What Actually Happens When You Build AI Systems With No Tech Background and No Team

My first attempt took eleven days and produced nothing I could use.

No team. No technical background. Thirty years of running a business by myself and a stack of evenings I had decided to spend on this. Eleven days later I had a folder of half-finished attempts and a fairly strong suspicion that this was for other people.

It was not. But nothing I had read prepared me for what the process actually feels like, so this is the version I wish someone had written for me. What really happens, what it costs, what breaks, and when you should not bother.

Week one: it feels like you are doing it wrong

You open the tool. You describe what you want. What comes back is close enough to be maddening and wrong enough to be useless. So you adjust and try again. And again.

That loop can go on for days, and while you are in it, the story you tell yourself is that you are not technical enough. That is the wrong diagnosis, and it is the one that makes most people stop.

What is actually happening is simpler. You are trying to hand off a job you have never described.

You know how to do the task. You have done it four hundred times. But knowing how to do something and being able to explain it step by step are two completely different skills, and only one of them transfers to an AI. This is the same reason hiring a virtual assistant fails for people with no processes written down. The help was never the problem. The handoff was.

The turn: forty boring minutes with a pen

Mine came when I closed the laptop and wrote the task out longhand.

Every step. In plain sentences. The way I would explain it to a person on their first day, including the small things I do without thinking. Where the information comes from. What I check. What I would never do. What "finished" looks like.
It took about forty minutes and it was genuinely dull. It was also the whole fix. When I typed that description in, the output was usable on the first try, and I had spent eleven days failing to get there by adjusting.

That is the piece almost nobody says out loud. The work of building an AI system is mostly the work of writing down what you already know.

What it actually costs

Money. Less than people expect, and I want to be specific rather than reassuring.

Claude's free plan is $0 and it is a real plan, not a demo. Claude Pro is $20 a month, or $17 a month if you pay for the year at $200 up front. Claude Max starts at $100 a month and no solopreneur reading this needs it. ChatGPT and Gemini sit in the same neighborhood on their paid consumer tiers.

And there is a good chance you are already paying for one without knowing it. Google folded Gemini into Google Workspace plans and dropped the separate AI add-on charge back in January 2025. If you pay for Workspace for your business email, you already own an AI. You have just never been introduced.

So the honest money answer is that you can build your first system for nothing, and the paid tier is worth it only once you know what you need it for.

Time. This is the real cost and I am not going to soften it.

Figure roughly three to five hours to get your first system genuinely working, spread across a few sittings. Most of that is not spent with the AI. It is spent writing down the task. After that, running the system takes minutes, and building the second one is much faster because you have already learned how to describe a job.

What you give up. A few evenings. That is the actual price, and for a solopreneur those evenings are not free.

What breaks, and what you should expect

You will build the wrong system first. Almost everyone does, and I did too. You pick the interesting task rather than the draining one. The interesting one is more fun to build and it changes nothing about your week. The draining one is the one that pays you back.

It will state things confidently that are not true. Anything with a number, a price, a date, or a name in it gets checked by you before it goes anywhere. Every time, no exceptions. That habit is part of the system, not a sign the system is broken.
It will write longer than you want. Cutting becomes a real part of the work. Budget for it.

Nobody is coming to check on you. This is the part that is genuinely harder alone. There is no colleague to say "that's not quite it," and no one to notice when you have been stuck on the same step for three days. Working without a sounding board is the actual disadvantage of doing this solo, and it is bigger than any technical gap.

The thing to do this week

One step. You can do it tonight with a notebook and no software open.

Think back over last week. Not the biggest task. Not the most important one. The one you did the most times. Answering the same kind of email. Writing the same kind of post. Pulling the same numbers into the same summary. Whatever you repeated most, that is your first system.

Then write out that task's steps in plain sentences, in order, the way you would explain it to someone on their first day. Include the parts you think are obvious. Especially those.

When you have that page, you have the hard part done. Paste it into whichever AI you already have access to, tell it that this is the job and you want it done this way, and give it one real example to work from.

If you cannot write the steps down, the AI cannot do the job. That is not a technology problem and it is not a you problem. It is just the order the work has to go in.

When you should not do this

I would rather tell you than have you waste a week.

If the task happens twice a year, leave it alone. Systems earn their keep through repetition. A task you do quarterly is not worth three hours of describing.

If the task requires your judgment at every step, leave it alone too. Pricing a nonstandard job. Deciding whether to keep a difficult client. Anything where the whole value is you thinking. AI can prepare the ground for those decisions. It should not make them.

If you are in the middle of something that needs all of you, a launch, a family situation, a busy season, this is not the week. The evenings this takes are real evenings.

The tool was never the hard part. Deciding what it should do was.

If you write that one page this week, I would like to hear which task you picked.

Sources

Claude plans and pricing, claude.com/pricing: Free $0, Pro $20 per month or $17 per month billed annually at $200, Max starting at $100 per month. Checked September 5, 2026.

Constellation Research, on Google dropping the Gemini add-on charge and folding Gemini into Workspace plans (announced January 2025): https://www.constellationr.com/blog-news/insights/google-workspace-drops-gemini-add-charge-raises-business-enterprise-plan-prices