🤖 AI, Developers and Engineering

Welcome to 2026, the wild world of AI. For the last month, I have been exploring modern AI models like ChatGPT 5.3 and Opus 4.6. It has filled me with excitement, fear, bitter-sweetness and hope. In this blog, I will be sharing experience and views of my own, my colleages and friends in the software industry.

Disclaimer: Career of no software engineer was harmed in the making of this blog

These according to me are the 5 stages of AI adoption in the software industry. Resemlence of these stages to anything else is purely coincidental.

Chapter I: Denial

Up until early 2025, the AI models were close to a new Software Developer. You give it a problem, it will try its best to solve it and bring the house down in the process. It needed strict monitoring and scope check. The results were good for small problems but giving it complex and architectural tasks was out of question. Software Engineers were relieved and just started using it as mostly autocomplete and documentation tool. Sometimes, if the AI behaved, it got to write test scripts. Actual productivity gains was different depending on who you ask. It just looked like a good replacement of Stack Overflow.

Chapter II: Anger

By mid 2025, almost every new developer interning was on the vibe-train. Tasks were getting completed faster but not like the way you would expect. Just like asking your wish from a tricky genie. Code reviewers were frustrated with the volume of work and sub-par code quality. Work which could be done in 5 lines of code was being done by creation of a new module, new helper functions and perfect documentation. AI "slop" in code was getting out of hand and memes about AI replacing software engineers were in full swing.

Chapter III: Bargaining

Even though AI companies are not making any money, there is huge investment on the potential. There are RAM shortages, GPU shortages, talk about putting data centers into orbit and until then, occupying Greenland to cool the ones on Earth. The AI models' performance on general benchmarks had plateaued. But, it was realised that key detail to extract better performance was context. If you give your entire codebase as context along with using MCP servers for your database, version control, release management platform, group chats and bank account then AI would definitely deliver you the next-gen flappy bird clone.

Chapter IV: Depression

With the AI potential getting realized, phased layoffs are increasing. Entire QA teams and departments are getting replaced by agentic workflows. People who are retained are suppposed to train the AI which would eventually boot them. Bigger question for a layman like me is if capitalism was trickle down then with AI, wouldn't it concentrate wealth further in the hands of few? Feels like these AI companies have an island where every industry wants to go to but can't say it out loud.

Chapter V: Acceptance

Recently, I was working on making a task serverless. I asked a popular coding agent to do it for me. The coding agent did a wonderful job of data validation but passed bulk rows of tables from one worker to other. This is a big no-no from design perspective. So, I completed one workflow myself and asked it to do the remaining. The agent was able to adapt to my coding style and deliver a quality result. This sent me into introspection on what I could have done better to get the best result first. I started asking my collegues and friends on how they use AI and I categorized their responses as follows:

The Delegator: Uses Claude Agent Teams and has build a mini engineering team with 2 pair programming developers, a reviewer, a tester and a Jira ticket updater (because why not?) . Each have their tasks defined in markdown files and also communicate with one another through a shared file. Everything is decoupled and specialized to emulate real world team workflows. The initial setup costed him some time but now he completes a week of workload in a day meanwhile also improving the internal processes of his mini engineering team when they get stuck. He had no comments when I asked what he did with the remaining days of the week. He asserted that this works for him because most of the tech stack used at his organization is open source or has an MCP server available.
The Skeptic: She has never bought in to the hype created by AI tech firms. Be it promises of achieving super intelligence by 2026 or replacing all Software Engineers. She has had enough of AI "slops" and only trusts it with general maintenance tasks. She has no preference or loyalty towards a model because - "all do the same shit".
The Reluctant Old-Schooler: He doesn't like AI. He believes it is just making us dumber. The process of making mistakes, learning and inventing something new crumbles if you pick and apply ready-made stuff. He doesn't want to use AI but other team members using AI are quickly finishing up thier work and so as to not become the blocker has to reluctantly give in.
The Vibe-Coder: He is all-in on the AI vibe-train. He has multiple subscriptions of AI models and dumps out code like a truck. Doesn't want to type code and spends all his time fidgeting with cursor rules. He points out that the end user doesn't care about the code or tech stack. Coming in first is what matters. "How worse can the amount of bugs be? And is that even an issue if I have an agent fixing things over iterations? Isn't that software engineering?"

What's Next?

