AI Didn't Give Me Free Time. It Turned Me Into Five People

More than a year ago I built a Python script that summarized my Gmail inbox. Nothing fancy: it pulled unread emails, ran them through an API, and spat out a digest so I didn’t have to open forty tabs before coffee. There was no Big Tech commercial product like it back then, none that I’d found anyway. Today you can get the same thing built into Gmail itself, or pick from a thousand SaaS tools that do it better. But at the time, writing code that actually worked, with no computer science background, no formal developer training, just some old PHP and WordPress scars and enough tech savviness to understand what a function does, felt like discovering a secret door in a house I’d lived in for twenty years.

That door led somewhere I didn’t expect. Seven months ago I shipped TAGiT, a Chrome extension with over 15,000 lines of TypeScript, built with zero prior TypeScript knowledge. Compared to that first Gmail script, TAGiT is a high school senior standing next to a first grader. And now I’m deep into building Qore, a daily moral dilemma game, which is a completely different kind of complexity again.

So yes, AI has changed how I work. But if you’re expecting the version of that story where I tell you I now work three hours a day and spend the rest on a beach, close this tab. That’s not what happened. What actually happened is stranger, more useful, and a lot more tiring than the pitch decks suggest.

The Productivity Myth Nobody Corrects

Here’s the assumption I keep running into, from other freelancers, from clients, from LinkedIn posts with too many emojis: AI does everything, saves you huge amounts of time, and makes you five times more productive. Part of that is true. The mechanism is just not what people picture.

You do not win free time. In my experience you can end up with less of it. What AI actually gives you is the ability to execute tasks that were previously impossible for one person, or that used to require hiring a specialist. It’s less “get your evenings back” and more “clone yourself into five people, each doing a job you never trained for.”

I’m a filmmaker and video producer by trade. Twenty years behind a camera, not a keyboard (well, apart from editing). Right now I can build a full software product by myself, start to finish. In the space of one project I go from director to software developer to product manager to marketer to salesperson to designer, switching hats depending on what the AI and I are working on that hour.

That sounds exciting written down. It is exciting. It’s also the reason my weeks got more hectic, not less.

From Gmail Scripts to Shipped Products

The gap between that Gmail script and where I am now isn’t really about lines of code. It’s about what I was willing to attempt. Back then, “build a tool for my own inbox” was already a stretch. Today I’ll sit down and say “let’s build a branching narrative engine that tracks a user’s moral reasoning across ten philosophical schools,” and treat that as a normal Tuesday. Not because I got smarter about software architecture in some formal sense. Because the tool for turning an idea into working code got radically more capable, and I got better at directing it.

That second part matters more than people give it credit for. Not having a developer background is a real deficit at the start. You don’t have the instinct to evaluate whether what the AI just wrote is actually good, or just plausible-looking. The way around that isn’t a shortcut, it’s just work: training yourself, reading, experimenting, and building the guardrails that catch the AI when it’s confidently wrong. Years of watching tutorials, reading newsletters, and just staying in the loop is what closes that gap, not a single trick.

The Months Nobody Sees

Here’s what the AI-building hype doesn’t tell you: AI accelerates learning and development, but a real product still needs architecture, testing, security, and polish, and none of that arrives pre-solved just because a model wrote your first working version fast. TAGiT didn’t take two days, whatever the highlight-reel demos on social media imply. It took months, and a chunk of those months was weeks of debugging, edge cases I never would have predicted, and production issues that only show up once real users touch the thing.

That’s the part of the story that doesn’t fit in a tweet. The exciting moment, the first time the AI writes something that actually runs, is real and it is genuinely thrilling. But it’s also the easy 20% of the work. The other 80% is the same as it’s always been for anyone shipping software: does it hold up under load, does it fail gracefully, is the data handled safely, does the UI make sense to someone who isn’t you. AI helps with all of that too, but it doesn’t do it for you in the sense of removing you from the loop. It shifts you from writing every line to reviewing, testing, and directing every line, which is a different skill and, in my experience, not a lighter one.

I’ve since layered other systems into my daily routine on top of the product work, including an automated system that gathers and briefs me on my day before I’ve had coffee. I won’t get into how it’s built here, that’s not really the point of this piece, but it exists because the same lesson applied: the first version worked in an afternoon, and then it took considerably longer to make it something I’d actually trust every morning without checking its work.

What “Five People” Actually Feels Like

Take a solo developer running one application, which is where I am with Qore right now. On any given day I need to hold in my head:

• development and architecture decisions
• the product itself, what it should become
• distribution
• marketing
• content
• app design
• legal terms
• SEO
• and increasingly, GEO, making sure AI answer engines can find and represent the product correctly

Two years ago that list belonged to ten different people in ten different departments. Now it’s a list I run through myself, with AI agents doing a lot of the heavy lifting on execution, but the thinking, the deciding, the noticing when something’s off, still sits with me. By default that’s too much to hold at once. It doesn’t feel like freedom, even though it technically is more capability than I’ve ever had. It feels like watching your own spider web expand outward in every direction at once, and knowing you’re still the one who has to feel every vibration on every strand.

