If you work in communications, PR or marketing, AI is probably already part of your job – but most often in presentations, strategies and “digital transformation” slogans, not in actually taking work off your plate. You know the pattern: another slide about the “artificial intelligence revolution”, and then you go back to Excel, Word and email, because the campaign won’t plan itself and the copy won’t edit itself.
I was in the same place.
I’m not a programmer, I’m a communications person. But at some point the redesign of my website trustedone.pl grew so much that I decided to treat it as an experiment: to see whether AI can be used not only for “writing texts”, but for real support of communication and marketing work. The stuff we do every day: planning, organising, checking quality and verifying facts.
This text is a fragment of that journey. In it, I explain how I built a small “team” out of several AI models – assistants and agents – that help me manage the website, content and background automations. No magic tricks and no promises that “AI will do everything for you”. I have one simple rule: technology is there to take the load off, while decisions and responsibility stay on the human side.
Assistants, agents and a communications person
When I first fired up a “coding” AI in a code editor, it felt like I had let into my backyard a very talented intern who completely doesn’t understand the context. It writes fast, knows every library in the world, but if you don’t explain exactly what you want, it can very confidently do something spectacularly stupid.
Today in this project the “intern” is no longer a single AI, but a small team: assistants and agents who work with me on the website, content and background automations. And I – a communications person, not a programmer – had to learn to manage them the way you manage a real project team.
If you work in PR or marketing, you don’t need your own lab-website to benefit from this. This setup can be translated into everyday work with campaigns, content and reports. Or monitoring trends in the Polish internet. Honestly – do you have time for that? I don’t, which is why we created https://trustedone.pl/chmura/ We don’t judge, we observe.
An assistant is not the same as an agent
To be precise: when I say “assistant”, I mean a tool like a “coding” model in an editor – a general-purpose AI that sits with me in the files, reads the project structure, suggests code, explains errors, asks about my intent. The perfect partner for so-called vibe coding – work where you don’t start from a detailed specification, but from a description of the effect you want to achieve and the “vibe” in which it should work. I explain how it should work and for whom, and the assistant proposes solutions that we then jointly refine into a “ready to use” version.
An agent is a completely different story. An agent doesn’t discuss or philosophise. An agent has one task and is supposed to execute it like an automaton: at a specific time check website performance, once a week clean up unnecessary working files, every day generate a report of what has changed in the project. There’s no “creativity”, but there is repeatability and iron consistency. And that’s exactly what I was missing when the project started to take on a life of its own.
My AI team: lead, builder, critic and fact-checker
In practice, it ended up with me building four main AI “collaborators” for the website:
- “Lead” – plans tasks, breaks them down into smaller steps, describes which files need to be changed and what criteria the final function or module should meet.
- “Builder” – just builds. Based on the plan, it writes code, prepares templates, integrates elements with each other.
- “Critic” – reads texts, looks at logic, catches inconsistencies, repetitions and corporate jargon before anything goes into communication materials.
- “Fact-checker” – trusts no one (including other models and… me). Checks data, dates and quotes in available sources and flags places where things don’t add up.
If you’ve ever worked in a larger PR or marketing team, this setup is familiar: strategist, copywriter, editor and researcher. Here the same thing happens, only part of the work is offloaded to models. And I’m still responsible for meaning, coherence and making sure nothing that comes out from under AI’s “pen” undermines trust in the brand and the project.
The key rule is simple: none of these agents publishes anything on their own. AI can prepare code, a website design, a text draft or an analysis, but the final click – whether in the CMS system or in the mailing tool – is mine. After one spectacular incident with “overzealous” suggestions in report data, I added a hard step to the system: mandatory fact-check and a human decision before every public use.
When an assistant helps, and when it gets in the way
After a few weeks of intensive work I see a clear pattern. The assistant is great wherever something needs to be understood, designed or explained.
For example:
- we break down error messages together,
- we plan the structure of a new module (sections, order, logic),
- we write human-readable documentation for what already works – so that someone outside IT can understand it.
The agent shines wherever the task is boring, repetitive and easy to describe in one sentence: “measure this”, “clean that up”, “save a report”, “run the tests and let me know if anything breaks”. For someone in PR or marketing it’s like the difference between a brainstorming session for a new campaign and sending a newsletter to several segments – one requires your presence, the other can be well automated if we once properly define the rules.
Of course, this is not free magic. Just structuring these roles, writing simple “job descriptions” for assistants and agents with task lists and quality criteria took me several hours. But after a few weeks you can see the effect: some things happen in the background, and I can focus on what I do best – strategy, content, conversations with people. Less digging into technical details, more actual conceptual work.
What’s in it for someone in PR who is “not technical”?
If you’re on the communications side, you may have two natural reactions. First: “this is all too IT-ish, I don’t have time for it”. Second: “if I don’t get a handle on this, I’ll be left behind”. The truth – at least from my experience – is somewhere in the middle.
You don’t have to be a programmer to start using such assistants and agents wisely. But you do have to know your own processes very well:
- know which tasks are repetitive and exhausting,
- which ones require your brain and experience,
- which ones can easily be handed over to an “agent” who doesn’t complain that it’s 6:00 a.m. and once again it has to check metrics or tidy up the materials.
In my case, this project has become a testing ground where I explore how far you can go with automation without losing control over the message and responsibility for the content. And that’s the conclusion that – I hope – may be useful for other communications and marketing people too: AI doesn’t have to be either a “magic black box” or a trendy slide-deck gadget. It can simply be a well-configured back office. And what does it look like in practice? Visit https://trustedone.pl/aktualnosci
What’s next?
The series of texts I’m writing is partly a diary of this journey and partly an invitation for other “humanities people with technical curiosity” who want to use AI for more than just generating another post. In the next pieces I’ll show concrete modules, automations and mistakes I made along the way – with enough detail to draw inspiration for your own projects, but without revealing the entire kitchen.
In the background, a small team of assistants and agents is at work, and in the foreground there is still a communications person making sure that technology stays a tool, not a goal in itself. If this approach resonates with you, stay with this series – the next parts will be more concrete and hands-on.