Posters without a designer
Fill in a spreadsheet, drag in the photos, download print-ready PDFs.
I'm Guus Witjes. I use AI to build small helpers for everyday work: posters, explainer videos, research and writing. You don't need a big project or a technical background: you start by describing what you want in plain language. This is what changed.
Ten examples from my own work. Each one started with a task that cost too much time. Click an example to see how it works.
Making things
Fill in a spreadsheet, drag in the photos, download print-ready PDFs.
AI writes the script, draws the animation and adds a voice.
AI cuts on the pauses, adds captions and puts each graphic on the right word.
Finding things out
AI reads public sources and each organisation's own plans, and ranks who fits best. A person decides at every step.
It runs on my own laptop, so nothing goes to the cloud.
The lesson: anyone can make something now. Knowing whether it is right is the hard part.
Writing
I write the argument. AI turns it into the rest, I check it and post.
A custom ChatGPT that asks who the text is for, checks it and rewrites it.
Preparing and organising
From the first mail to the follow-up a month later. AI prepares, I check.
My notes, rules and examples in one place, so AI works with my knowledge instead of guessing.
And just for fun
My laptop picks a new photo at every login: snow in winter, green in spring, lakes in summer, red leaves in autumn. A small script, nothing more.
All the photos and clips from a phone become one film on music. The maps are drawn on my laptop, so nothing about the trip goes online.
You start with a task, not with technology. You describe it in plain language, the way you would explain it to a new colleague.
An example of how a build starts. The first version is rarely perfect: you try it on real work and ask for changes, as you would with a colleague.
A task you repeat every week, that costs time and has a clear result.
Your rules, your examples, what good looks like. That is what makes AI good at your work.
Ask for something small and test it on real work the same day.
Say what is wrong, the way you would to a colleague, and use it again next week.
In a workshop with Guus Witjes you pick one task from your own work and leave with a working AI helper for it. No technical background needed.
Get in touchNo. You describe what you want in plain language, in ChatGPT, Claude or Gemini. What you do need is a clear task and the knowledge to judge whether the result is right. That knowledge is yours, not the AI's, and it is the part that makes the result good.
It ranges from an hour to a few weeks. The personas took about an hour once the research was in, and the explainer video a few hours. The partner scan took under 3 hours of my own time. Bigger helpers, like the workshop preparation, grow over weeks of real use.
You decide what goes in. Two of these examples, the meeting transcription and the poster tool, run entirely on a laptop, so nothing goes online. For everything else: use the AI tools your organisation allows, and keep personal data out of what you type.
Yes. In my workshops you build your first AI helper on a task from your own work, using the four steps on this page. You leave with something that works, plus the method to build the next one yourself.
More builds
Problem. A separate project app became a second source of truth next to my files.
A local dashboard on top of my own files: tasks, content, workshops and hours. It writes back to the files, so there is one source.
Result. 43 open tasks moved out of the old app. Orphaned notes, files nothing linked to, went from 628 to 120.
Built with Claude Code and Node.
Problem. Teams ask "can AI do this for us?" and there was no repeatable route from that question to a working system.
Eight phases with a template each, from intake through four working documents (workflow, knowledge, rules, brief) to build, test and handover.
Built with templates and Claude Code.
Problem. Most of the tokens in my AI work went to re-reading the same context.
An audit that shows where your tokens go, plus hooks that save on them automatically.
Built as a Claude Code plugin with Python.
Problem. Slides drift off-brand as soon as someone starts moving boxes around.
A slide editor where every slide stays on-brand, because content, layout and style are kept apart.
Built with Claude Code, Next.js and TypeScript.
Problem. Websites show up in Google but not in the answers of ChatGPT and Perplexity.
Checks a website step by step for AI search: a quick scan, a checklist, robots.txt and structured data.
Built as a custom GPT. Live and in use.
Problem. Your best knowledge sits in your head, where an AI assistant cannot use it.
Copy-paste prompts that get your knowledge, judgement and voice out of your head and into instructions for an AI assistant. I use them in workshops.
Built as plain prompts. Works in ChatGPT, Claude and Gemini.
Turns a training request into a session design with learning goals, timings and a price estimate.
Built as a Gemini Gem and a Claude Project.
A five-minute daily news read in four sections, with an angle for a post on the best items.
Built as a Claude Project.
Scans a text for banned words and typical AI sentence patterns, reports every match and gives a score out of 50.
Claude Code skill.
Scores a draft against the original brief and improves it round by round until the score stops rising.
Claude Code skill.
From goal to slide plan to a finished PowerPoint on a corporate template, with quotes, images, steps and big numbers instead of a stack of bullet slides.
Claude Code skill, python-pptx.
Turns a PDF into an e-book that reads well on a Kindle, and checks every figure on a contact sheet.
Claude Code skill, Python.
Writes five alternative openings for a LinkedIn post, each from a different angle.
Claude Code skill.
The full skill library
Next to the builds above, the library holds 87 more skills. Each skill does one job, and Claude Code loads it only when that job comes up.
A recruitment team makes its own posters in the house style, without waiting for a designer. One HTML file that runs offline in the browser.
AI finds the organisations in a region that fit best, reads their own plans and turns the best ones into a card per organisation. A human sets the criteria and decides at every pause point.
Example from one region.
Four rules: public sources only, a source for every claim, an explanation for every score, no personal data.
An outdated explainer video, rebuilt with AI: script, animation, voice and MP4, without an agency.
One chain from the first request to the follow-up a month later. I lead where it counts; AI prepares the rest and I check it.
One argument per week becomes a full week of content. AI writes, I review and post by hand.
What step 4 makes from one hub, every week:
Three helpers that share one set of tools: one for videos of someone talking to camera, one for screen recordings and one that cuts long recordings into short clips.