Claude Code for Non-Developers: I Built 100 Course Slides

Last updated: June 2026

Quick answer

Claude Code for non developers is now a real design and content production tool, not just a coding assistant. In May 2026 I built nearly 100 professional course slides using Claude Code, with no design skill of my own. The honest caveats: token costs were significant, not every slide was perfect on the first pass, and the workflow required back-and-forth correction. The final output did not look like slop, which is the bar that matters.

TL;DR

I built ~100 course slides in a new visual style using Claude Code, with zero design background. The slides were production quality, not throwaways.
Token usage was heavy. Claude generated HTML, held design memory, and wrote internal progress reports across the run. This is not a free workflow.
The reason it worked is the meta-workflow. I fed Claude my website and screenshots first to anchor a visual style, then iterated on design before any content was added.

Who this is for

This article is for non-engineers using Claude Code for production work that is not "writing software." If you are a course creator, content marketer, founder building lead magnets, analyst producing internal training, or PM making investor decks, this case study is for you. The same workflow that built my slides will build your sales decks, internal documents, and lead magnets.

If you are coming to Claude Code totally fresh, start with the Claude Code tutorial for beginners and come back when the basics feel familiar.

What did I actually build?

In May 2026 I rebuilt the first section of our AI coding course inside Claude Code. The output:

~100 slides in a new visual style
Each slide is HTML, designed by Claude after I anchored the style with my own site as reference
Professional design quality, the kind that does not embarrass me in front of paying students
Built entirely by a self-described "horrible designer" (that is me)

I also built a PDF of the 18 most common AI terms for people moving from Python into AI, as a free gift for YouTube viewers. Same workflow. Same tool. No design skill required on my end.

The reason this matters: Claude Code is being adopted by developers first, but the highest leverage use cases for non-developers are quietly arriving. Slides, PDFs, internal documents, lead magnets. All built by people who would never have opened a design tool a year ago.

The honest version of the workflow

I want to be precise about how this actually worked, because the AI-content space is full of "I built X in 5 minutes" claims that fall apart on inspection.

Step 1: Anchor the visual style before any content

This is the step most people skip and the reason most AI-generated slides look like slop.

I started by sending Claude the aitutorcode.com website and screenshots of pages I liked. Then I had a long, iterative conversation about the visual style I wanted: spacing, color usage, the navy and ochre palette, typography mood, visual density. No content yet. Just style. By the time we had a style I liked, Claude had a clear visual memory of what "looks like aitutorcode" means.

Step 2: Feed the curriculum content in chunks

Only then did I start sending it the actual slide content (topics, bullet points, code examples for the AI coding course). Claude rendered each section into slides using the visual style we had agreed on.

This was the part that took token volume. Claude generated HTML, held design memory across the run, and wrote internal progress reports. The token consumption was, in my words, "incredible amounts of usage." I do not have a clean dollar figure to share because it ran across multiple sessions, but I want to be honest: this is not a cheap workflow.

Step 3: Correct, iterate, do not trust the first pass

Not every slide came out perfect. Some had layout mistakes, a few had factual errors in example code, one section had the wrong slide order. I had to go back and forth with Claude on every cluster.

Claude Code is a collaborator, not a vending machine. The students of mine who get the most out of it treat it that way. The pattern of asking Claude to ask you clarifying questions is what I call meta-prompting and it is the actual skill that matters here.

Step 4: Inspect, ship, accept it is not literally perfect

The final slide deck is good enough that I use it with paying students. It is not the slide deck a $200-per-hour design agency would have produced. It is closer than I expected and the cost was a tiny fraction of an agency engagement.

The bar I held the work against was: does this look like slop? The answer was no. That was the win.

Why is "did not look like slop" the only bar that matters?

AI-generated content has a real problem. Most of it is technically correct and visually dead. Bullet points with stock gradients. PDFs that read like they were assembled by an algorithm. Decks where every slide could be from any company in any industry.

If your AI-generated output is recognizably AI-generated, you have not saved time. You have offloaded your professional reputation to a tool that produces something a discerning viewer will discount.

The slide work I did in May passes the eye test for one reason: I spent the first chunk of the run anchoring style before generating any content. The model knew what "good" looked like for my brand before it started rendering. This is the same pattern I see with my students who use Claude at work. The ones who get useful output are the ones who spend the upfront tokens defining what good looks like.

How does this change Claude Code's audience?

Claude Code was built for developers. That positioning is now incomplete.

What I am seeing in tutoring sessions: non-developers using Claude Code to produce HTML documents, PDFs, slide decks, internal training, and lead magnets. Once Claude can produce styled HTML reliably, it can produce almost any document format a working professional needs.

