You can't secure what you don't understand.

Security is rarely a default state. In most cases, it's an additional layer. But to enforce it, you need to understand:

1. Common Misconfigurations & Mistakes

2. Common Attacks & Their Defenses

3. Regression Bugs: Every update might bring some new bugs.

But one thing that you fundamentally need to secure a perimeter is the understanding of the perimeter. The most important realization a cybersecurity professional can have is that every perimeter is inherently insecure. And you can't secure it completely but you can make it less vulnerable. Understanding how that perimeter is vulnerable & quantifying it is the first task. Then, figuring out what sort of security can be enforced in order to secure it is the second one.

You can't secure the level that you can't understand.

There's layers to software or hardware. Every individual layer has some function and each layer has an added abstraction. Understanding of all the levels is pretty hard to gain but you can only secure the level that you understand. You can't secure the level that you don't. You have to leave it to god with the prayer that whomsoever has designed the lower levels has taken care of the security aspect.

These are usually your vendors that you licence your goods from. But the truth is security is always an afterthought.

Security isn't the essential thing while making software.

Security isn't considered as the essential thing while making software. But loss of security is an essential event. Because to run the software, we need a Minimum Viable Product (MVP) to show the stakeholders. And stakeholders don't really care about security (cuz most of them aren't very technical). The software should function as intended.

Startups need to move & ship fast so security testing can be removed from your standard Software Development LifeCycle to reduce steps.

This mindset results in the hiring of Cybersecurity folks as a separate entity / team when the product gains momentum. So in some sense, this lack of security is very important for me in getting hired. But the truth is no matter how hard & secure you code, there's always vulnerabilities there!

Personal Segue

• In cybersecurity, simplification is a crime. You must do the deep dive into the details. There’s only 1% possibility that you require that level of depth at a particular point in time. But when you do, you can’t be saying, “I don’t know what’s happening”. You’re the guy everyone reaches to when hell breaks loose. Why? Cuz they think you must know what’s going on. You’re the last hope. And the last hope can’t afford to be dumbfounded because they chose to simplify.

The launch of Grokipedia is a direct challenge to the established order, reigniting the age-old debate that "Whoever controls information, controls the narrative." For years, the community-driven Wikipedia has served as the world's most accessible archive, the place we return to to make sense of the past. Now, an AI-first alternative enters the field, promising to be a more "truth-seeking" source.

The AI-Powered Alternative: A Question of Sustainability

Grokipedia is fundamentally different from its rival. While Wikipedia relies on a global community of human volunteers for writing, editing, and vetting articles, Grokipedia's content is entirely generated and maintained by xAI's conversational model, Grok.

Musk announced the rollout on X (formerly Twitter), claiming, "Version 1.0 will be 10X better, but even at 0.1 it's better than Wikipedia imo." The current version went live with nearly 900,000 articles, a fraction of Wikipedia's millions and features a minimalist homepage with a search bar.

Key distinctions include:

• Content Creation: AI-generated and 'fact-checked' by the Grok model, not human editors.

• Editing: While users cannot directly edit pages in v0.1, Musk stated that in future versions they will be able to. Currently, users are able to "ask Grok to add/modify/delete articles," which the AI will then process or reason why it’s not possible. (I wonder if this can be jailbroken)

• Transparency: Unlike Wikipedia’s open edit history, Grokipedia currently lacks full transparency on content changes.

The AI Paradox: Bias, Hallucination, and Control

The decision to rely solely on AI moderation presents a new set of critical problems that many experts believe makes the current model unsustainable. Grokipedia replaces the human bias that Musk decries with the inherent risks of algorithmic bias and hallucination.

LLMs like Grok are trained on vast datasets from the internet including Wikipedia itself meaning they risk reproducing the biases already embedded in that data. Even if Grok is designed to "purge out the propaganda," the AI itself has demonstrated instances of generating inaccuracies and controversial responses in the past. When an encyclopedia's core editor is an AI that can hallucinate confidently presenting false or misleading information as fact, the pursuit of a robust, reliable knowledge source becomes a significant challenge.

The model is thus a high-stakes experiment: an attempt to recreate a robust, crowd-sourced encyclopedia like Wikipedia using a centralized, AI-governed approach. This raises the critical concern that while Musk hopes to "purge out the propaganda," his own well-documented preferences and biases which he openly supports and which directly influence his platforms may translate into a new form of one-sided information control. Whenever a businessman supports "free speech," it often remains sketchy, with the true support extending only until their interests or profits are hampered.

