Inside the AI Transformation: A Senior Engineer's Brutal Truth About the Future of Coding Jobs

Jemin Patel, who started at Amazon in 2016 pre-AI era, shares unfiltered insights on how AI is transforming software engineering. From 30% AI-written code to changing hiring practices - the uncomfortable truths about engineering careers.

July 22, 2025

5 min read

By Rachit Magon

Computer science employment has dropped 26.5% - the steepest decline since the 1980s, before the internet existed. Google's chief scientist admits AI already writes 25% of their code. Yet most conversations about AI's impact on engineering jobs happen in boardrooms, not with the engineers living through this transformation.

Today, we're diving deep into the uncomfortable reality with someone who's witnessed it firsthand: Jemin Patel, a senior software engineer who started as an Amazon intern in 2016 when AI coding assistants were science fiction. Over nearly a decade at Amazon, Sprinkler, and multiple startups, he's watched the profession evolve in real-time.

His perspective? "You don't need an army of people anymore - you need an army of agents." But he's not pessimistic about the future. He's excited. Here's why, and what it means for every engineer navigating this transition.

Key Takeaways: The New Reality of Engineering Careers

The Brutal Truth:

• 30% of code is now AI-generated, even at smaller companies
• Companies aren't hiring junior developers - only experienced ones
• Traditional career progression (intern → junior → senior) is breaking down
• Interview processes will shift from DSA to "can you work with AI agents?"

Immediate Adaptation Strategies:

• Master prompt engineering and AI tool orchestration
• Learn to break down complex problems for AI consumption
• Build domain knowledge alongside technical skills
• Develop model fine-tuning and deployment capabilities

The New Career Path:

• College grads: Can you work with agents?
• Senior engineers: Can you build and orchestrate that army of agents?
• Future unicorns: Single developers managing AI agent swarms

Q: You've been in tech for nearly a decade, from the pre-AI era to now. Give us the real picture - how has your day-to-day work changed?

Jemin: I started my journey as a software engineer at Amazon, and throughout this decade I've learned a lot. The first two companies - Amazon and Sprinkler - gave me holistic experience. I learned how to write better code, scale products, different technologies.

Then my journey started toward startups. In those initial companies I was a backend engineer, but at startups you have to wear multiple hats - frontend, backend, DevOps. The major thing I learned was handling infrastructure. I converted monolithic applications into microservice-based architecture, learned GCP and AWS, creating Kubernetes clusters.

Now after the startup experience, you can throw me an idea and I can build an entire startup or product from scratch, end to end.

But nowadays things are different. You don't need an army of people - you need an army of agents that can build a product for you. There could be a solo founder, a unicorn founder with just one developer. That could be a thing in the future. I'm really excited for the upcoming days in this AI wave.

🔥 ChaiNet's Hot Take: This evolution from backend specialist to full-stack to "AI orchestra conductor" represents the new career trajectory. The engineers thriving aren't just coding - they're architecting systems where AI agents do the implementation while humans provide strategic direction.

Q: Computer science employment has dropped 26.5% - the lowest since the 1980s. Google admits AI writes 25% of their code. Are we in denial about how fast AI is displacing us?

Jemin: No, I'm with that statement. It is moving fast. Even in my current days, 30% of code is written by AI. But it's not like I'm blindly following AI code - I need to spend more time reviewing because you have to be very careful. It's not going to work all the time.

My time has increased on reviewing code and having discussions with people. The percentage of AI-written code is increasing day by day. I can say 30% of code is written by AI in my day-to-day life currently.

🔥 ChaiNet's Hot Take: The 30% figure aligns with industry reports, but Jemin's emphasis on review time is crucial. AI isn't just replacing coding time - it's shifting the skill focus from writing to understanding, reviewing, and orchestrating. This creates a higher bar for engineers, not a lower one.

Q: If 30% of code is AI-written, does that mean you need 30% fewer developers?

Jemin: Even six months or a year back, we had projections - we need this many people to deliver these features within this timebound. Now things are getting changed. With fewer developers, you can deliver the same amount of output within the given time range.

So we do not hire fresh people - we just hire experienced people, because that's what they bring: experience and quality code. That's our strategy nowadays. The headcount is getting decreased day by day as models get evolved, and that count is going to increase.

But people have to focus on how you can develop your own model, how to tune it based on your requirements. Those capabilities people should know to save themselves.

🔥 ChaiNet's Hot Take: This hiring shift is the most concerning trend for junior developers. The traditional entry path is disappearing. Companies are essentially saying: "We'll take the productivity gains from AI, but we still need humans who can provide oversight and strategic thinking - and those humans need experience."

