity #GameChanger

โš  End of AI voice mode. Everything above this line was written by Claude, on request, in about four seconds. Everything below it is Warren.

Yerp... that's the AI "slop" we know and hate. The "I didn't put any thought into this and outsourced my thinking completely" vibe. It sprinkled in emoticons, em dashes, machine gun sentences. That's the joke and the whole point of the example. The facts are true but the voice is entirely "not me". It's not "wrong" but it definitely isn't "right" either. ## That's why you get me Does Warren still code? You betcha biscuits I do. Does the Claude bro do everything? Nope, and it would be crazy to let it. It does the donkey work so I don't have to, which means I can deliver more value on the important things. Claude is a lever, the judgement and experience I bring are the fulcrum to create massive leverage (yes... I wrote that all by myself). The phone ringing at 1am and being answered by me and the knowledge I bring: that's the moat... it's not a river of code. That's why you get me.