I also had the privilege of having some "What do you think of the future?" conversations and would like to share some interesting points that came up:

End of new JavaScript Frameworks? I am kind of on the fence for this one. AI could very well just use a unified framework. This could also apply to other languages and frameworks.Yay but nay, sounds boring.
Era of traditional coding interviews seems to be over. However, the skill of problem pattern recognition still remains valuable i.e you need to know which algorithms to apply in the given situation but how to do it with C within 30 min getting all 50 test cases correct with no SegmentationFault is no longer a metric of competence. Langauage agnostic design patterns and larger system architecture would likely dominate the conversation. In general, a polyglot and a generalist would be more preferable to a language speacialist by the companies.
Freelancing and contracts would become more prevalant. Engineers would develop systems, install agents and fly away only to be called if there is some problem. Note to self: Learn to estimate charges from a mechanic.
Opportunities in other disciplines like finance, consultancy, supply chain etc would increase. Demand to create agentic workflow for their mundane administrative tasks would surge. DevOps would silently become MLOps/AIOps which seems to have the potential of becoming one of the highest paid jobs.
Alternatives to existing sloppy apps with snappier features or UI could be much rewarding. Also, a good way of giving the finger back to the big tech laying you off (joking... don't sue me), fixing mistakes they couldn't. Speaking of mistakes, JavaScript could finally be re-written in Ruby. Hey, if you don't have to read the code or fix it, then why not?
Becoming managers is a very real possibility where developers just overlook code getting written by AI agents while you direct them with instructions. How do you feel about becoming the very thing you hated?

Conclusion

The wild world of AI is here to stay. We could be the first generation who sees the most progress in science, medicine, technology, literature (questionable), communication and various fields than any other human in the history of this planet.

But, I do have AI fatigue sometimes. Like, when I was writing this blog. I used multiple LLM models and prompts. Every result I got was fancier, more factual and better. Still, it didn't feel my own. I wonder if in future something imperfect but authentic becomes more valuable than something perfect and generated. And for that reason, this is my first and the last draft. If you find any punctuation or grammatical errors, don't let me know because I'm not gonna fix it :)

🤖 AI, Developers and Engineering

Disclaimer: Career of no software engineer was harmed in the making of this blog

These according to me are the 5 stages of AI adoption in the software industry. Resemlence of these stages to anything else is purely coincidental.

Chapter I: Denial

Chapter II: Anger

Chapter III: Bargaining

Chapter IV: Depression

Chapter V: Acceptance

The Delegator: Uses Claude Agent Teams and has build a mini engineering team with 2 pair programming developers, a reviewer, a tester and a Jira ticket updater (because why not?) . Each have their tasks defined in markdown files and also communicate with one another through a shared file. Everything is decoupled and specialized to emulate real world team workflows. The initial setup costed him some time but now he completes a week of workload in a day meanwhile also improving the internal processes of his mini engineering team when they get stuck. He had no comments when I asked what he did with the remaining days of the week. He asserted that this works for him because most of the tech stack used at his organization is open source or has an MCP server available.
The Skeptic: She has never bought in to the hype created by AI tech firms. Be it promises of achieving super intelligence by 2026 or replacing all Software Engineers. She has had enough of AI "slops" and only trusts it with general maintenance tasks. She has no preference or loyalty towards a model because - "all do the same shit".
The Reluctant Old-Schooler: He doesn't like AI. He believes it is just making us dumber. The process of making mistakes, learning and inventing something new crumbles if you pick and apply ready-made stuff. He doesn't want to use AI but other team members using AI are quickly finishing up thier work and so as to not become the blocker has to reluctantly give in.
The Vibe-Coder: He is all-in on the AI vibe-train. He has multiple subscriptions of AI models and dumps out code like a truck. Doesn't want to type code and spends all his time fidgeting with cursor rules. He points out that the end user doesn't care about the code or tech stack. Coming in first is what matters. "How worse can the amount of bugs be? And is that even an issue if I have an agent fixing things over iterations? Isn't that software engineering?"

What's Next?

I also had the privilege of having some "What do you think of the future?" conversations and would like to share some interesting points that came up:

End of new JavaScript Frameworks? I am kind of on the fence for this one. AI could very well just use a unified framework. This could also apply to other languages and frameworks.Yay but nay, sounds boring.
Era of traditional coding interviews seems to be over. However, the skill of problem pattern recognition still remains valuable i.e you need to know which algorithms to apply in the given situation but how to do it with C within 30 min getting all 50 test cases correct with no SegmentationFault is no longer a metric of competence. Langauage agnostic design patterns and larger system architecture would likely dominate the conversation. In general, a polyglot and a generalist would be more preferable to a language speacialist by the companies.
Freelancing and contracts would become more prevalant. Engineers would develop systems, install agents and fly away only to be called if there is some problem. Note to self: Learn to estimate charges from a mechanic.
Opportunities in other disciplines like finance, consultancy, supply chain etc would increase. Demand to create agentic workflow for their mundane administrative tasks would surge. DevOps would silently become MLOps/AIOps which seems to have the potential of becoming one of the highest paid jobs.
Alternatives to existing sloppy apps with snappier features or UI could be much rewarding. Also, a good way of giving the finger back to the big tech laying you off (joking... don't sue me), fixing mistakes they couldn't. Speaking of mistakes, JavaScript could finally be re-written in Ruby. Hey, if you don't have to read the code or fix it, then why not?
Becoming managers is a very real possibility where developers just overlook code getting written by AI agents while you direct them with instructions. How do you feel about becoming the very thing you hated?