Q: If companies aren't hiring junior developers, how does anyone become a senior engineer anymore?

Jemin: Things are changing. One year back, junior expectation was: write an API that does this task. If you could handle it in one or two days, that was fine - basic expectation from fresh grads.

Nowadays, those things get done within 1-2 hours. You just need to test it. Even junior or senior engineers can do it within 1-2 hours. So obviously people won't expect you to do the same task in one or two days.

Those graduating recently or in upcoming years have to increase their expectations. Companies are expecting more from you. You have to learn something else - not just coding. They may ask you to develop this feature within these many days, and you should learn either to develop code yourself or how to write better prompts to ChatGPT that can give you code within 1-2 hours.

You have to be a prompt generalist - know how to communicate with the model in a better way so it gives better output. In coming days, in interviews people may ask: "You have to do this assignment within 1-2 hours. You can use anything - Cursor, ChatGPT, whatever. You'll be given an IDE, just finish this task within the timebound."

The interview process may change. Things are changing, so you have to adapt and learn.

🔥 ChaiNet's Hot Take: This is a fundamental shift in what "junior developer" means. The new baseline isn't "can you code?" but "can you effectively direct AI to solve complex problems?" It's like asking someone to be a manager before they've been an individual contributor.

Q: Will competitive programming and DSA interviews become irrelevant?

Jemin: That will change, but not drastically. Companies still have to adapt their interview processes. Eventually it will change, but in upcoming days they'll ask whether you can work with agents or not. If you can work with an army of agents, then we can hire you.

The expectation from a college grad would be: can you work with agents? The expectation from a senior engineer would be: can you build that army?

🔥 ChaiNet's Hot Take: This reframes the entire skill hierarchy. Instead of junior→senior being about code complexity, it becomes about AI orchestration complexity. Juniors manage individual agents, seniors design and coordinate agent swarms.

Q: AI tools have evolved from writing boilerplate to optimizing database queries and architecting microservices. When you look at your job, could AI do 70% of it?

Jemin: All these models work with context knowledge. Gemini 2.5 Pro has 1 million tokens as context - you can give 1 million pieces of information and get answers. But not all models work at that level. Claude has 200k tokens, ChatGPT has lower.

Even bigger companies can't fit all their data in a single prompt and get perfect output. Having holistic knowledge gives better answers than just chunks of knowledge.

What I'm doing - you can't give the entire company's code repo knowledge to the model right now. You have to break down your problem. If you have an entire feature, break it into smaller chunks. For individual breakdown tasks, you can give context about specific files and ask for development.

You have to learn how to ask questions to LLMs with the given context knowledge. I'm not seeing more than 70-80% of tasks done by AI because I'm not just coding - I need to manage people, be involved in architecture discussions, guide people.

Previously, maybe one year back, I had to manage both coding and meetings. Nowadays, with AI help, I can write code in less time. It's helping me write code while also helping team members make architectural decisions.

Code review is something I believe LLMs haven't reached the mark yet. I still have to be involved in all code reviews thoroughly, even though we have Copilot as a code reviewer. I'm not getting expected review comments - that's still yet to come.

🔥 ChaiNet's Hot Take: The context limitation is crucial - it's why senior engineers remain valuable. They can break down complex systems into AI-digestible chunks and synthesize results back into coherent architecture. This "decomposition and recomposition" skill becomes critical.

Q: You mentioned it's like that movie I, Robot where you had to ask the right question. Is prompt engineering the new core skill?

Jemin: Yes, exactly. I've been involved with Scale AI - one of my friends was working on training models, and I know how they train models with millions of tokens. They have procedures, SOPs on how to train models.

Based on that journey, you understand how LLMs work and how we need to ask them questions. Right now, what we know works best is: give the holistic view of the entire feature, ask for a roadmap. It will give you step-by-step procedures, then go deeper into individual steps.

That's how you should ask LLM questions to get the right answers. First create a roadmap, then go deeper one by one into each feature.

🔥 ChaiNet's Hot Take: This reveals why prompt engineering isn't just about crafting clever queries - it's about understanding information architecture and problem decomposition. It's closer to systems design than creative writing.

Q: You've been remote working since before it was cool. How do you compare remote vs. office work, especially as companies push return-to-office?

Jemin: In remote work, trust is very important. It's all about how transparent you are with your management - how open to the team. Even if you're facing an issue, stuck somewhere, it's better to communicate openly: "Hey, I'm facing this issue, I'm stuck at this point." So they trust you.