The Case for Coexistence: Different Lenses on Reality

Instead of viewing Grokipedia purely as a replacement, an emerging perspective suggests the real value may lie in its coexistence with Wikipedia. If Wikipedia is seen as the product of consensus-driven human deliberation and secondary sourcing, the historical default, Grokipedia can be seen as the product of an automated, proprietary algorithm trained to present a different "truth."

For the public, having two vastly different knowledge bases, one human-curated and non-profit, the other AI-curated and for-profit offers a unique opportunity to look at reality from different lenses. Users can cross-reference information, compare editorial framing, and identify competing narratives on sensitive topics. This forces a higher degree of critical thinking from the reader, who is now tasked with navigating a pluralistic digital archive rather than simply relying on a single dominant source.

Built on the Rival's Foundation

In a striking irony, early Grokipedia articles were found to be heavily derived from, and in some cases copied nearly verbatim from, Wikipedia pages, complete with a disclaimer stating the content is "adapted from Wikipedia, licensed under Creative Commons Attribution-ShareAlike 4.0 License." This dependency on its rival highlights the practical difficulties of building a comprehensive knowledge base from scratch and underscores the value of the unpaid labor of the volunteer editors Musk seeks to replace.

The Wikimedia Foundation, which hosts Wikipedia, issued a firm response, stating, “Wikipedia’s knowledge is and always will be human... This human-created knowledge is what AI companies rely on to generate content; even Grokipedia needs Wikipedia to exist.”

What Lies Ahead

The rollout of Grokipedia v0.1 marks the start of a fascinating, if precarious, experiment on the future of knowledge. Whether Grokipedia can overcome its initial challenges of algorithmic bias and limited scope remains to be seen. Its ultimate value might not be in its ability to replace Wikipedia, but to force a new era of critical engagement with online information. Archives are what we return to, and only time will tell if Grokipedia evolves into a trusted archive or simply a high-tech echo chamber with an algorithmic slant, yet one that provides a necessary counter-perspective in the ever-shifting control of the global narrative.

Disclaimer: The article is written for Laymen understanding so there’s no going into the nitty-gritty details on the nature of AI Agents. AI Agents here refer to LLM based Agents and not SLM ones. The core argument is: LLM API based Agentic Systems have failed in Enterprise as stated in the MIT report.

The AI Bubble is Deflating

The notion that AI agents will seamlessly automate the workplace is collapsing. A recent MIT report confirms this skepticism, revealing that 95% of generative AI pilots in enterprises are failing to yield measurable returns. This "implementation gap" proves the current model is unstable.

While India is a global leader in rapidly scaling its Digital Public Infrastructure (DPIs) (like UPI and Aadhaar), it generally remains slow in adopting and realizing the potential of other truly novel tech. However, this sluggish pace, which allows India to test the stability first, is strategically sound when faced with an unstable technology bubble.

Misapplication Instead of Broken Technology

The MIT Report’s claim that 95% of enterprise AI agent pilots fail does not mean that AI agents lack utility; rather, it highlights widespread issues with how enterprises apply and integrate these technologies. The overwhelming evidence from the report indicates that failures are most often due to misaligned strategies, inadequate integration into business workflows, and misplaced investment, rather than problems with the core AI technology itself.

• Why Most AI Agent Pilots Fail: Enterprises often deploy AI agents in areas where they are ill-suited, focusing too heavily on sales and marketing use-cases with low ROI, while overlooking high-value automation opportunities in back-office and operational domains.

• The Build Trap: Many companies attempt to build their own AI tools in isolation, resulting in stalled projects; in contrast, partnering with specialized vendors and integrating solutions into specific workflows achieves a much higher success rate, nearly 67% compared to about 33% for in-house builds. Although isolation results in increased security, it doesn’t guarantee accuracy.

• The 5% Success Stories: The small proportion of enterprises that do succeed with AI agents see meaningful ROI, typically by targeting cost elimination (e.g., business process outsourcing reduction), deploying deeply-integrated solutions, and continuously measuring and improving outcomes. These cases prove significant business value is attainable, but achieving it requires strategic focus and organizational readiness, not generic AI hype.

The Economics: Compute Debt vs. Giants

The high-cost nature of AI development and operation is unsustainable for many startups:

• Compute Debt: Pure-play AI companies are deeply in debt for their compute bills and are unlikely to reach breakeven.

• Incumbent Advantage: The only companies likely to survive this bubble are those like Meta and Google, which treat AI as a side gig subsidized by existing, massive revenue streams.