You cannot be equally good at everything, even with AI doing most of the manual labor. That’s the honest trade-off nobody selling AI productivity wants to put on the slide. It’s a genuine compromise: you get to explore, build fast, and produce things that would have taken a team and a budget I never had, almost as fast as you can think them. But it’s not the “go to the beach while the AI runs your business” fantasy. We’re not there. The tools are powerful and still unpredictable enough that they need real human effort, testing, and correction to work reliably. It’s not intuitive for the average person to walk in and immediately extract the best version of what these tools can do.

I think we’re heading toward one of two outcomes. Either we end up in a two-speed world (like I previously predicted), where a smaller group of people who’ve put in the hours to actually direct AI well provide that leverage as a service to everyone else, or the technology plateaus enough that the average user catches up on their own. But underneath both of those sits a bigger question I don’t think gets asked enough: who actually controls the AI. Right now that’s a handful of corporations, not any of us, and I have real doubts that where they take it serves the people using it. Concentrated power has a track record, and it isn’t generosity. For the near future, I think the two-speed world holds regardless, the skill gap and comprehension gap make sure of that. The people who’ve put in the hours can still fight their way onto the first wagon. Whether that stays possible longer term is a different question, one that depends less on individual effort and more on who owns the wagon.

The Part AI Still Can’t Do For Me: Getting Noticed

If you ask me what’s hardest, even with every AI tool at my disposal, it’s not writing code and it never was. It’s distribution and marketing. The barrier to actually building a product has collapsed. Anyone can build now. Which means the entire competitive advantage shifted to who can get their thing in front of people and make them care, and that’s still mostly a human skill.

Here’s a concrete version of that problem: outreach. AI makes cold email trivially easy to generate at scale, for me and for literally everyone else building something right now. Which means the person on the receiving end of your “personalized” outreach is getting ten times more of it than they used to, and it all reads the same. The result is that generic AI-assisted outreach is becoming less effective the more people use it, precisely because it’s so easy to produce. The human touch, actual personalization, visible proof that you did your homework on this specific person, is worth more now than it was three years ago. That’s the opposite of what the hype implies.

Where AI genuinely helps on the distribution side is the operational layer underneath the strategy: automating social posting, running A/B tests, doing insight analysis, surfacing leads worth a look. It removes the grunt work. It does not replace the judgment call of what to say and to whom, and that gap is what needs to be filled.

Think about it from the other side of the inbox for a second. If cold outreach takes one person an hour to write by hand, and now takes an AI thirty seconds to generate at scale, everyone building something is going to lean on that thirty seconds. Which means every inbox, including yours, is filling up with messages that all sound competent and all sound the same. The tools that were supposed to make outreach more efficient are quietly making it less effective per message, because the scarcity shifted. It’s not attention that’s scarce anymore, it’s a message that actually proves someone paid attention to you specifically. That’s a strange inversion: the more powerful the AI tools get for everyone, the more the advantage moves back toward old-fashioned human specificity.

Building Qore: A Different Kind of Test

TAGiT taught me how to ship a technical product. Qore is teaching me something else. It’s a daily moral dilemma game: users get a scenario, pick from three choices, and their decisions branch out from there. Over time, their pattern of choices builds a thinking profile that maps them to one of ten philosophical schools of thought. They can play with friends and compare how differently everyone reasons through the same dilemma.

My background is philosophy-adjacent by curiosity, not by coding for branching narrative logic. And that’s exactly where AI makes mistakes I can’t always catch by feel. A multi-step, branched-out story has a lot of surface area for the logic to drift, a choice that should lead somewhere coherent instead loops back into nonsense, or a branch that technically works but reads as flat and lifeless. A human eye still has to review that for clarity and coherence. Where it gets interesting is that you can build a second layer of AI agents whose only job is to verify the first layer’s output against the rules and guardrails you’ve defined, so the system checks and corrects itself before it ever reaches me for review. That’s not “AI replacing my judgment.” It’s AI catching AI, with my judgment still setting the rules both layers have to follow.

Every product I’ve built has made the next one faster to reason about, even when the product itself doesn’t pan out the way I first imagined. That’s been true from the Gmail script to TAGiT to Qore, and I expect it to keep being true for whatever I build next.

What I’d Actually Tell Another Freelancer

If you’re a solo operator, not an enterprise with a department for every function, here’s the honest version: AI will let you attempt things that used to require a team you couldn’t afford. It will not hand you spare hours. It will hand you more roles, more decisions, and more surface area to manage, and it will wear you down in a different way than a heavy shoot week used to. The productivity is real. The relief is not automatic, you have to build it deliberately, through the guardrails, the review habits, and the judgment calls that AI still can’t make for you.

I still don’t know if Qore will find its audience the way I hope. I didn’t know that with the other tools either, and I definitely didn’t know it with TAGiT six months before launch. What I do know is that I’ll come out the other side of this one with more experience directing these tools than I had going in, and that compounding is the actual return on investment here, more than any single product’s outcome.

So the real question isn’t whether AI makes you more productive. It does. The question worth sitting with is whether you’re ready for what that productivity actually costs you, and what you’re going to do with five roles’ worth of capability once you have it. If you’re navigating something similar, or want to talk through how this applies to a video or content project, reach out, or browse more of what I’ve been building and breaking over on the blog.