I did not set out to be an 'agent person'
Six months ago I would have rolled my eyes at the phrase 'agentic workflow.' It sounded like crypto people had moved to AI. But I kept running into the same problem: there were tasks in my workflow that were too small to delegate to a human and too repetitive to do myself. Competitor research. SEO audits. Generating blog post drafts. Monitoring design trends. Cross-posting content.
These are not design tasks. But they eat into design time. A typical week for me included maybe 40% actual design work and 60% everything around it — research, communication, documentation, marketing. I needed a way to automate the surrounding 60% so I could focus on the core 40%.
That is when I started experimenting with agent platforms. Not because I wanted to be on the cutting edge. Because I was tired of doing the same boring things every week.
What agent platforms actually are (plain version)
Forget the buzzwords for a second. An agent platform is a system where you give an AI a goal — like 'research competitor pricing and create a summary' — and it figures out the steps, executes them, and returns a result. It is different from asking ChatGPT a question. ChatGPT gives you one response. An agent can search the web, read multiple pages, compare data, write a report, and save it to your notes — all without you guiding each step.
The three I use: Hermes is my main agent — the one I am talking to right now. It lives on my server and has access to my files, my terminal, my browser. I use it for complex tasks that need multiple steps: auditing my website's SEO, checking performance, writing and deploying code changes. n8n is my automation layer — it connects services together. When a new blog post goes live, n8n triggers an agent to generate social media posts, update my sitemap, and notify me. I also use agent features built into OpenCode and Claude for development-specific workflows.
The key word here is delegation. I do not tell the agent 'fix this button.' I say 'audit my portfolio site for performance issues, identify the top three problems, and implement the fixes.' The agent decides how to approach that. That is what makes it an agent and not just a chatbot.
The workflows I actually run
Competitor and market research used to take me half a day. Now I give an agent a domain — say, 'research travel voucher platforms in the Middle East' — and it searches, reads competitor sites, pulls pricing, analyses UX patterns, and writes a summary. I review the summary, verify the key claims, and use it as a starting point. The agent does not replace my analysis. It replaces the grunt work of opening 40 tabs.
SEO is another one. My agent checks my site weekly: broken links, missing meta tags, performance regressions, keyword gaps. It produces a report with specific fixes. Some fixes it applies directly — like updating meta descriptions or compressing images. Others it flags for me to review. This used to be a monthly manual audit that I often skipped. Now it happens automatically.
Content drafting. I write all my articles myself — the voice, the stories, the opinions. But the agent handles research synthesis. I feed it my notes, it organises them into themes, pulls relevant quotes from my past work, and suggests a structure. I still write every sentence. But the blank page is less blank when I start.
The boring stuff: generating image alt text across my entire site. Converting design specs into structured data. Checking Figma files for component consistency. These are things nobody wants to do, and they are exactly the kind of tasks agents excel at.
The learning curve nobody warns you about
Agent platforms are not consumer products. Setting up Hermes required configuring a server, understanding environment variables, setting up model providers, and debugging cryptic error messages. n8n's visual workflow builder is easier to grasp, but chaining 5 services together still requires thinking like a programmer — conditionals, error handling, data transformation.
The first weekend I set up Hermes, I spent six hours just getting it to connect to a model provider. The documentation assumed I knew what a gateway was. I did not. I Googled. I broke things. I fixed them. By Sunday evening it was working, and that felt genuinely satisfying — but I would not recommend this to someone who hates technology.
The tools are improving fast. Six months from now, setting up an agent might be as simple as installing an app. Today, it still requires patience and a willingness to read error logs. If you are a designer who gets frustrated when Figma plugins do not install correctly, wait another year before diving into agents.
What I would tell a designer considering this
Start with one workflow. Do not try to automate everything at once. Pick the single most annoying repetitive task in your week and ask: could an AI agent do 80% of this? If yes, set it up. Live with it for a month. Then add the next one.
Keep your expectations realistic. An agent will not understand your brand voice without explicit instructions. It will make mistakes. It will occasionally go down a bizarre rabbit hole — I once asked my agent to research 'design trends' and it spent 20 minutes reading a blog about interior decoration. You need to supervise.
But when it works, it really works. Last month my agent caught a broken image on my portfolio, fixed the file, updated all the references, and pushed the change — all while I was sleeping. I woke up to a commit message I did not write, on a fix I did not request, and the site was better for it. That moment felt less like using a tool and more like having a very junior but very diligent teammate.