What I'd Tell Someone Starting With AI Tools Today

A friend asked me last week what AI tools they should start with. Not a developer friend — a person who runs a small business, does their own marketing, handles their own operations, and has heard enough about AI to know they should be using something but not enough to know what. They wanted the honest version, not the Twitter version. Here's what I told them, expanded into something useful for anyone in the same position.

Start With One Tool, Not Five

The single biggest mistake people make when starting with AI tools is subscribing to everything at once. They sign up for ChatGPT Plus, Claude Pro, Midjourney, Jasper, and some AI email writing tool — all in the same week. Total cost: $80-100/month. Total usage after 30 days: ChatGPT occasionally, everything else forgotten. I've seen this pattern dozens of times, and it happens because the AI tool landscape is overwhelming by design. Every tool's marketing tells you it's essential. None of them are.

Pick one general-purpose LLM. Just one. Use it for a month. Learn its strengths, discover its weaknesses, build habits around it. After a month, you'll know enough to evaluate whether you need something else — and what specifically that something else should do.

If you're asking me which one, I'd say Claude if your work is primarily text-heavy — writing, analysis, coding, reasoning. I'd say ChatGPT if you want the broadest feature set — web browsing, image generation, code interpreter, plugins — in a single subscription. Both are $20/month. Both are good. The difference matters less than the habit of using either one consistently.

What I would not recommend as your first AI tool: any specialized tool. Don't start with Midjourney for images, Jasper for marketing copy, or an AI video generator. Specialized tools are for people who've already identified a specific gap in their workflow and need a specific solution. If you're just starting, you don't know what your gaps are yet. The general-purpose tool will show you — because the places where it's inadequate are the places where a specialized tool might actually be worth the money.

Learn to Prompt Before You Learn to Pick Tools

The skill that matters most in the first month isn't tool selection. It's prompting. And by "prompting" I don't mean the mystical "prompt engineering" that has its own cottage industry of courses and certifications. I mean the basic skill of telling the AI what you actually want, with enough specificity that it can give you something useful.

Most people's first prompts look like this: "Write me an email to a client." That's the equivalent of walking into a restaurant and saying "bring me food." You'll get something. It probably won't be what you wanted. A better prompt: "Write a follow-up email to a client named Sarah who I met at a conference last week. We discussed a potential partnership around content distribution. I want to suggest a 30-minute call next week to explore the idea. Tone should be professional but warm, not salesy. Keep it under 150 words."

The difference between those two prompts is specificity — and specificity is the single skill that separates people who find AI tools useful from people who find them disappointing. Every piece of context you add — who the audience is, what the goal is, what tone you want, how long the output should be, what to include and what to avoid — makes the output meaningfully better.

Here's the mental model that helped me most when I was learning: treat the AI like a smart employee who just started today. They're capable and willing, but they know nothing about your context. Everything you'd tell a new employee before asking them to do a task — the background, the constraints, the preferences, the definition of "done" — you need to tell the AI. The people who get the worst results are the ones who treat AI like a mind reader. The ones who get the best results are the ones who treat it like a competent stranger who needs a good brief.

Three specific prompting habits that make the biggest difference for beginners.

First, include examples. "Write a subject line for a newsletter" produces generic output. "Write a subject line for a newsletter. Here are three previous subject lines that performed well: [examples]. Match that style." produces output that's calibrated to your audience and voice. Examples are the highest-leverage input you can provide.

Second, specify the format. "Summarize this article" produces a block of text that may or may not be the length or structure you wanted. "Summarize this article in 3 bullet points, each one sentence, focusing on the practical implications" produces exactly what you can use. The format instruction takes five seconds to add and saves five minutes of reformatting.

Third, iterate. Your first response from an AI tool is almost never the final product. Treat it as a first draft and give feedback. "Make the tone more conversational." "Cut the second paragraph — it's redundant." "The opening is too generic — start with the specific problem instead." Each round of feedback gets you closer to what you actually want, and the AI remembers the conversation context, so each revision builds on the last.