The Transformation of Work

AI's impact on employment is a massive force of change, particularly for the Indian workforce:

• Displacement at the Base: AI is primarily automating low-knowledge, highly repetitive jobs in knowledge work. While these roles are often not large in volume elsewhere, these low-knowledge jobs are massive in numbers in India (particularly in ITES/BPO), making the overall employment effect substantial.

• Mid-Level Augmentation: Mid-level jobs are not disappearing. Instead, the subprocesses involved in them are getting automated quickly. AI acts as an extended brain to gather perspectives, but replacement is risky.

• Creative Fields: In creative areas like Music, Graphics, and UI/UX, AI is fundamentally taking away very low level creative work, challenging entry-level professionals.

• The Leadership Gap: There is immense value and a critical gap for professionals who combine management skills, business acumen, and a deep, technical understanding of AI. These leaders are crucial for bridging the divide between technology and genuine business value.

The Non-Negotiable Human Role

AI cannot touch roles where accountability, reliability, and sustained focus are critical:

• High Stakes and Compliance: AI cannot take away anything that requires Security, Regulations, and Compliance or High Stakes Stuff, which are jobs where accuracy is paramount and a single mistake can cost a severe damage.

• The Context Flaw: While AI can initially demonstrate attention to detail in the first few prompts, it disappears as soon as the context window overflows. This fundamental technical limitation means that a human must always oversee high-accuracy tasks.

• Unexplored Territory: The entire ecosystem is further complicated by the fact that AI's legal boundaries and data laws regarding it are an unexplored territory. The final shape of the AI economy hinges on future legal decisions.

Conclusion

The current AI landscape is defined by a critical paradox: immense technological capability meets profound implementation and economic instability. For India, the failure of enterprise AI pilots is not an obstacle, but an opportunity to avoid the financial sinkhole of the bubble. The biggest challenge remains protecting the massive volume of low-knowledge jobs that fuel the country's middle class.

Comedy provides a clear and relatable example of defamiliarity in action. A skilled comedian doesn't invent new situations; rather, they take a commonplace subject, such as a frustrating social interaction or a mundane daily chore, and expose its inherent absurdity. By presenting these moments from a fresh, often exaggerated, angle, the comedian forces the audience to step back and rediscover their own experiences. This act of re-viewing creates the humor, turning a routine event into a source of unexpected insight and shared recognition.

This principle extends far beyond comedy to all other art forms. A painter, for instance, does not simply replicate a scene, but uses the very medium of paint and canvas as a source of a new perspective. They rebuild the scene with color, light, and form, and in doing so, we are forced to perceive the world not objectively, but through the medium itself. Similarly, a poet uses novel word choices and structures to give everyday language a new, powerful resonance. In each case, it is the artist's perspective that matters most. They are not merely observers but active participants in the process of reconstruction, taking the raw materials of reality and reassembling them from a new angle. This reconstruction is what defines art, offering us a way to discover the world from a viewpoint that is both new and personal.

Ultimately, art does not change the physical reality of the world, but it fundamentally alters our perception. It is a profound process of taking a step back from the familiar to view the world from a new angle, your angle. Because our perception is inherently subjective, defamiliarity challenges our pre-existing notions, evoking different and deeply personal emotions in each individual. In this act of discovery, you are empowered to reconstruct reality for yourself, uncovering new facets of the world that were previously hidden, and sometimes uncomfortable. This personal journey of re-viewing is what makes art a vital and transformative force.

• Have a checklist of tasks.

• Set a starting time & mark it with an alarm.

• Set an ending time. (A focus intensive sitting must not be of >90 minutes)

• Clear all distractions.

• Zone Out & Stay Attentive.

Why This Focus Method? (Motivation)

In an age of constant pings, endless notifications, and a never-ending to-do list, finding true, deep focus can feel like an impossible task. We often sit down with the best intentions, only to find our minds wandering, our phones beckoning, and our productivity plummeting. But what if there was a simple, repeatable method to cut through the noise and harness your attention?

Inspired by proven time-management techniques, this 5-step "Focus Method" is designed to help you zero in on your tasks, achieve your goals, and reclaim your precious mental energy. Let's dive in.

Step 1: Choose Your Mission (Goal & Checklist)

Before you can focus, you need to know what you're focusing on. Vague intentions lead to vague results. This step is about making your objective crystal clear, specific, and achievable within your dedicated focus block.

Your Goal: Make it precise, aligned with a larger objective, and trackable for one sitting.

• Example: Instead of "Work on my report," try "Write 2 pages of the introduction for the quarterly report."