Building trust is very important in remote work. If they trust you, you can work from anywhere. If they don't trust you, they'll call you to the office.

Getting a remote job initially is very difficult because nobody will trust you. But after you work with a remote company, the next company becomes easier because you already have that trust-building process.

Communication is very important. You have to communicate on Slack or messages - "this is what I have done, this is what I'm going to do next." Whenever you face an issue, don't spend too much time researching. Speak up: "I'm facing this issue and doing this research to solve it." You may get help.

Nobody will reach out to you because you're working remotely. Nobody is watching you. You should know how to unblock yourself, how to communicate clearly with the team.

Another thing - even if I go to an office job, I can get better pay, stocks and everything. But now I feel the trade-off between social life and the amount of money you're making. At a certain point, social life is important because that's something I'm losing.

When I was at Amazon, I was commuting more than two hours in the Amazon cab. I was spending 60 hours in a month just on commuting. I don't want to do that. 60 hours with my family will be much more rewarding than urban commuting.

🔥 ChaiNet's Hot Take: This economic analysis is powerful - 60 hours monthly commute equals 720 hours yearly, or 30 full days. Remote work's value proposition becomes even stronger as AI handles more routine tasks, making location-independent deep work more valuable.

Q: Traditional career paths (intern→junior→senior→lead→principal) are changing. What are you optimizing for in your career now?

Jemin: That's changed drastically. Initially, I was thinking about why companies value senior people most - because they have context knowledge. If you join Amazon 10 years back and stay 10 years, that's huge for Amazon because you have context knowledge: how the codebase works, how things are changing.

Context knowledge is very important. If you want to make yourself valuable, you should know about the company more - domain knowledge, not just coding. You should know the business value. Domain knowledge will make you different among other people.

Now things are changing. Previously you were working with people, now you need to know how to work with agents. You need to ramp up yourself - how to deal with multiple agents, do multiple things at a time. You should know both technically and managerially.

In technical terms, you need to know how to create your own model. Obviously you can use base models like Gemini, but you should know how to create your own model on top of it, how to deploy those models, how to create your own agent that people can use.

Previously you were simply coding based on business requirements. Now you have to think ahead: how can I automate this particular business activity through AI?

Now you need to change your technical vision - how to deploy your own model, create your own model. Maybe if you have a very good idea, you can create your own company as well. Just a single person in a company can do anything with AI - you have coding person, marketing person, SEO experts, everything.

I was using operative.sh - you just give an idea, it can create the entire website for you. It can do testing live in front of you. It can open the URL, test for you, analyze logs, check logs: "This is the error log, this is the landing error I'm getting, let's fix it." Just give the prompt and it will do it. A single prompt will give you the entire website.

🔥 ChaiNet's Hot Take: This shift from "domain expertise + coding" to "domain expertise + AI orchestration + model creation" represents a fundamental reframing of senior engineering roles. The most valuable engineers will be those who can bridge business needs with AI capabilities.

Q: Do you believe we'll see single-person unicorns soon?

Jemin: I'm saying it is possible. If you already have entrepreneur experience - sometimes as an engineer-entrepreneur you know how to build a product, but when you need to sell it, marketing becomes difficult for engineers. You need marketing expertise.

But now domain knowledge is there. People can use LLMs to get marketing strategy as well. You can hire people, and there are agents available with expertise in marketing and sales domains. I think you can build a single-person unicorn.

🔥 ChaiNet's Hot Take: This isn't just about technical feasibility - it's about the democratization of capabilities that were previously team-dependent. The limiting factor shifts from implementation capacity to strategic vision and execution excellence.

Q: Do indie projects still make sense when AI agents can do everything?

Jemin: Yeah, you can still do it. As I said, if you want to survive in this upcoming AI wave and you have an idea, you should start building on top of it. Ultimately, you are solving people's problems.

If you're genuinely solving people's problems, people will pay for you. It's not like the problem is already solved - it's how optimally or efficiently you're bringing the solution. People will reward that particular solution - cost effective, time effective, whatever.

Even if the problem is already solved but you have a better way to solve it, go for it. Build it. People will love it. People will pay for it.

So even if you have an idea, just go for it. Just build it. You don't need years of time to build the product now. You just need a few months to build the product.

🔥 ChaiNet's Hot Take: The reduced development time creates more opportunities for experimentation. Instead of betting years on one idea, you can test multiple concepts in months. The winners will be those who iterate fastest and understand user problems deeply.

Q: What's the behind-the-scenes conversation about hiring in this new AI world?

