My Actual 2026 AI Workflow: What I Use, When, and Why the VPN Runs All Day
People ask what my setup actually looks like. Not the aspirational stack — the real one, with the rough edges included.
People ask what my setup actually looks like. Not the aspirational stack — the real one, with the rough edges included.
Here it is.
What’s Running When
| Time | Task | Tool | VPN Required? |
|---|---|---|---|
| 8:00 AM | News and research scan | Perplexity Pro | Yes |
| 8:30 AM | Outline articles, structure ideas | Claude | Yes |
| 10:00 AM | Write drafts, polish prose | ChatGPT | Mostly |
| 12:00 PM | Build internal tools, code scripts | Cursor + Claude API | Yes |
| 2:00 PM | Generate images for articles | Midjourney | Yes |
| 4:00 PM | Screenshot analysis, multimodal tasks | Gemini | Yes |
| 6:00 PM | Final edits, publishing | ChatGPT + Claude | Varies |
The VPN runs from the moment I sit down until I close the laptop. I don’t toggle it on and off. It’s just on. The same way WiFi is on.
8:00 AM — Perplexity for Context
I start the day catching up. What shipped in AI tools overnight. What my regular sources published. Whether anything I was tracking has changed.
Perplexity Pro is better than Google for this because it actually reads the pages rather than just returning them. I’ll ask “what are the significant AI announcements from the last 24 hours?” and get a synthesized summary with citations. Faster than manually reading ten tabs.
Why VPN: Perplexity Pro is region-restricted. The basic version is more accessible, but the Pro features — deeper research, the better model backends, file analysis — require a connection from a supported region. Without the VPN, I’d be using the downgraded version and not getting the same synthesis quality.
8:30 AM — Claude for Structure
Once I have a handle on what’s worth writing about, I start organizing.
My process: I dump everything I’ve collected — research notes, interesting quotes, competitor headlines, half-formed thoughts from the past few days — into a Claude Project. Then I ask it to help me see patterns and structure them into outlines.
Claude’s 200K context window makes this possible in a way that other models can’t match. I’m not summarizing before I paste — I’m pasting everything. An entire research thread, multiple articles, a week of notes. Claude handles the volume.
Why VPN: Claude has the strictest geographic restrictions of anything in my stack. Without the VPN, it simply doesn’t load. There’s no workaround, no alternative with the same context capacity and reasoning precision. This is the most VPN-dependent tool I use.
10:00 AM — ChatGPT for Writing
Claude structures. ChatGPT writes.
I take the outlines from Claude and expand them in ChatGPT. The writing rhythm feels more natural to me — I’ll draft a section, paste it in, say “tighten this, keep the same points,” and the output is usually close to what I want.
I also use it for quick questions while drafting. “Explain TLS 1.3 changes in one sentence.” “Give me three ways to phrase this that don’t sound like marketing copy.” Fast, doesn’t break focus.
Why VPN: ChatGPT doesn’t block consistently, but it blocks unpredictably. On a flagged IP, you get blocks, CAPTCHAs, or throttled responses at random intervals. With a clean IP through the VPN, I don’t think about this at all.
12:00 PM — Cursor for Code
Midday I switch to development work — building internal tools for the site, writing scripts, maintaining infrastructure.
Cursor replaced VS Code for me about eight months ago. The AI autocomplete is fast enough that it doesn’t feel like waiting — it’s predicting as you type, not after. The Composer generates components and routes from plain English descriptions. I spend less time writing boilerplate and more time on the logic that actually matters.
Why VPN: Cursor’s AI features depend on Claude and GPT-4o APIs running behind the scenes. Those APIs have the same geographic restrictions as the standalone services. Without the VPN, Tab completions start failing, Composer throws errors, Chat becomes unreliable. The editor still opens, but the AI — which is the reason to use Cursor — doesn’t work.
2:00 PM — Midjourney for Images
Every article needs a hero image. I’m not a designer, but I need the images to look intentional and professional.
Midjourney is the fastest way I’ve found to get from “I need an image that conveys X” to an image that actually works. Describe what I want in a few sentences, run 4 variations, iterate on the best one. Fifteen minutes for something I’d spend hours hunting for in stock photo libraries — and it doesn’t look like a stock photo.
Why VPN: The web interface and the image editor are geo-restricted. I can use the Discord bot without the VPN, but the bot is slower and doesn’t have the image editing tools. The VPN unlocks the version that’s actually faster to work with.
4:00 PM — Gemini for Multimodal
Late afternoon is for tasks that don’t fit in text. Screenshots of competitor landing pages that I want to analyze. Charts and data visualizations from reports. Anything where the input is visual.
Gemini handles this better than anything else I’ve tried. Upload an image, ask questions about it, get detailed analysis. I use it less than the other tools, but when I need it, there’s no substitute.
Why VPN: Google phases Gemini releases by region. Connect through the US and you get the current model. Connect from certain other regions and you’re on an older version. For multimodal tasks where model quality directly affects the analysis, the difference is real.
The Infrastructure Layer
I tried free VPNs when I first set this workflow up. Then a cheap paid one. Both were problems in different ways.
Free VPNs: their IPs are on every blocklist. Connecting through them sometimes made things worse — Claude blocked me faster on a flagged VPN IP than on my original IP.
Cheap paid VPN: mediocre infrastructure. Speeds were inconsistent. Some servers worked for specific tools, some didn’t. I was debugging the VPN instead of doing work.
The shift was moving to a VPN with enough server depth that there were always clean IPs available: NordVPN. 6,000+ servers, audited no-logs policy, fast enough that AI response times feel normal. The VPN has become invisible, which is what I needed it to be.
What the Workflow Costs Without the VPN
Not theoretical — I tested this for a few days when I was on travel and had a bad VPN setup:
- Perplexity Pro downgraded to basic search. Research takes significantly longer.
- Claude doesn’t load. I lose the tool I use for structuring and long-document analysis.
- ChatGPT throws blocks intermittently. Writing sessions get interrupted.
- Cursor AI features degrade. Development slows down.
- Midjourney web interface unavailable. Image work takes longer through the bot.
- Gemini on an older model. Multimodal quality drops.
That’s not the same job. It’s a version of the job with most of the efficiency stripped out. For someone who does this work professionally, that cost is real and daily.
Setup Notes
Hardware: MacBook Pro. Nothing unusual about the machine matters here.
VPN configuration: connected to a US server, kill switch enabled, split tunneling set up so only browser and work apps route through the tunnel. Banking apps, streaming services, anything that needs my real location stays on the normal connection.
Morning routine: open laptop, click connect in the VPN app, open the tools. Done. The five seconds of clicking connect is the entire setup overhead.
The VPN That Keeps Your AI Workflow Running
Over 6,000 servers in 110+ countries. Clean IPs that bypass AI service blocks. Independently audited no-logs policy. 30-day money-back guarantee.
Get NordVPN — Save up to 72%