• Example: Instead of "Study for the exam," try "Complete chapters 3 & 4 practice problems in the calculus textbook."

Is your goal multi-staged or complex? Create a mini-checklist for this session! Have not more than 5 tasks for a single sitting.

If your goal involves several smaller actions to reach completion within your focus block (e.g., "Research and outline blog post"), break it down further.

Step 2: Block Out Your Time (Set Your Alarm)

Commitment is key. Simply thinking about doing a task isn't enough; you need to schedule it. This step turns your intention into a concrete appointment with yourself.

• Action: Pick a specific start time and set an alarm for it. This external cue helps you transition into work mode and overcome inertia.

• Pro Tip: Treat this scheduled block like a non-negotiable meeting. You wouldn't skip a meeting with your boss; don't skip this one with your productivity.

Step 3: Define Your Deep Work Sprint (Choose Your Limit)

Our brains aren't built for endless, uninterrupted concentration. Science shows that focused attention works best in defined bursts. This step sets the boundaries for your intensive work period.

• Action: Decide on a fixed time limit for your focus session, typically 30 or 50 minutes. During this period, your commitment is to work constantly without interruption.

• Why it works: Knowing there's an end in sight makes even daunting tasks feel manageable. It prevents burnout and allows you to commit fully for a set duration. Experiment to find what length works best for your personal focus span.

Step 4: Purge the Pings (Clear Distractions)

This is perhaps the most critical, yet often overlooked, step. Distractions are focus killers, and they come in many forms: physical and digital.

• Physical Distractions:

  • Tidy your workspace. A cluttered desk often leads to a cluttered mind.

  • Close your door if possible.

  • Inform housemates/colleagues that you'll be unavailable for your focus block.

• Digital Distractions:

  • Put your phone on silent and place it out of reach (ideally in another room).

  • Close all unnecessary browser tabs and applications.

  • Turn off notifications for email, social media, and messaging apps.

Step 5: Engage Your Mind & Body (Zone Out & Stay Attentive)

With your goal set, time blocked, and distractions cleared, it's time to mentally and physically prepare for deep work.

• Mentally Prepare: Take a few deep breaths before you start. Remind yourself of your specific goal for the session. Visualize yourself successfully completing the task. This primes your brain for focus.

• Stay Attentive: As you work, consciously bring your mind back whenever it wanders. Don't judge yourself; simply acknowledge the distraction and gently redirect your attention to your task. Your physical posture can also influence your alertness—sit upright, maintain good posture.

Done! You’re set to focus.

Vibe coding is when teams rely on AI-generated code that feels right, looks neat, and even runs fine, but lacks the depth of security, compliance, and architectural alignment that enterprises depend on.

It’s fast, it’s seductive, and if not handled carefully, it’s a recipe for long-term pain.

Developers Love Writing, Not Reading

If you’ve worked with developers, you know this to be true. Most of us love writing code. We hate reading and deciphering it. We thrive when building new things, not when tracing someone else’s thought process.

AI changes this dynamic in a fundamental way. Instead of writing, developers now spend more of their time:

• Validating machine-generated outputs

• Reading long stretches of code they didn’t personally author

• Reconciling what the AI produced with existing architecture and practices

This is mentally exhausting. Because as humans, we’re naturally better at reviewing what we generate. AI-generated code lacks that internal context. The result? Zoning out, missed issues, and fatigue.

And it’s not just about attention spans. Vibe-coded codebases tend to be massive in volume. No one can realistically read thousands of lines dropped in by an AI in one sitting. Teams often only realize after days of effort that the generated codebase is poorly aligned, inconsistent, or in the worst case, fundamentally unusable.

When “Good Code” Isn’t Secure

The real trap of vibe code is that it can look solid. AI-generated code often looks solid. It compiles, it runs, and it appears to follow best practices. But enterprises need more than code that runs. They need code that is:

• Secure against modern threats

• Compliant with internal and external regulations

• Consistent across systems, so that productivity scales instead of erodes

AI doesn’t guarantee any of that.

Think of enterprise software as layers of trust carefully built over time. Every line of code is written within a framework of compliance, compatibility, and standards. When vibe coding skips these rules, it introduces subtle cracks. One skipped compliance check, one deviation in architecture, and suddenly productivity at scale begins to suffer. (or worse, it might introduce a hole that compromises the entire system)

Enterprises don’t just build functioning code, they build on standards, compliance, and long-term reliability. AI doesn’t guarantee those.

Loss of consistency isn’t a stylistic issue, it’s a business productivity issue.

