DEV Community: Nikita Veras

The Real Price of AI

Nikita Veras — Thu, 12 Mar 2026 05:53:07 +0000

First time i actually hit the limits of my Cursor $20/month plan. So I decided to check how much I’d been using it, and I noticed that the request history actually shows the price of each query.

Then I found my most expensive conversation — 4.13 $. I was like what the hell. I honestly didn’t realize Opus 4.6 was that expensive. And it doesn’t even say that it’s the high-thinking mode — I was just selecting the “thinking” one.

Honestly, I feel like once the marketing hype around AI cools down, subscriptions are going to get way more expensive than they are now. Models will get smarter, they’ll spend more compute “thinking”, and companies will eventually need to start making their investment back.

And at that point, $200 for Claude probably won’t even sound that crazy.

What’s the most expensive AI conversation you’ve ever had? 👀

Cursor basically rebuilt MCP.

Nikita Veras — Fri, 20 Feb 2026 07:23:11 +0000

Cursor just rolled out an update with plugins.

To me, it looks like MCP — but in a new wrapper, and honestly, it feels much better. Before, the AI would decide on its own what to call, which endpoints to hit, what actions to trigger. Now you can explicitly tell it what to do and get a much more predictable result.

The most useful plugins for me so far are Figma and Context7. The latter lets you research library docs — which I’ve been doing a lot lately.

OpenClaw another AI agent 😅

Nikita Veras — Tue, 17 Feb 2026 09:23:00 +0000

This tool made quite a buzz online — to the point where Mac Mini sales supposedly spiked. So what is it?

Basically, it’s “hands” for your AI — a way for it to interact with the outside world: chat apps, notes, calendars, and a bunch of other tools. The integrations list is pretty impressive — 50+ and counting.

The main selling point? It can run 24/7 and supposedly doesn’t lose context — meaning it “remembers” everything.

But here’s the catch: it doesn’t ship with its own model. You have to plug in your own — either a local model or an API key for something like GPT or Claude.

I’m a bit skeptical for one simple reason:
If it runs 24/7, you basically have to validate its output 24/7.

That said, it could be really cool for automating repetitive workflows.

For example, imagine a calendar analyzer running all the time, checking your schedule and suggesting:
“Hey, move this meeting — you’re already drained at that time.”
(You’re a zoomer after all 😄)

And of course, the security question doesn’t magically disappear. Giving an always-on agent access to sensitive data is… a big deal.

Have you had any real experience with agents — not just AI chat that edits your files, but something that actually acts on your behalf?