AI is scary good - How everything has changed in 1 year đź’€
Exactly a year ago, I watched our engineering team sweat over an AI voice calling module. It took them a solid month/month and a half of sprints, tickets, and endless standups to ship a very basic, clunky version. It was great at the time, not belittling their work.
Cut to this week. I needed a similar module for a completely new use case. The tech team was crying about bandwidth, and the existing system had issues so I decided to take matters into my own hands and made one using Claude Code.
I am a pure non-engineer. I don't code. But I built the whole damn thing in literally 3-4 days.
And here’s the wildest part - it’s actually much much better than what our devs built last year. It handles human interruptions really well, has crazy good noise cancellation, and the overall voice experience is just smooth and works for Indian accents well too.
You rarely get to see a direct like-for-like comparison like this in tech. 1 month by a team of actual SDEs vs 3 days by a non-techie using an AI agent. It has completely blew my frickin mind.
The world has fundamentally changed in just 12 months. If you are just writing boilerplate code, taking a fat CTC and calling yourself a software engineer, I genuinely don't know how you survive the next 2 years.
PS: None of this is because of my skills, except maybe I knew what to build and had a vision for the product. Most of this is advancements in models. But still, I feel unchained!
One interview, 1000+ job opportunities
Take a 10-min AI interview to qualify for numerous real jobs auto-matched to your profile 🔑
Just for once deploy that code to production, many non-engineers are building POCs using AI and telling devs are slower, just please deploy that code to production, take full ownership including integrating with existing systems and don't ask anything from devs. After doing the actual deployment, please let me know your experience of AI
Ohh wao in tcs a non techincal guy is doing development, and also bro if you are non technical then how you have identified the integration part if you do not know the codebase🙂 how you have integrate you new module to existing codebase is claude doing integration so and if claude is doing this then is tcs really allowing any AI to have access to codebase.
I understand why you, as a non technical person, think that this is incredible. And in many ways it is.
But coding isn't difficult when you do it at a small scale with single user.
What you're missing is -
- It will likely fail in weird, unexpected, irrecoverable ways when you scale up to more users
- It will be extremely difficult to make any changes a year from now
- When things do break at scale, you'll have no real chance at fixing it in a way that won't just make it break soon again
Your like-for-like comparison is bullshit.
Think if it like this, when a team of devs get a product call saying the company with 100 restaurants wants to start selling bread. Developers build an entire system to source flour and yeast. To stabilize costs of raw materials when markets fluctuate. Then they find ways to repeatedly produce the exact same bread in all 100 restaurants. They need an entire infrastructure to do the accounting for the costs, to track how many loaves are made and are selling, etc.
All you've done is bought a bread machine and made one singular loaf of bread. Now admittedly the bread machine is very useful, the actual process of baking the bread is easier and faster for sure.
But this is why your comparison is bullshit. You don't have what it takes to run a restaurant chain just because you made one decent loaf of bread.
Engineering is and will remain a high learning curve skill. Especially as more and more new engineers lack depth and more people like you build slop we need to fix.
đź’Ż
the thing you built isn't just a module, it's a receipt. it shows the value has shifted from 'can you code it?' to 'can you specify it?'. your non-engineer brain just out-specified an entire engineering sprint from last year. this isn't about your tech team being slow. it's about your product sense now having a compiler.
Who are you? What are you here for?
AI writes code for sure which works for the given use case. But in production different kinds of people use different functions in different scenarios and different environments it needs very conscious handeling which AI cannot comprehend. AI writes beginning code with methods or controllers but there is lot to development beyond that. AI is really useful in saving time with routine requirements and companies behind them are over promoting the products because the capital investments on AI is extremely high and unless every business adapts AI they are far behind making profits. For sure AI is extremely good and useful for some use cases but the narrative that it is replacement to Humans is just AI companies strategy for funds. It's very difficult to reduce hallucinations in AI as the context in a complex flow grows and grows we know where to stop looking where to go deep but even the most advanced models like Opus 4.6 hallucinate a lot if the context grows. Use AI to fasten but look underneath the hood!!
If I only had a virtual knife or something 🤣
please, pretty please... deploy the code in productionđź«Ł Do the full cycle as SDE's not just code, that's boring now, isn't it?
đź« đź« đź«
I have an app idea in mind which is a subscription based service. Can I make this on Claude?
Yup, you can. But if you're planning to generate revenue from it and make it public, then it's advisable to have it reviewed by experienced software engineers at different stages: Architecture, development, testing, security, deployment and then later on scaling and maintenance.