Don’t Ship the Vibe Code (At Least, Not Often)

Shipping vibe code into production is like running your company on duct tape. It might hold for now, but sooner or later, it breaks.

If teams must ship AI-generated code, then enterprise discipline applies:

• Mandatory code reviews by senior engineers who understand the architectural and compliance context

• Rigorous testing including unit, integration, and security

• Compliance validation to ensure architectural consistency

Otherwise, the new code risks becoming “absolute”, completely disconnected from the standards and compatibility rules of the existing codebase.

For Product Managers: When Vibe Coding Makes Sense

Vibe coding isn’t evil, it’s just misplaced when used for production. The one place it shines? Prototyping.

For product managers, it can be a game-changer:

• Skip spec-writing when testing ideas

• Prototype quickly with AI and vibe code

• Validate whether the feature even deserves serious engineering

Too often, specs are written, reviewed, and polished, only for everyone to realize the feature is a bad fit or worsens the UX. With vibe coding, PMs can show stakeholders how it will look and feel without months of wasted cycles.

And in urgent cases, this allows PMs to show, not tell, demonstrating how a feature might look or behave without bogging down engineering teams.

The Takeaway for Enterprises

• For developers: Don’t fall for the neatness of vibe code. Treat it as suspect until validated.

• For managers: Use it for speed, not structure.

• For enterprises: Protect consistency, compliance, and security at all costs.

Vibe coding is a powerful accelerant. But accelerants can either fuel innovation, or burn everything down.

So don’t ship the vibe code. Use it to explore and prototype, but build production systems the way enterprises always have: carefully, consistently, and with compliance.

Let me clarify, I’m not talking about general-purpose SLMs that are just smaller versions of their larger counterparts. The true power lies in models meticulously trained on highly specific data about a particular niche, all while retaining a surprising degree of generalization. The primary decision-making architecture will work with respect to its specialized training data, while broader, general-purpose reasoning will be handled by a separate, language-based architecture.

The challenges in accomplishing this are significant, but not insurmountable:

• Lack of Quality Data: The biggest question is, where do you find the niche data to train these SLMs? Unlike the vast, general web data that fuels LLMs, this information is often proprietary, expensive, and requires deep domain expertise to curate.

• Training Costs: While SLMs are smaller, the initial training cost to create a specialized, high-performance model can still be high. This often means that only enterprises with substantial resources can afford to build them, though this barrier is slowly lowering.

• Loss of Accuracy: SLMs have a reputation for being notorious for hallucinations and inaccuracies, but I believe this is a misconception. When trained on a constrained, highly-relevant dataset, they can be remarkably accurate within their domain. The key is to recognize their limitations and use them only where they are designed to perform.

So, why do I have this opinion?

• The High Cost of LLMs: LLMs kill a lot of compute per token. When you pass a General Purpose LLM like GPT a query, it treats all the tokens in the same way. It might achieve accuracy, but at the cost of high compute and, consequently, high financial cost. In enterprise, where millions of queries are processed daily, cost is of utmost importance.

• The Inefficiency of Monolithic Architectures: What if there was an architecture that could only trigger certain neural networks in isolation with respect to the query given? This would achieve a similar level of accuracy to a large LLM but at a fraction of the compute cost. This is exactly what a hybrid SLM-based architecture enables. Architectures like Mixture of Experts (MoE) and Mixture of Recursions (MoR) are currently being tested in enterprise, but they have their own challenges. A key hurdle with MoE is that all experts often need to be loaded into VRAM for execution, even if only a few are used for a specific query. This can make implementing these models for multidisciplinary or cross-functional queries, which require loading many different experts, pretty hard to achieve effectively. MoR, on the other hand, is an emerging architecture that focuses on reusing a shared stack of layers, dynamically assigning different recursion depths to individual tokens. While it promises to improve parameter efficiency and adaptive computation, it's still in the early stages of development and has its own engineering complexities.

• The Data Privacy Problem: The other major issue with these general-purpose LLMs is that they are often SaaS applications. When you send a query, you're sending your data to an external provider, and there is no guarantee of data privacy. For many industries, from healthcare to finance, this is a non-starter. A private, in-house SLM solves this problem completely by keeping all proprietary data on-premises.

The future of Agentic AI isn't about one monolithic model to rule them all. It's about a federation of specialized, efficient, and cost-effective SLMs, each an expert in its own domain, orchestrated by a general-purpose reasoning engine. This architecture will deliver the performance and specificity enterprises need, without the prohibitive costs and privacy risks of today’s one-size-fits-all models.