Jemin: It's not like "if you can work with AI, we'll pay you more." Even existing companies are having internal training modules - they train internal employees how to use LLMs. They give assignments: "finish this assignment using LLMs."

Good companies are investing in their employees with these courses. But obviously, if they have very junior-level developers who are just doing piece-of-code work that's being done by AI agents, their job is uncertain, to be honest.

For example, Facebook has billion-dollar revenue. Even one line of code can impact that billion-dollar revenue. That's why they hired thousands of junior-level employees who could do one line of coding and save their billion dollars. But now things are changing - they can solve these problems internally with LLMs, so they don't need those people.

Obviously, layoffs will happen in that scenario. But those who are long in a company, have good domain knowledge, and that particular vertical is performing well - that job is secure.

In upcoming days, there won't be junior developers. There will be an army of agents. You need to deal with that. You need to know how to handle that army of agents, not people.

Those days are coming. Be prepared so you won't be surprised. Don't stay away from AI agents - you need to get deep into it and understand what's actually happening and what's going to come.

🔥 ChaiNet's Hot Take: This frank assessment reveals the strategy: companies will invest in retraining valuable employees while cutting roles that AI can replicate. The message is clear - become valuable enough to retrain, or become vulnerable to replacement.

Q: If you could give advice to your past self before the AI revolution, what would it be?

Jemin: I think I have friends who are already in ML. I learned a few courses in advanced machine learning, but we didn't choose that particular path - we went into normal software engineering careers.

But if you could have gone into that particular domain - I don't know, Facebook might reach out saying "we're giving you a $10 million signing bonus." People are there like Perplexity's founder - he just passed out in 2017 from IIT Madras, worked with top-notch companies like Google and OpenAI, learned how to create LLMs. Now he has Perplexity.

Even Google is scratching their head: "What should I do?" They're in panic mode. Google's going to be replaced very fast. They're in pressure. Even big companies are under pressure about how to go with that particular speed so nobody can overtake them.

It's not like we're just seeing movies. We're seeing how they're responding to particular market situations. But they're very sure.

Yeah, Perplexity is getting their own browser. You just give them a prompt and it can order from Amazon. You ask: "Based on my chart history, what do you think I'm deficient in terms of minerals?" It can give answers. Then you ask: "Which is the best medicine for this supplement?" It can find those, then "Can you order for me on Amazon?" It's going to order for you.

People are heading toward that complexity. It's a paid version, but people will pay those who want to save time.

🔥 ChaiNet's Hot Take: This regret about not pursuing ML earlier reveals the opportunity cost of the current transformation. The founders and engineers who built the AI systems disrupting everyone else positioned themselves at the center of the revolution rather than on its periphery.

The Uncomfortable Truth: Engineering Is Becoming AI Orchestration

Jemin's decade-long journey from Amazon intern to senior engineer offers a front-row seat to the most significant transformation in software engineering history. His insights reveal three uncomfortable truths:

1. The Junior Developer Path Is Broken Traditional career progression assumes you start by writing simple code and gradually handle more complexity. But AI now handles that simple code better than humans. New graduates must start at a level that previously required experience.

2. Experience Matters More Than Ever Companies want developers who can provide context, break down complex problems, and orchestrate AI tools effectively. This creates a "experience gap" - junior roles disappear while senior roles become more demanding.

3. The Future Is Agent Orchestration The most successful engineers won't be those who write the best code, but those who can design, deploy, and manage systems of AI agents. It's less about programming and more about AI architecture.

Three Strategic Responses:

For Current Students: Don't just learn to code - learn to architect AI systems. Focus on prompt engineering, model fine-tuning, and agent coordination. The interview question won't be "can you implement a binary tree?" but "can you design a system where AI agents solve this business problem?"

For Junior Engineers: Race to gain domain expertise and AI orchestration skills. The window for traditional junior roles is closing, but opportunities for AI-savvy problem solvers are expanding.

For Senior Engineers: Your experience becomes your moat, but only if you combine it with AI mastery. The engineers who thrive will be those who use AI to amplify their experience, not replace it.

Jemin's excitement about the future isn't naive optimism - it's strategic positioning. He understands that in a world where anyone can generate code, the value lies in knowing what to build, why to build it, and how to orchestrate the AI systems that will build it.

The transformation is accelerating. The question isn't whether AI will change engineering jobs - it's whether you'll be directing the change or displaced by it.

Connect with Jemin: You can find him on LinkedIn for insights into navigating the AI transformation in engineering careers.

The next few years will determine whether you're part of the solution or part of the displacement. The engineers who thrive will be those who learn to conduct the AI orchestra, not compete with the individual instruments.