User type	What they build with Claude Code	Time savings	Skill required
Developer	Code, infrastructure, scripts	30 to 70%	Existing dev knowledge
Content creator	Slides, PDFs, lead magnets, scripts	50 to 80% on visual production	None for design, basic terminal comfort
Analyst/PM	Internal training, dashboards, reports	40 to 60%	Basic Python helps but not required

Time savings figures are rough estimates based on AI Tutor Code's observed student outcomes, not formal benchmarks.

The highest leverage use cases for Claude Code right now are the ones where the user has good taste but no production skill. Course creators, marketers, founders, internal trainers. None of them are developers.

The token-cost reality

Producing ~100 slides this way burned significant tokens. The Claude Code Max plan at $100/month with the doubled 5-hour limits (rolled out May 6, 2026) is the tier that makes this kind of work practical. The era of doing this on a $20 plan is over. I cover the pricing shift in the $20 AI coding tool era piece.

For context, Anthropic's pricing page shows the per-token rates for direct API use. The Pro and Max plans roll those costs into a flat subscription.

The math for content creators: if Claude saves you 20 hours of design or production work in a month, the $100/month Max plan pays back many times over.

What I would do differently

Anchor the style with more reference material, not less. Next time I would send 10 screenshots from sites I admire and a written one-page style guide before any content.
Generate in batches of 10 slides, not long runs. The longer the session, the more design drift creeps in.
Write the full content outline before opening Claude. The split between "what to say" (me) and "how to say it visually" (Claude) is sharper than I gave it credit for.

Common mistakes I see

Skipping the style anchor. People open Claude Code and start asking for slides. The output looks generic because Claude has no anchor for what "good for your brand" means. Fix: spend the first 20% of your tokens on style alone, no content.
Trusting the first pass. Adult learners often assume AI output is take-it-or-leave-it. Claude Code is iterative. Plan for 2 to 4 rounds of correction on every batch.
Underestimating token costs for non-code work. Slide and PDF generation burns more tokens than people expect because the model is producing styled HTML and holding design memory. Budget accordingly. The Max plan exists for a reason.

What to do next

Pick the path that matches what you are trying to build.

If you are a course creator or content producer, the slide workflow above is reproducible. Open Claude Code, anchor your style with reference material, then generate in batches. Expect to spend the first session on style and the next several on content.

If you are a founder or solopreneur, the lead magnet workflow (PDF, one-pager, ebook) is the highest leverage starting point. Same pattern: anchor style first, then generate. I built our 18-AI-terms PDF in a single afternoon.

If you are an analyst or internal trainer, treat your first 20 hours with Claude Code as design investment. After that, you will be producing internal training material in a fraction of the time you used to spend.

If you want to walk through your specific production project with someone who has actually done this work, that is what a single tutoring session can do. Book a free 15-minute Discovery Call.

Frequently Asked Questions

Can a non-developer actually use Claude Code?

Yes, more easily than most people expect. Basic terminal comfort is the only real prerequisite. The interface is text, not code. If you can write a clear paragraph of English, you can drive Claude Code through 80% of non-code use cases.

How many slides can Claude Code produce in a day?

In my run, I averaged 30 to 40 slides per focused day after the style anchor was set. The first day is mostly style work and produces few slides. Days 2 onward are content-heavy and produce more.

What does the token cost actually look like?

I will not give a precise dollar figure because my run spanned multiple sessions on the Max plan, but the practical answer is: the $100/month Claude Code Max plan is the right tier for content-heavy work. Pro at $20 is not enough for runs at this volume.

Is the slide quality really as good as a human designer?

No, and I will not pretend otherwise. A senior designer would do better work. The honest comparison is against the slides I would have made myself, which was nothing useable. Claude Code gets a non-designer to "professional enough for paying students," which is the bar that matters for most non-design-led work.

What is the single highest-leverage tip for this workflow?

Anchor style before content. Spend the first chunk of your run defining what visual quality looks like for your brand using reference material. Every minute spent here saves five minutes on the back end correcting slop.

Does this work for sales decks and investor materials?

Yes, and arguably better than for course slides because investor decks have a clearer visual canon. If you feed Claude a few investor deck references you admire, it can produce a same-tier deck in a single session. The same anchor-then-generate pattern applies.

Ready to put Claude Code to work in your job?

If you are serious about using Claude Code for production work in your role, stop watching demos and start building. Book a free 15-minute Discovery Call and we will walk through the exact workflow for your situation. No pitch, just a conversation about your work.

Book a Free Discovery Call →

Written by AI Tutor Code, private 1-on-1 online tutoring for professionals learning Python, AI, and modern ML tools. 200+ students taught. 3,000+ hours of private tutoring delivered. 4.9/5 average rating.