Pinned Tweet
Hello everyone, I'm Harikesh Singh 👋
Working as a Lead DevOps Engineer with 8+ years of experience in DevOps, Cloud & Infrastructure Engineering.
Starting a new adventure:
The Builder Journey 🚀
Through this space, I'll share insights on DevOps, Cloud, AI/ML, GenAI, MCP & Software Engineering.
Always learning, building, and exploring what's next.
Let's connect and grow together!
RAG has become a popular approach for building AI applications.
But is it always necessary?
1-Can a sufficiently large context window solve the problem?
2-Would fine-tuning be more appropriate?
3-Does the application actually need external knowledge?
4-How much retrieval complexity is justified?
5-Are we choosing RAG because it fits or because it's trending?
Are we solving the problem or following the architecture trend?
1
3
134
AI can generate code, explain concepts, and solve common programming problems.
But how should junior engineers develop real skills?
1-Should they focus more on debugging?
2-Should they build complete applications?
3-Should they learn system design earlier?
4-Should they understand infrastructure?
5-How will they gain experience if AI handles basic tasks?
Are we preparing the next generation of engineers for the right challenges?
1
1
71
Your pipeline is green. The deployment completed successfully.
But is that enough?
1-Has application health been verified?
2-Are error rates within acceptable limits?
3-Has database compatibility been checked?
4-Can you roll back without data loss?
5-Has the release been validated from the user's perspective?
Is a successful pipeline the same as a successful deployment?
5
5
115
AI can suggest architectures, generate Terraform, and write application code.
But let's ask:
1-Can it understand every business constraint?
2-Can it predict long-term maintenance costs?
3-Can it identify hidden architectural dependencies?
4-Can it make the right trade-offs under uncertainty?
5-Would you trust it with a critical production decision?
Is technical judgment the last major human advantage?
A technical decision in one service can affect five other teams.
A deployment change can impact production reliability.
A database decision can affect infrastructure costs.
Great engineers understand that their responsibility doesn't end at their own codebase.
3
4
184
Watching tutorials, collecting certifications, and experimenting with new tools can feel productive.
But real engineering growth happens when you build something, break it, troubleshoot it, and understand why it failed.
Knowledge becomes valuable when you can apply it.
2
2
58
An incident isn't just an operational failure.
It reveals assumptions that weren't tested, dependencies that weren't understood, and failure scenarios that weren't considered.
Fixing the immediate issue restores service.
Understanding why it happened makes the system better.
3
5
137
Code reviews were primarily about checking what developers wrote.
Now AI can generate entire pull requests.
The focus is shifting towards validating architecture, business logic, security, performance, and maintainability.
Reviewing code may become less about syntax and more about engineering decisions.
3
5
147
Knowing Kubernetes is valuable.
But deploying everything on Kubernetes doesn't make someone a better engineer.
Understanding containers, networking, resource management, orchestration, and operational trade-offs matters much more than memorizing kubectl commands.
Tools are not engineering judgment.
4
7
274
Automating deployments is great.
But what happens when an automated deployment fails halfway through?
Without execution logs, clear status reporting, failure notifications, and rollback mechanisms, automation can make troubleshooting even harder.
Automation should make operations easier, not invisible.
3
5
85
An application can scale with more compute resources.
Engineering teams need something different.
Clear ownership, reusable components, documentation, architecture standards, and effective communication.
Scaling technology is difficult. Scaling engineering collaboration is another challenge entirely.
2
4
91
Every architecture looks impressive when all services are healthy.
The real test begins when databases slow down, networks fail, queues fill up, and dependencies become unavailable.
Reliable engineering starts by assuming things will eventually break.
8
10
214
A piece of code can pass every test and still introduce architectural problems.
AI can generate solutions quickly, but understanding their long-term impact requires context, experience, and judgment.
The ability to question generated code is becoming a critical engineering skill.
6
3
147
More commits don't mean better engineering.
More pull requests don't mean better software.
More lines of code don't mean more value.
The real question is simple: Are we solving meaningful problems or just producing more activity?
2
84
Unpopular opinion: Being technically brilliant is not enough to become a great leader.
You also need communication, empathy, decision-making, and the ability to make other people successful.
If your team cannot function when you're on vacation, that's not job security.
That's a knowledge-sharing failure disguised as importance.
8
9
320
A team where everyone agrees with the manager isn't necessarily a healthy team.
Sometimes disagreement is the strongest evidence that people care enough to challenge decisions.
3
4
206
The fastest way to lose a team's trust is to blame people for problems created by poor processes.
Fix the process, improve the system, and take responsibility for the outcome.
8
10
324
Unpopular opinion: If your code works but creates a security vulnerability you knowingly ignored, you're not a great engineer.
Technical success without responsibility is incomplete engineering.

