I’m a Java developer. At my day job, we don’t use AI tools. But outside of work, AI has become part of most of the things I do.
Here’s my actual setup — not theory, not sponsored content, just what I genuinely use every day.
Writing and research — Claude
When I started AI Baltics, I had a research problem. The EU AI Act is a 144-page legal document. National AI laws across Estonia, Latvia, and Lithuania are mostly in the local languages — Latvian on likumi.lv, Estonian on riigiteataja.ee, Lithuanian on lrv.lt. Almost no English-language summary exists comparing all three.
I could have spent six weekends translating, cross-referencing, and synthesising. Instead I worked with Claude as a research partner. Long conversations where I’d paste in raw legal text in Latvian, ask Claude to extract the substantive provisions, then push back when something looked wrong. Then verify against the original. Then move to the next document.
The pattern that emerged: Claude is excellent for structuring research, terrible as a source of research. Every claim needed verification against the primary text. But the synthesis — turning ten hours of legal-text reading into a clean side-by-side comparison — went from “I’ll never finish this” to “this is shippable in two evenings.”
Same pattern for blog drafts. I write the structure and the angle myself. I give Claude my rough notes and ask for a first pass. Then I rewrite about 60-70% of what comes back, because Claude’s voice isn’t mine. The output isn’t AI writing — it’s me writing faster, with fewer staring-at-blank-page moments.
One specific thing I never let Claude do alone: write the opening paragraph. The opening sets voice, and voice is the thing AI gets least right. I always write the opener myself. Claude takes over from paragraph two.
Building the AI Baltics website — Claude Code
The infrastructure for AI Baltics runs on two Oracle Cloud free-tier ARM instances: one hosting WordPress, one hosting Mailcow for self-hosted email. Setting that up by hand would have been a weekend of work — installing Nginx, configuring PHP-FPM, hardening SSH, getting MariaDB tuned, configuring SPF/DKIM/DMARC for the email server, and so on.
Instead, Claude Code connected to the servers via SSH and ran the whole sequence under my supervision. I’d describe what I wanted (“install WordPress with Nginx, PHP 8.2, Redis object cache, configure for the aibaltics.com domain”). Claude Code would propose the steps, I’d approve, it would run the commands. When something failed — and things failed plenty — it would diagnose, propose a fix, and continue.
The honest version: I held the architecture in my head. I knew what I wanted Nginx to do, what the security baseline should be, what the WordPress configuration looked like. Claude Code wrote the executable layer.
Specific moments where this saved real time: getting Mailcow’s mail-tester score above 9/10 took maybe 45 minutes of iterative DNS tweaking, with Claude Code reading the test output and proposing the next adjustment. By hand I’d have spent half a day Googling DKIM record formats.
What it didn’t replace: deciding what to host where, the security model, and which trade-offs were acceptable. Those are still my calls. Claude Code executes the calls; I make them.
P.S. DNS related things I still needed to configure myself. Claude could suggest what to do, but I needed to do that by myself anyway as he doesn’t obviously have access to my domain registrant.
Business cards — Claude + a Riga print shop
I needed business cards for AI Baltics. Latvian print shops have specific requirements: CMYK colour profile, 2mm bleed, 85×55mm, text converted to outlines (so the printer doesn’t need my fonts), no trim marks on the final file. None of which I knew before starting. As I said, I’m a developer — and a bit of a Creator now — but I’m not a designer.
The whole process — from “I want business cards” to “here’s a print-ready PDF” — happened in a single Claude conversation over maybe 90 minutes. I described the brand identity (teal primary, slate text, the AI Baltics logo). Claude generated the design as SVG, exported as CMYK PDF, applied the bleed, outlined the text. I sent the PDF to a print shop here in Riga (no sponsored content here, as you remember), they confirmed it passed their pre-flight checks, and the cards came back perfect. 300g Invercote, matte anti-scratch laminate.
What surprised me: the friction was almost entirely in the print specifications, not the design. I knew what I wanted the cards to look like. Translating that into “what does a Latvian print shop actually need to receive” was the part that would have taken hours of back-and-forth with a designer or print broker. Claude handled that translation in minutes.
This is also the moment when I started taking AI tooling seriously beyond writing. If it can take me from concept to print-ready file in one conversation, what else can it remove friction from?
Presentations — Claude
The webinar slide deck was generated programmatically. 20 branded slides with icons, tables, timeline graphics, and consistent design — all from a single conversation. I submitted the content as text, Claude produced a PPT file as output. A couple of tweaks and it was ready.
The friction it removed: I’m comfortable writing the content of a presentation. I’m slow at making it look professional — choosing fonts, aligning icons, building consistent timeline graphics across 20 slides. The deck would have taken me six hours by hand. With Claude, it took ninety minutes including the tweaks.
Email drafting — Claude
I drafted the emails to the print shop in Latvian, but I had Claude help with the technical terminology and tone. My day-to-day Latvian is conversational; the specialised vocabulary for print specifications was outside it. Claude handled the precise terms while I kept editorial control over the tone.
What AI can’t do (yet)
Where I still do everything manually: actual coding decisions at work, visual design judgment that requires taste rather than execution, relationship building, strategic decisions about AI Baltics.
The pattern is clear — AI doesn’t replace my work, it removes the friction around it. The creative thinking, the judgment, the taste — that’s still human. But the execution? AI makes it 10x faster.
Want practical AI content for the Baltics? Subscribe to AI Baltics — short, weekly issues, no spam.
Webinar on 7 May: AI Laws in the Baltics — what changes on 2 August 2026 and what to do about it. Free, live on YouTube. Register here