Praveen Deshpande

Just a guy trying to make sense of things as he goes.

AI as a Sidekick and Devs Need to Stay in the Driver's Seat

I've been seeing a lot of reels on my YouTube feed about "Will AI replace human Devs." My opinion based on my own experience: it is all about control.

Where AI Actually Makes Sense

MVPs, PoCs, front-end experiments. Go ahead and vibecode. No real users, no revenue, something breaks and nobody cares. Output is all that matters here and AI gets you there fast. Completely valid.

Where You Need to Be Careful

The moment you are building a serious enterprise grade application with a genuine user base, AI is your sidekick. Devs are still the protagonists.

You need to know what every piece of your code does. Not because AI cannot write it but because when something breaks in production you need to know exactly where to look. Hand that control over entirely and the moment AI hits a wall or produces something wrong your whole team slows down hard. Nobody knows the codebase anymore. Debugging becomes a guessing game.

Vendor Lock-in But Make It AI

Same way we talk about cloud vendor lock-in the same thing is starting to happen with AI tools. Your workflows, your rules, your context and habits all get built around one specific tool or model. Switching becomes painful. You are dependent before you even realise it.

Cloud lock-in, AI lock-in. Same story.

These Pricing Plans Are All Over the Place

I paid for a plan and could not use it properly because changes were happening in the background. Turns out this is not just me. Major AI coding tools have flipped their pricing structures multiple times in under a year. Flat subscriptions to token-based billing, usage caps introduced overnight, new sign-ups paused without much warning. Developers running heavy sessions are seeing their costs jump significantly with no heads up.

Maybe this settles down as things mature. But right now you are building your entire workflow on a foundation that keeps shifting under you. That is a real risk.

The Server Load Problem Nobody Thinks About

When your tool calls out to an AI vendor server that server is dealing with traffic unlike anything traditional infrastructure handled before. With ecommerce you can anticipate spikes. A sale, a launch, a seasonal rush. With AI tools you genuinely cannot. A viral moment, a new model drop, a pricing change that triggers a surge and suddenly the servers are under pressure nobody saw coming.

When that happens you get throttled. Rate limits kick in, usage windows shrink and your team hits a wall mid-workflow. If you do not know your own codebase at that point you are stuck. The vendor is down and so are you.

Stay in control. AI is a great sidekick. Just do not hand it the wheel.


What are your thoughts? Let me know: reach out.

Opinion, AI

⬅ Previous post
Taming Messy RSS Feeds with n8n and a Bit of Vibecoding

Next post ➡
K8s Log #1 - Setting Up a Local Cluster and Surviving the Confusion

Subscribe for more!

You don't have to keep coming back here to read my latest posts. Subscribe via RSS to get updates whenever I publish something new.