The $0 Stack

Before you spend any money, use the free tiers. Claude has a free tier. ChatGPT has a free tier. Perplexity has a free tier. Google's Gemini has a free tier. Use all of them — not simultaneously, but sequentially. Spend a week on each one, doing the same types of tasks. By the end of the month, you'll know which interface you prefer, which model gives you the best results for your specific work, and whether the paid tier is worth upgrading to.

The free tiers have real limitations — usage caps, older models, fewer features. But for someone who's starting out, those limitations won't matter for at least the first few weeks. You'll hit the usage cap before you hit the quality ceiling, and that's actually useful information: if you're hitting the free tier's limits, you're using the tool enough that paying makes sense. If you never hit the limits, you don't need the paid version.

The $0 stack for a month of exploration: Claude free (for writing, analysis, and reasoning), ChatGPT free (for general tasks, web browsing when available, and image generation via DALL-E), and Perplexity free (for research with citations). Total cost: nothing. Total capability: enough to evaluate whether AI tools are useful for your specific work.

What to Automate and What to Keep Manual

The automation question comes after the habit is established — usually month two or three. By then, you've identified the tasks where AI saves you time and the tasks where it doesn't. The next step is figuring out which of the time-saving tasks are repetitive enough to automate.

The rule is simple: automate what you've already done manually at least four times. Not once. Not "I think I'll need to do this regularly." Four actual times. This filter eliminates the cathedral trap — the tendency to build elaborate automation systems for tasks you'll do twice and forget about. If you've done something four times, you know the task well enough to automate it correctly, and you know you'll use the automation.

For non-technical users, the automation path starts with the AI tool's built-in features. ChatGPT's custom GPTs, Claude's Projects, Perplexity's Collections — these are lightweight automation that saves your context, preferences, and instructions so you don't have to repeat them every time. This is where most people should start and where many people should stop. The value of saved context is enormous and the implementation cost is zero.

If you're more technical, or if your automation needs outgrow the built-in features, that's when tools like Zapier or Make enter the picture. Not before. And if you're significantly technical, n8n self-hosted. But these are month-three-or-later decisions, not week-one decisions.

The Subscriptions to Avoid

Some specific advice on what not to buy when you're starting out.

Don't buy any tool that markets itself primarily as an "AI writing assistant" with its own interface. Jasper, Copy.ai, Writesonic — these tools wrap the same underlying LLMs (GPT, Claude) in a specialized interface with templates and workflows. The templates are sometimes useful, but the underlying capability is identical to what you get from the LLM directly. You're paying a premium for a coat of paint. Use the LLM directly and write your own prompts — you'll learn more and spend less.

Don't buy any "AI course" before you've spent at least a month using the free tools. The best AI education is experimentation, not instruction. After a month of daily use, you'll know what you actually need to learn — and most of it will be specific to your use case, which no generic course covers well.

Don't buy annual subscriptions until you've used the monthly plan for at least three months. The annual discount is a commitment that assumes you'll still be using the tool in a year. The AI tool landscape changes fast enough that a tool you love in March might be leapfrogged by June. Pay monthly until you're sure, then lock in the annual rate.

And don't subscribe to more than two paid AI tools at the same time until you've been using AI daily for at least three months. Two is enough to cover your needs while you're learning. More than two creates subscription sprawl, decision fatigue about which tool to use for each task, and a monthly bill that makes you feel committed even when you're not getting value.

The Timeline

Week 1-4: Use one free-tier LLM for everything you can think of. Learn to prompt. Discover what it's good at and bad at for your specific work.

Month 2: If you've hit the free tier limits, upgrade to the paid version. Start exploring one additional tool for the specific gap your first tool doesn't cover well.

Month 3: Evaluate whether you need automation beyond the built-in features. Set up saved contexts (custom GPTs, Claude Projects) for your recurring tasks. Consider one specialized tool if you've identified a specific need.

Month 4+: You now know enough to make informed decisions. Add tools based on demonstrated need, not marketing. Cancel anything you haven't used in 30 days.

That's the boring version. It takes four months to build a useful AI tool stack. Anyone telling you they can get you set up in a weekend is selling you something. The setup isn't the hard part — the habit is. And habits take time.

This article is part of the Weekly Drop at CustomClanker — one take, every week, no fluff.