The central challenge of working with AI

Gábor Pintér — Fri, 06 Nov 2026 14:17:38 +0000

The central challenge of working with AI is speed vs. quality.

Work too slow and your career/business will get behind.
Work too fast and the quality of your work will suffer.

This has always been a dilemma, even before AI. What's new with AI is that this tension is amplified. Now you can both go extremely fast and do extremely shitty work too.

Two steps to fight this: diagnosis and implementation.

First, diagnose what your real problems are.

Is speed truly a problem? Did your employer/client really ask you to speed up 10x? Or is it somewhat an artificial fear fueled by billion dollar companies trying to justify their stock?

The speed problem can be fixed easily. Open ChatGPT, type in your problem and voilà! Now you have a quality problem.

I see quality problem more often than the speed problem. People (including me) started to rely professionally on AI too much. Speed went up in the short term and quality dropped. Eventually, speed went down in the long term too.

This clumsy implementation brought in other factors too. I didn't enjoy coding as much as before. I got distracted more easily. I got more comfortable but less confident. I started to work on multiple problems at the same time and the continuous context switching made me exhausted. Copy-paste more and more.

Eventually I became a meat proxy. A human channel through which AI models communicate.

So what's the fix? How do I increase the quality of my work and enjoy what I do while staying in business?

I believe the keys to a successful shift are gradualism and moderation.

Instead of going from zero-AI to full-AI mode, you take small steps. Instead of jumping from one side of the spectrum (slow but good) to the other extreme (fast but shitty) you navigate towards the middle.

As a developer I introduced AI-free sessions for myself. I even installed a second IDE with no AI integration in it. During these sessions I look around the codebase more, I text-search more, I make mistakes and I spend time staring at loops. I organise the code, separate blocks, and label them. I treat the code as my craft. Like in the old days.

The goal is a more human adaptation, without letting hype and fear override my professional responsibility.

I believe not leveraging AI is a mistake. But letting AI kill your curiosity, your craft, and your confidence is also a mistake. True professionals adapt, but they also reject.

Blog paywall with Astro, Cloudflare, Clerk and Stripe

Gábor Pintér — Mon, 04 May 2026 11:48:29 +0000

If you want to introduce a paywall to your Astro blog - like I do - then Cloudflare Pages, Clerk, and Stripe offer a pretty sweet setup.

In this post I'll show you how you gate some of your content with subscriptions or one-time payments. We'll use Astro for our content, Cloudflare Pages for hosting, Clerk for authentication and Stripe for collecting payments.

Get started with Cloudflare Pages and Clerk

Deploy Astro to Cloudflare Pages. Cloudflare Pages supports SSR for Astro, which means we can evaluate requests on the server side. Deploy directly from GitHub for continuous integration.
Create a free Clerk account. Clerk will handle creating and authenticating your users. Their free tier is extremely generous.
Use Cloudflare's middleware logic to evaluate requests based on URL, e.g.: /blog/exclusive-posts/*
In your middleware.ts, use Clerk to validate user authentication. Redirect to a paywall when the user is not logged in.

// middleware.ts (simplified)
import { defineMiddleware } from "astro:middleware";
import { createClerkClient } from "@clerk/backend";

export const onRequest = defineMiddleware(async (context, next) => {
    const { request, url } = context;
    const pathname = url.pathname;

    // Only gate the exclusive-posts pages.
    if (!pathname.startsWith(EXCLUSIVE_POSTS_PREFIX)) return next();

    const secretKey = getClerkSecretKeyFromEnvironment();
    const publishableKey = getClerkPublishableKeyFromEnvironment();

    const clerkClient = createClerkClient({ secretKey });

    // Get auth state
    let authState;
    try {
        authState = await clerkClient.authenticateRequest(request, {
            publishableKey,
            secretKey,
            acceptsToken: "session_token",
        });
    } catch {
        return new Response(null, { status: 302, headers: { Location: paywallRedirectLocation(pathname) } });
    }

    // Get userId
    const userId = getUserIdFromSessionClaims(authState);
    if (!userId) {
        return new Response(null, { status: 302, headers: { Location: paywallRedirectLocation(pathname) } });
    }

    // User is logged in, continue
    return next();
});

Signing in is only part of the story: for a real paywall you also need to check Clerk user metadata (or equivalent claims) so an active subscription or valid one-time purchase grants access, and redirect everyone else, including logged-in users without entitlement, to your paywall.

At this point you can test this setup manually, without integrating Stripe. Create users manually in the Clerk Dashboard, log in with them and try to access content on your protected URLs. You can even create separate development and production environments in Clerk.

Once you are ready, you can move on to the next phase, which is Stripe integration.

Stripe integration

Create Stripe Payment Links. Payment Links give you pretty sweet customization options, like mandatory billing address fields and custom fields for newsletter subscriptions, for example. I've created a monthly and a yearly subscription, and one for one-time payments too.
Create a Cloudflare Worker that will get called by a Stripe webhook. Place this worker within your Astro repo, so you can deploy it together with the site. The task of this worker is to handle Stripe event notifications and react to them: create a new user in Clerk if payment has been processed and no user is found for the given email address, or update user metadata when subscription changes.
Create a Stripe webhook in the Stripe Dashboard. Provide your Cloudflare Worker URL as the endpoint. Make sure to include the following events:

checkout.session.completed
customer.subscription.created
customer.subscription.updated
customer.subscription.deleted

Webhook signing and secrets

Stripe signs each webhook request with an HMAC of the raw body and a timestamp, and sends the result in the Stripe-Signature header. Your Worker should read the body as text (not parsed JSON first), verify the signature with your endpoint’s signing secret, and only then JSON.parse the payload. If verification fails, respond with 400 and do not update Clerk.

When you add the endpoint in the Stripe Dashboard, Stripe shows a Signing secret (it starts with whsec_). Treat it like a password: store it only in Cloudflare, not in git.

For a Worker in its own folder (for example workers/stripe-webhook/), define the bindings your code expects (STRIPE_WEBHOOK_SECRET, CLERK_SECRET_KEY) and set the values with Wrangler:

# from the worker directory; use your Worker name / env flags as in wrangler.toml
npx wrangler secret put STRIPE_WEBHOOK_SECRET --config wrangler.toml --env prod
npx wrangler secret put CLERK_SECRET_KEY --config wrangler.toml --env prod

Wrangler prompts you to paste each value; it stores them encrypted on Cloudflare’s side. Repeat with --env dev for a development Worker and use a separate Stripe webhook endpoint (or the Stripe CLI forwarding URL) that points at that URL, with its own signing secret.

In my own setup, I also use three metadata fields on Clerk users that store information about their subscription status.

hasOngoingSubscription: boolean
ongoingAccessUntil: Date
oneTimeAccessUntil: Date

hasOngoingSubscription becomes true when they subscribe and switches to false when they cancel. Upon cancellation the ongoingAccessUntil field gets filled with a date. When someone performs a one-time purchase, their oneTimeAccessUntil field gets filled with a date.

As a bonus, you can create a static Thank You page in your Astro blog. Use confetti.js for a ceremonial feel. Use the URL of this page as the redirect URL in your Stripe Payment Links, so your users land on a nice page after successful payments that gives them instructions about the next steps.

This post has been originally published on my blog.

DEV Community: Gábor Pintér

The central challenge of working with AI

Blog paywall with Astro, Cloudflare, Clerk and Stripe

Get started with Cloudflare Pages and Clerk

Stripe integration

Webhook signing and secrets